SiMa.ai
可信的半导体原生 Physical AI 平台,战略验证真实;但据报约 USD1.4 billion 的估值没有披露收入锚点,且远高于上市边缘 AI 可比公司。
SiMa.ai 是一家可信、背书充足的 Physical AI 芯片公司;但据报道约 USD1.4 billion 的估值已经计入尚未独立可见的执行成果, 也明显高于上市边缘 AI 可比公司,因此在看到经审计财务数据前,纪律性结论应是继续研究。
封面要素
公司概况
SiMa.ai 是一家无晶圆厂边缘 AI 半导体公司,2018 年成立,总部位于加州 San Jose。公司销售全栈 “Physical AI” 平台,把专用芯片(MLSoC,第二代品牌为 Modalix,采用 TSMC N6 工艺)与 Palette 软件套件(SDK、无代码 Edgematic 工具、端侧 LLM 框架 LLiMa,以及 2026 年推出的智能体式 Palette Neat 环境)打包。Modalix 于 2025 年 8 月量产,可在约 10 瓦以内运行 LLM、Transformer、CNN 和生成式 AI。自 2018 年以来,公司大约经历九到十轮融资,累计融资约 USD355 million,最新一轮是 2025 年 8 月由 Maverick Capital 领投、超额认购的 USD85 million Series C,并在约 2026 年 4 月获得 Micron 战略投资。公司瞄准机器人、汽车、工业自动化、航空航天与国防、智能视觉和医疗,代表性合作包括 STIGA、TRUMPF、LTTS、VVDN 和 Synopsys。
- 官网
- sima.ai
- 成立时间
- 2018-01-01
- 创始人
- Krishna Rangasayee
- 创立地点
- San Jose, California, USA
- 总部
- San Jose, California, USA
- 产品
- SiMa.ai 以芯片、系统级模组(1,000 件量级下,8GB 版本 USD349、32GB 版本 USD599)和开发套件(USD1,499)形式销售 Modalix MLSoC,并把 Palette 软件工具链捆绑其上,而不是作为独立经常性许可出售。
- 客户
- 需要在边缘以高能效完成端侧 AI 推理的机器人、汽车、工业自动化、航空航天与国防、智能视觉和医疗领域 OEM 与系统集成商。
- 商业模式
- 通过直销和分销商(Macnica、Enclustra、ThinkRobotics)销售 Modalix 芯片、系统级模组和开发套件;Palette 软件用于提高设计导入黏性,而不是作为独立经常性收入出售。
- 阶段
- Late-stage private; August 2025 Series C and April 2026 Micron strategic investment
- 融资情况
- 自 2020 年 Dell Technologies Capital 领投 Series A 起,经历 2021 年 Fidelity 领投 Series B、2025 年 8 月 Maverick Capital 领投且 StepStone Group 加入的 Series C,约九到十轮共融资约 USD355 million。约 USD1.4 billion 投后估值来自 The Information 报道,公司未确认。
执行摘要
主要优势
- 有差异化、低功耗的边缘 / Physical AI 芯片(Modalix MLSoC,TSMC N6,低于 ~10W),再配 Palette 软件套件; 它瞄准的市场正在增长,云端 GPU 经济性塞不进嵌入式设备的功耗和散热预算。
- 战略和商业验证强:Maverick Capital 领投的 August 2025 Series C 获超额认购,Micron 在 April 2026 战略投资, 2026 Edge AI and Vision Alliance 年度产品奖,以及重点合作(STIGA、TRUMPF、LTTS、VVDN、Synopsys)。
- 创始人 CEO Krishna Rangasayee 扎根半导体行业,曾任 Xilinx SVP 和 Groq COO;董事会与投资人阵容也强, 包括 Fidelity、Point72、Dell Technologies Capital、StepStone,Intel CEO Lip-Bu Tan 也在董事会。
- 边缘 AI 并购市场活跃,愿意支付战略溢价(NXP/Kinara USD307 million;onsemi/Synaptics ~USD7 billion), 给公司提供了可信的战略退出路径。
主要风险
- 据报道约 USD1.4 billion 的估值缺少公开财务锚:SiMa 不披露收入、毛利率、烧钱速度或现金跑道; 第三方收入估算意味着约 28-39x 远期倍数,相比 Ambarella 约 5-8x 市销率明显偏高。
- NVIDIA 的 Jetson 平台和 CUDA/JetPack 生态已经扎根;若被其挤出,SiMa.ai 可能只能守一个细分位。
- 同行压力显示板块脆弱:Hailo 估值据报道跌破 USD500 million,并转向压力 SPAC 路径; Blaize 经营亏损接近 USD103.8 million,且收到持续经营警示。
- 资本密集型 fabless 模式单一依赖 TSMC N6 代工,加上 BIS 出口管制和尚未验证的 ISO 26262 车规认证, 带来融资、供应和资质风险。
- 围绕约 USD355 million 优先股资本的清算优先权堆叠未披露,下行情景退出时可能吞掉大部分回款,继续稀释也很可能。
未决问题
- 经审计收入、毛利率、经营烧钱速度、现金余额和现金跑道。
- Series C 条款清单(清算优先权、参与权、优先顺位)、完整股权结构表和稀释历史。
- 可靠收入数据,以及独立的 MLPerf 级 Modalix 代际基准测试。
- 按金额加权的 design-win 管线,包括量产时间表和头部客户收入集中度。
- 代工与 OSAT 第二来源认证路线图,以及任何具体多元化时间表。
目录
01公司概览
1.1 身份、总部与商业模式
SiMa.ai 是一家无晶圆厂边缘 AI 半导体公司,总部位于加州 San Jose 的 333 W. San Carlos St., Suite 1100。公司成立于 2018 年,出发点是下一轮 AI 将在边缘而非云端取胜。公司把自身品类称为 “Physical AI”:机器智能嵌入看得见、会移动、能行动的设备,例如机器人、无人机、车辆、摄像头和工厂设备;这些场景无法接受云端延迟。它把自己定位为完整系统供应商,而不是单点芯片卖家。 商业产品是两层全栈平台,公司称为 “SiMa.ai ONE”。硬件层是 MLSoC(Machine Learning System-on-Chip),第二代品牌 Modalix 于 2025 年 8 月量产;该芯片采用基于 Arm 的架构和 TSMC N6 工艺,支持 CNN、Transformer、LLM 和生成式 AI 工作负载,功耗低于 10 瓦。软件层是 Palette 套件,包括 SDK、无代码 / 低代码 Edgematic 可视化工具、端侧 LLM 框架 LLiMa,以及 2026 年推出的智能体式 “Palette Neat” 环境。商业模式是芯片加软件产品销售:SiMa.ai 销售芯片、系统级模组(商业 1K 件定价从 8GB SoM USD349、32GB SoM USD599 起)和开发套件(USD1,499),并配套 Palette 软件栈。这使它成为一家以硬件牵引的公司,靠机器人、汽车、工业自动化、航空航天与国防、智能视觉和医疗等领域的出货与设计导入变现。[CO001, CO002, CO003, CO004, CO005, CO006]
| 指标 | 数值 / 状态 | 日期 / 期间 | 信心 | 缺口或备注 |
|---|---|---|---|---|
| 成立时间 | 2018 | 创立 | 高 | 公司和独立来源中的创立年份一致 |
| 总部 | San Jose, California, USA | 当前 | 高 | 公司还称在八个国家设有办公室 |
| 阶段 | 后期私营(Series C 已完成;Micron 2026 年战略轮) | 2025-08 to 2026-04 | 高 | 截至运行日期未披露 IPO 或收购 |
| 最新定价轮 | Series C,$85M,超额认购 | 2025-08-01 | 高 | The Information 报道该轮洽谈规模为「>$100M」;差异未解 |
| 已披露总融资 | $355M | 截至 2025-08 | 高 | 不包括 2026 年 4 月 Micron 未披露金额的战略投资 |
| 报道估值 | 约 $1.4B(融资后,报道值) | 2025 | 中 | 公司拒绝披露;数字来自第三方报道 |
| 先前估值锚 | 约 $960M(PitchBook) | 2024-08 | 中 | 该数字支撑报道中的约 45% 上调 |
| 领投方(Series C) | Maverick Capital | 2025-08-01 | 高 | 也领投了 2024 年 4 月轮次 |
| 旗舰产品 | MLSoC Modalix(第二代),已量产 | 2025-08 | 高 | 低于 10W;TSMC N6;基于 Arm |
| 软件套件 | Palette 软件套件(SDK、Edgematic、LLiMa、Palette Neat) | 2024 to 2026 | 高 | 无代码 / 低代码与智能体式工具链 |
| 收入 / 年化收入 | 未公开披露 | 截至 2026-07-23 | — | 需要管理层确认 |
| 员工数 | 公开资料无法支撑 | 截至 2026-07-23 | — | 留存来源没有钉住当前数字 |
结合公司披露和独立报道。官方 $85M / 未披露估值的表述与第三方 $100M+ / $1.4B 报道不一致时,表格保留差异,而不是抹平。
[CO001, CO002, CO007, CO018, CO019, CO023]SiMa.ai 的身份、产品栈、客户、资本和依赖如何连接。
1.2 领导层、治理与董事会
领导层由创始人兼 CEO Krishna Rangasayee 锚定。他的公开履历对一家硬件初创公司来说很厚:在 Xilinx 工作近 18 年,先后升任整体业务高级副总裁兼总经理、全球销售执行副总裁;随后担任 AI 芯片公司 Groq 的 COO,此前还在 Altera 和 Cypress Semiconductor 任职,并拥有 25 项以上国际专利。这种创始人-市场匹配——来自 FPGA 世界、曾经放大一家数十亿美元半导体业务的销售与运营负责人——是投资论点核心,也把实质性的关键人风险集中到单一高管身上。 董事会和管理梯队进一步强化半导体背景。董事会主席 Moshe Gavrielov 是退休 Xilinx CEO,也在 TSMC、Cadence、NXP 董事会任职;成员包括 Intel CEO、前 Cadence 掌门 Lip-Bu Tan,Scott Darling(Dell Technologies Capital)、Andrew Homan(Maverick Capital)、Jake Flomenberg(Wing Ventures)和 Mike Dauber(Amplify Partners)。运营层包括汽车业务总裁 Harry Kroeger,他来自 Bosch-Daimler 圈层,曾有管理董事会经历,并曾在 Tesla 和 Rivian 董事会任职。一个值得尽调的缺口是联合创始人名单:公司材料称 SiMa.ai 由创始团队启动,本报告任务也列出 Manish Garg(CTO)、Yann Lepenant(首席架构师)和 Manu Prasad(工程执行副总裁)为联合创始人;但独立公开记录始终只把 Rangasayee 列为创始人,因此其他联合创始人仍是公司提供的信息,尚未得到独立佐证。[CO011, CO012, CO013, CO014, CO015, CO016]
| 人员 / 群体 | 当前角色 | 相关过往背景 | 重要性 | 关键人 / 披露备注 |
|---|---|---|---|---|
| Krishna Rangasayee | 创始人兼 CEO | 在 Xilinx 18 年(SVP / GM、全球销售 EVP);Groq COO;Altera、Cypress;25+ 项专利 | 核心产品、资本和商业化人物;半导体创始人与市场匹配度很强 | 对单一高管的关键人依赖非常高 |
| Moshe Gavrielov | 董事会主席 | Xilinx 退休 CEO;TSMC、Cadence、NXP 董事 | 深厚的晶圆厂和 EDA 治理可信度 | 非执行角色;带来行业网络而非日常运营 |
| Lip-Bu Tan | 董事会成员 | Intel CEO;Cadence 退休 CEO 兼董事长 | 释放顶级半导体和 EDA 关系信号 | 也是 SiMa.ai 早期天使投资人 |
| Harry Kroeger | 汽车业务总裁 | Bosch-Daimler 管理委员会;曾任 Tesla 和 Rivian 董事 | 锚定由 Series C 资助的汽车商业化推进 | 运营负责人;汽车是既定路线图重点 |
| Scott Darling | 董事会成员(总裁) | Dell Technologies Capital | 代表最早的机构领投方 | 投资方关联董事 |
| Manish Garg / Yann Lepenant / Manu Prasad 三人 | 按报告要求列名的联合创始人(CTO / 首席架构师 / 工程 EVP) | 留存公开来源未详述 | 若得到佐证,可补齐技术创始团队 | 公司提供;公开记录中未独立确认 |
董事会和管理层姓名已与公司自身领导层 / 治理页面核对;额外联合创始人名单由公司提供,并标记为未验证披露缺口。
[CO011, CO012, CO013, CO014, CO016, CO017]1.3 融资历史、投资者与估值
SiMa.ai 自 2020 年以来完成约九轮披露融资,公司公告与第三方追踪数据拼出的历史基本一致。Dell Technologies Capital 于 2020 年 5 月领投 $30 million Series A;Fidelity Management & Research 于 2021 年 5 月领投 $80 million Series B,并于 2022 年 5 月领投 $30 million 延展轮;MSD Partners 于 2022 年 10 月领投 $37 million Series B1 延展轮;2023 年中又获得 ~$13 million;随后 Maverick Capital 在 2024 年 4 月领投 $70 million 轮,并在 2025 年 8 月 1 日领投超额认购的 $85 million Series C,使累计披露融资达到 $355 million。StepStone Group 作为新投资者加入 Series C,Micron Technology 又在约 2026 年 4 月进行一笔金额未披露的单独战略投资,加深存储与计算伙伴关系。 估值是公开披露刻意留白之处。SiMa.ai 明确拒绝披露 Series C 后估值,Business Standard 也确认这一点。独立报道填补了空白,但并不干净:The Information 报道称 SiMa.ai 正在以约 $1.4 billion 估值融资超过 $100 million——较 PitchBook 所称前一年 8 月约 $960 million 估值溢价超过 45%——这也是独角兽追踪机构和后续报道广泛把 SiMa.ai 称为 ~$1.4 billion 独角兽的原因。由两名知情人士给出的 $100M+ 数字高于公司官方 $85 million 口径,因此确切轮次规模和估值都仍是“报道但未确认”的事实,需要用原始融资文件核验。[CO018, CO019, CO020, CO021, CO022, CO023]
| 利益相关方 / 投资方 | 角色或轮次背景 | 战略意义 | 公开记录显示 | 尽调问题 |
|---|---|---|---|---|
| Maverick Capital | 2024 年 4 月轮次与 2025 年 8 月 Series C 的领投方 | 近期最显眼的支持方;拥有董事会席位 | 公司新闻稿及追踪数据库均列为领投方 | 核实持股比例、董事会权利和按比例跟投条款 |
| StepStone Group | Series C 新进投资方 | 后期成长性背书 | 公开记录列为 Series C 新参与方 | 核实配额及任何信息权 |
| Micron Technology / Micron Ventures | 约 2026 年 4 月战略投资(金额未披露) | 面向边缘 AI 的存储与计算供应链协同 | 追踪数据库和融资报道披露;未出现在 2025 年 8 月新闻稿 | 核实金额、时间,以及是否绑定商业 / 供应协议 |
| Fidelity Management & Research | 领投 Series B(2021 年)及 2022 年延伸轮 | 锚定早期机构融资 | 多轮 Series B 均列为领投方 | 核实持续持股及是否有老股交易 |
| Dell Technologies Capital | 领投 Series A(2020 年);持续投资方 | 最早机构支持方;连接 Dell 生态 | 多个轮次均被列名 | 理清战略投资还是财务投资,以及是否参与董事会 |
| MSD Partners | 领投 Series B1 延伸轮(2022 年 10 月) | Michael Dell 家族资本 | 公开记录列为 $37M 延伸轮领投方 | 核实持股及权利 |
| Lip-Bu Tan / Michael Dell(天使投资人) | 天使 / 个人投资方 | 高信号个人背书 | 融资报道列名;Tan 也是董事 | 核实天使份额与机构份额划分 |
公开信息支持一个战略密度高、半导体属性很重的投资人底盘,但没有披露各投资方对应的控制权、 持股比例或商业义务。
[CO018, CO019, CO020, CO021, CO022, CO025]截至 2026-07-23 报告日的关键成熟度、资本和警示指标。
1.4 规模、布局与牵引力
SiMa.ai 把自己呈现为全球分布式运营。公司称以 San Jose 总部为核心,在八个国家设有办公室;公布的办公室清单包括美国(San Jose)、德国(Stuttgart)、印度(Bengaluru)、以色列(Modiin)、日本(Tokyo)和韩国(Seoul)等。商业进展以垂直行业而非硬收入描述:公司反复点名机器人、汽车、工业自动化、航空航天与国防、智能视觉和医疗作为目标及活跃市场,并列举早期部署,如机器人领域的自主移动机器人和机器人割草机制造商 STIGA,以及与 Synopsys(汽车边缘 AI)和 L&T Technology Services(LTTS)的战略工程与商业化联盟。 公开记录没有给出截至本报告日期仍准确的一组核心运营指标。2026 年收入或收入年化运行率、累计出货量、毛利率和可核验当前员工数都未披露;投资者材料与数据库只提供“客户采用增长”等定性表述。由于员工数、收入和客户数都是波动事实,并会锚定烧钱速度和牵引力分析,本概览把它们列为明确缺口,并给出具体尽调路径,而不是导入未经验证的数字。简言之,SiMa.ai 看起来是一家真实的跨国边缘 AI 供应商,有实际设计导入动能和标杆伙伴,但其规模目前由地域、垂直行业和合作关系支撑,而不是由审计财务或出货指标证明。[CO027, CO028, CO029, CO030, CO031, CO032]
2018 至 2026 年,SiMa.ai 的成立、产品发布、融资节点与合作伙伴关系时间线。
1.5 里程碑与负面信号
公司时间线连贯,且由产品推进。SiMa.ai 2018 年以硬件-软件协同设计假设创立;约 2022 年推出其称为行业首个带完整软件栈的专用 MLSoC;2023 年把第一代 MLSoC 推入生产部署;2024 年发布 Modalix MLSoC 和 Palette Edgematic 无代码工具;2025 年则用 Modalix SoM 与 LLiMa 框架把生成式 AI 带到端侧,并完成 Series C。融资时间线从 2020 年 Series A 延伸到 2025 年 8 月 Series C 和 2026 年 4 月 Micron 战略投资;2026 年公司又推出 Palette Neat 作为智能体式开发环境。 负面框架对尽调同样重要。第一是估值不透明:公司拒绝披露 Series C 估值,广泛引用的 ~$1.4 billion 数字基于匿名信源而非公司声明,同时报道的 “>$100M” 轮次规模与官方 “$85M” 冲突。第二是竞争强度:SiMa.ai 所处市场由 NVIDIA 占据主导份额,即便 Hailo 这类融资充足的边缘 AI 对手,也被描述为难以在 NVIDIA 生态面前把试点转成规模;这对任何挑战者都是结构性逆风。第三是披露缺口:联合创始人记录不完整、当前财务与员工指标缺失、股权结构经济条款未披露,都意味着股权故事依赖尚未进入公开来源的私人数据。[CO034, CO035, CO036, CO037, CO038, CO039]
| 日期 | 事件 | 类型 | 金额 / 估值 / 状态 | 参与方 | 意义 |
|---|---|---|---|---|---|
| 2018 | SiMa.ai 在 San Jose 创立 | 创立 | 公司成立 | Krishna Rangasayee 与创始团队 | 为 Physical AI 论点打下法律和技术底座 |
| 2020-05 | Series A 轮 | 融资 | $30M | Dell Technologies Capital(领投) | 首笔机构资金;验证边缘 AI 芯片押注 |
| 2021-05 | Series B 轮 | 融资 | $80M | Fidelity(领投)、Dell、Amplify | 把芯片研发推向规模化 |
| 2022 | 首个专用 MLSoC 与完整软件栈出货 | 产品 | 第一代 MLSoC | SiMa.ai | 建立软硬件协同设计差异化 |
| 2022-10 | Series B1 延伸轮 | 融资 | $37M | MSD Partners(领投) | 衔接量产爬坡 |
| 2023 | 第一代 MLSoC 进入生产部署 | 规模化 | 生产部署 | 机器人、工业、智能视觉客户 | 在实际场景验证协同设计论点 |
| 2024 | 发布 Modalix MLSoC 和 Palette Edgematic | 产品 | 第二代平台发布 | SiMa.ai、Synopsys、Arm、TSMC 等参与方 | 迈向多模态 / GenAI 与无代码部署 |
| 2024-04 | Maverick 领投轮次 | 融资 | $70M(PitchBook 报道估值约 $960M) | Maverick Capital(领投) | 为 Series C 估值上跳设定上一轮锚点 |
| 2025-08-01 | 完成 Series C | 融资 | $85M;累计融资至 $355M;报道估值约 $1.4B | Maverick(领投)、StepStone | 达到独角兽里程碑;支持全球扩张和汽车业务 |
| 2025-08 | Modalix 量产,SoM 与 LLiMa 发布 | 产品 | 已出货;SoM 起价 $349 | SiMa.ai、Enclustra、Arm、TSMC、Synopsys 等参与方 | 第二代平台进入商业供货 |
| 2025-09 | 与 L&T Technology Services 建立战略联盟 | 合作 | 联合开发联盟 | SiMa.ai、LTTS | 把销售落地拓展到移动出行、工业、医疗健康 |
| 2026-04 | Micron 战略投资 | 融资 | 金额未披露 | Micron Technology | 加深存储 / 计算供应链协同 |
| 2026 | 发布 Palette Neat 智能体环境 | 产品 | 智能体 AI 工具 | SiMa.ai | 平台迈向自主端侧智能体 |
单一事实时间线。若干产品节点来自公司叙事,只能定位到年份或月份;融资日期和金额由新闻稿及独立追踪数据库交叉佐证。 估值为报道口径,非公司确认。
[CO001, CO007, CO018, CO019, CO024, CO034]1.6 展示材料
02市场分析
2.1 市场边界、纳入支出与替代方案
SiMa.ai 卖的不是通用 “AI” 产品;它位于边缘 AI 推理芯片层——单颗 MLSoC(“Modalix”)把 Arm 应用核心、专用 ML 加速器和视觉 / DSP 模块融合在一起,使计算机视觉、Transformer 和生成式 AI 模型能在本地以约 10 瓦以内运行,并配套 Palette 软件工具链。这把纳入支出的边界收得很窄:专用边缘 AI 处理器、系统级模组、开发套件,以及把训练好的模型变成可部署边缘应用的配套软件。它明确排除云端 / 数据中心训练芯片、通用 CPU、独立传感器和终端用户 AI SaaS 席位;这些属于相邻市场。 SiMa.ai 瞄准的买家——机器人、工业自动化、汽车 / ADAS、航空航天与国防、智能视觉和医疗设备制造商——越来越需要实时推理;在这些场景里,云端往返太慢、太耗电,或让敏感数据暴露过多。因此他们拿来比较的现状替代品不只是其他边缘 AI 初创公司(Hailo、Ambarella、Blaize),更重要的是 NVIDIA Jetson 平台;Jetson 主导边缘机器人计算。此外还有 Qualcomm 的通用 SoC、嵌入式 GPU、FPGA,某些情况下也包括继续依赖云端推理。SiMa.ai 的位置由它相对这些选择所声称的更高每瓦性能和单芯片、软件优先体验来定义,因此相关市场最好从边缘 AI 硬件 / 芯片层出发,并叠加 Physical AI、机器人和汽车推理视角。[CM001, CM002, CM003, CM004, CM005, CM006]
| 细分 / 类别 | 纳入口径 | 排除口径 | 买方 / 付款方 | 为什么重要 |
|---|---|---|---|---|
| 边缘 AI 推理芯片(核心 SAM) | 专用 MLSoC、NPU、边缘 AI 加速器、IP | 云端 / 训练 GPU、通用 CPU、传感器 | 机器人 / 工业 / 汽车产品 P&L 负责人 | 这是 SiMa.ai Modalix MLSoC 的直接市场 |
| 边缘 AI 模组和开发工具 | 系统级模组、开发套件、Palette 软件 / SDK | 独立企业 AI SaaS 席位 | 嵌入式工程团队、项目预算 | 覆盖 SiMa.ai 的 SoM 和软件附加收入 |
| Physical AI / 机器人计算 | 机器人和自主机器的端侧计算 | 机械执行器、底盘、非 AI 电子器件 | 机器人 OEM、AMR 与人形机器人厂商 | SiMa.ai 主打的高增长需求框架 |
| 汽车 / ADAS 推理 | 用于感知和驾驶辅助的车载 AI SoC | 动力总成、纯娱乐信息系统、非 AI ECU | Tier-1 供应商与 OEM 项目预算 | 与 Synopsys 合作路径相连的大相邻市场 |
| 航空航天、国防和边缘视觉 | 面向国防 / 工业视觉的加固型边缘推理 | 通用 IT、云分析、后台 AI | 国防主承包商、工业和零售集成商 | SiMa.ai 瞄准的高价值、供应可保障利基市场 |
边界以 SiMa.ai 的 MLSoC / Palette 定位为锚,再扩到买方实际评估的 Physical AI、汽车和国防相邻市场。
[CM001, CM002, CM003, CM004, CM005, CM006]边缘 AI 价值链从 IP 和晶圆代工出发,经过 SiMa.ai 的 MLSoC 与模块,最终到达垂直 OEM 和终端部署。
[CM001, CM002, CM006, CM020, CM026]2.2 通过边缘 AI、Physical AI 和汽车视角测算市场
没有一个公开数字能干净覆盖 SiMa.ai 的机会,因此更稳妥的做法是保留多个视角,而不是硬凑一个 TAM。最窄的硬件口径下,MarketsandMarkets 估算边缘 AI 硬件市场 2025 年为 USD26.14 billion,到 2030 年增至 USD58.90 billion,CAGR 为 17.6%;这一层最直接对应 SiMa.ai 的芯片和模组。更宽的边缘 AI 定义把软件和服务纳入后,规模更大、增速更快:Fortune Business Insights 认为边缘 AI 将从 2025 年 USD35.60 billion 增至 2034 年 USD445.75 billion(32.5% CAGR),Global Market Insights 估算为 2025 年 USD25.2 billion 到 2035 年 USD225.5 billion(24.7% CAGR),Grand View Research(经 Axis Intelligence)估算为 2025 年 USD24.9 billion 到 2033 年 USD118.7 billion(21.7% CAGR)。这些估计在 2025 年约 USD25-36 billion 的基数上相当集中,但终值因口径和时间窗不同而明显分化;应交叉校验,而不是简单平均。 第二个视角是 SiMa.ai 自己推向市场的 Physical AI。MarketsandMarkets 估算狭义 Physical AI 计算 / 软件层到 2032 年达到 USD15.24 billion,并从 2026 年起以 47.2% CAGR 增长;更宽的 Physical AI 生态估计(所有机器人、自主机器与基础设施)在 2026 年约 USD383 billion,2040 年进入数万亿美元;AI 机器人市场则预计从 2025 年 USD6.11 billion 增至 2030 年 USD33.39 billion。第三个汽车视角也直接相关,因为 SiMa.ai 设有汽车业务单元,并与 Synopsys 合作:MarketsandMarkets 估算汽车 AI 市场从 2025 年 USD18.83 billion 增至 2030 年 USD38.45 billion(15.3% CAGR),汽车 AI SoC 细分到 2034 年约 USD13.0 billion(15.6% CAGR),ADAS 市场从 2021 年 USD20.73 billion 增至 2030 年 USD74.57 billion。诚实结论是分层的:直接 SAM 看边缘 AI 硬件,需求方向看 Physical AI 与机器人,汽车 / ADAS 则是体量大、既有玩家重的相邻市场。保留的公开来源没有隔离 SiMa.ai 自身可服务份额,因此这些数字界定机会边界,而不是公司收入。[CM008, CM009, CM010, CM011, CM012, CM013]
| 口径 / 发布方 | 地域 | 规模 | CAGR / 增长 | 方法提示 | 置信度 | 局限 |
|---|---|---|---|---|---|---|
| 边缘 AI 硬件 / MarketsandMarkets | 全球 | USD26.14B(2025)至 USD58.90B(2030) | 17.6% | 专用边缘 AI 硬件 / 芯片 | 高 | 仅硬件;不含软件 / 服务 |
| 边缘 AI / Fortune Business Insights | 全球 | USD35.60B(2025)至 USD445.75B(2034) | 32.5% | 广义边缘 AI,含软件和服务 | 中 | 口径最宽;窗口最长,抬高终点规模 |
| 边缘 AI / Global Market Insights | 全球 | USD25.2B(2025)至 USD225.5B(2035) | 24.7% | 广义边缘 AI 市场 | 中 | 10 年窗口拉大分歧 |
| 边缘 AI / Grand View(经 Axis) | 全球 | USD24.9B(2025)至 USD118.7B(2033) | 21.7% | 广义边缘 AI 市场,硬件占比 51.8% | 中 | 来自聚合方,不是原始页面 |
| Physical AI 计算 / MarketsandMarkets | 全球 | 到 2032 年达 USD15.24B | 47.2% | 较窄的 Physical AI 计算 / 软件层 | 中 | 新品类;定义仍在成形 |
| Physical AI 生态 / TechRT | 全球 | 约 USD383B(2026)至 USD3.26T(2040) | 多倍增长 | 覆盖机器人 / 自主系统 / 基础设施全生态 | 低 | 作为 SiMa.ai 直接 SAM 过宽 |
| AI 机器人 / TechRT | 全球 | USD6.11B(2025)至 USD33.39B(2030) | ~40%+ | AI 机器人系统市场 | 低 | 聚合二手估算 |
| 汽车 AI / MarketsandMarkets | 全球 | USD18.83B(2025)至 USD38.45B(2030) | 15.3% | 汽车 AI,含 ADAS / 自主驾驶 | 中 | 比单纯车载 AI 芯片更宽 |
| 汽车 AI SoC / Mobility Foresights | 全球 | 到 2034 年约 USD13.0B | 15.6% | 车载 AI 系统级芯片收入 | 中 | 时间窗口长;单一发布方 |
| ADAS / NextMSC | 全球 | USD20.73B(2021)至 USD74.57B(2030) | 14.2% | 完整 ADAS 系统,不限芯片 | 中 | 系统层口径;芯片只是子集 |
这些口径要合看,不能相加。边缘 AI 硬件最接近 SAM 代理;Physical AI 和汽车数据圈定需求方向与相邻市场。 没有任何口径单独拆出 SiMa.ai 能拿到的份额。
[CM008, CM009, CM010, CM011, CM012, CM013]从广义物理 AI 生态逐层看到 SiMa.ai 核心边缘 AI 硬件 SAM,以及已服务份额未披露。
各层只是方向性参考,不能相加,因为底层单位、口径和市场定义不同。
[CM008, CM009, CM014, CM016, CM017, CM019]以一致的 USD 十亿口径展示边缘 AI 市场的低 / 基准 / 高来源支持区间:当前基准较一致,近期分歧更大。
单位为 USD 十亿。2025 基年使用 Grand View(低)、MarketsandMarkets 硬件(中)和 Fortune Business Insights(高);2026 近年使用同一批发布方的下一年数据。
[CM009, CM010, CM011, CM012]2.3 买方分层、预算所有者与采用路径
SiMa.ai 的市场横跨多个买方类型,并非单一同质客户;公司围绕垂直行业组织商业化:机器人、工业自动化与机器视觉、汽车 / ADAS、航空航天与国防、智能视觉 / 零售,以及医疗设备。使用者通常是嵌入式或 ML 工程团队,需要把感知模型或生成式 AI 模型塞进受散热和功耗约束的产品;买方是掌握 BOM 和路线图的产品、平台或工程负责人;付款 / 预算所有者通常是硬件产品 P&L 或项目预算,而不是云预算项。采用触发点因细分不同而变化——机器人看电池续航和散热上限,汽车看功能安全和延迟,国防看加固和供应链保障,医疗看隐私 / 监管约束——但共同点都是要把推理搬到端侧。 采用路径通常从开发套件和 Palette 软件评估开始,经过系统级模组设计导入(SiMa.ai 公布 SoM 的 1K 件定价为 8GB USD349、32GB USD599,开发套件为 USD1,499),在某个平台赢得设计导入后进入量产。由于这是硬件设计导入,销售周期长,导入后的转换成本高;参考客户和生态伙伴(Arm、TSMC、Synopsys、L&T Technology Services、Enclustra)能实质性缩短评估期。这种动态两面:SiMa.ai 赢下一个芯片位时会受益,但也偏向在位者(NVIDIA Jetson);大多数工程团队已经熟悉默认选择,感知集成风险最低。[CM020, CM021, CM022, CM023, CM024, CM025]
| 细分 | 买方 | 用户 | 付款方 / 预算负责人 | 工作流 | 采用触发点 | 含义 |
|---|---|---|---|---|---|---|
| 机器人 / Physical AI | 机器人产品负责人 | 嵌入式 / ML 工程师 | 硬件产品 P&L | 机器人 / AMR 中的感知与端侧 GenAI | 电池与散热限制、实时控制 | 每瓦性能是决定性指标 |
| 工业自动化 / 机器视觉 | 自动化平台负责人 | 视觉 / 控制工程师 | 资本开支 / 项目预算 | 缺陷检测、预测性维护 | 延迟与本地数据控制 | 设计导入周期长,但拿下后粘性高 |
| 汽车 / ADAS | Tier-1 / OEM 项目负责人 | ADAS 软件团队 | 车型项目预算 | 传感器融合、感知推理 | 功能安全、延迟、单车成本 | 最大相邻市场;巨头密集、周期慢 |
| 航空航天与国防 | 国防主承包商 / 集成商 | 系统工程师 | 列装项目预算 | 加固型边缘视觉和自主能力 | 供应保障、功耗、加固 | 高价值、低数量、认证严格 |
| 智能视觉 / 零售 / 医疗健康 | 设备或解决方案 OEM | 产品工程师 | 设备 BOM 预算 | 端侧分析和监控 | 隐私、监管、功耗包络 | 隐私要求把推理推向端侧 |
买方地图结合 SiMa.ai 公布的垂直领域,以及边缘 AI 硬件设计导入中常见的预算负责人和采用触发点。
[CM020, CM021, CM022, CM023, CM024, CM025]目标细分市场按预算负责人、采用触发因素、运营优先级和采用路径映射。
[CM020, CM021, CM022, CM023, CM024, CM025]2.4 增长驱动、采用约束与估值相关性
SiMa.ai 所在市场的顺风因素在独立分析机构页面中有充分记录。最主要驱动是延迟:本地推理消除云端往返,对机器人、机器视觉和安全关键控制是决定性因素。数据隐私和主权要求敏感视频、工业和医疗数据留在设备上,以满足监管。能效——SiMa.ai 每瓦性能叙事的核心——让 AI 能在电池供电和受散热约束的产品上运行。IoT 和 5G 连接扩散(全球 5G 连接数 2023 年升至 1.76 billion,预计 2028 年达到 7.9 billion)扩大了可承载分布式智能的设备基数;端侧生成式与智能体 AI 兴起,也提高了每台边缘设备必须处理的计算强度。 约束同样实质,也是可触达市场低于表面 TAM 的原因。安全是双向的:边缘部署扩大攻击面,并增加安全启动和监控成本。异构硬件 / 软件之间的生态碎片化和互操作问题拖慢部署,也抬高切换风险。边缘 AI 集成技能短缺、工业 / 汽车严格监管与功能安全要求、先进边缘芯片高成本和长设计导入周期,都会拉长采用。最上层是竞争结构:NVIDIA 在边缘机器人计算中占主导,多数边缘 AI 芯片挑战者难以放大。因此,市场增长比持久拿份额更容易证明。对估值而言,这意味着市场确实大且在扩张,但 SiMa.ai 据报 ~USD1.4 billion 的估值最终必须由设计导入和收入支撑,而不是只靠 TAM。[CM028, CM029, CM030, CM031, CM032, CM033]
| 驱动 / 约束 | 方向 | 时间 | 证据 | 为什么重要 | 尽调问题 |
|---|---|---|---|---|---|
| 低延迟本地推理 | 驱动 | 当前 | Wevolver 与 GM Insights | 云端往返对机器人 / 控制太慢 | 哪些 SiMa.ai 工作负载最需要端侧低延迟? |
| 数据隐私与主权 | 驱动 | 当前 | Wevolver 与 Polaris | 敏感视频 / 工业 / 健康数据留在本地 | 需求有多少由隐私驱动,多少由成本驱动? |
| 能效 / 每瓦性能 | 驱动 | 当前 | MarketsandMarkets 与 Axis | 让电池 / 散热受限产品跑 AI | SiMa.ai 的每瓦性能领先能否顶住下一代对手? |
| IoT 与 5G 设备扩张 | 驱动 | 当前至中期 | Polaris(5G 连接数从 1.76B 到 7.9B) | 扩大承载边缘智能的装机基础 | 哪些设备品类能转化为设计定点? |
| 端侧生成式与智能体 AI | 驱动 | 当前至中期 | MarketsandMarkets 的 Physical AI 口径 | 提高单设备计算强度 | Modalix 的余量能否匹配 GenAI 模型增长? |
| 安全与攻击面扩大 | 约束 | 当前 | Wevolver | 边缘部署扩大攻击面,也增加安全启动成本 | SiMa.ai 如何满足边缘安全要求? |
| 生态割裂与技能缺口 | 制约 | 当前 | GM Insights + Wevolver | 拖慢部署;抬高切换 / 集成风险 | Palette 能把集成摩擦降到什么程度? |
| NVIDIA 主导地位与创业公司出清 | 制约 | 当前 | CNBC + AIMultiple | 现有巨头仍是默认选择;多数边缘 AI 创业公司做不大 | 哪些证据显示 SiMa.ai 正从 Jetson 手中拿下设计插槽? |
多项因素有两面性;驱动因素放大市场,制约因素则抬高独立边缘芯片挑战者的执行门槛。
[CM028, CM029, CM030, CM031, CM032, CM033]2.5 冲突估计与剩余尽调缺口
严谨的市场章节应保留未知项。第一项缺口是定义分散:公开边缘 AI 市场值在 2025 年约 USD25-36 billion 的基数上大致一致,但终年预测相差数千亿美元,因为各家口径对软件、服务、数据中心相邻支出、地理范围和时间跨度的纳入 / 排除不同。把任一预测当成标准答案都会扭曲机会。第二且更重要的缺口是公司特定数据:保留的公开来源没有披露 SiMa.ai 的收入、出货量、设计导入数量或已服务市场份额,因此这些规模视角只能划定机会边界,无法桥接到 SiMa.ai 的 SAM 或 SOM。 第三项缺口是竞争份额数据:虽然 NVIDIA 被广泛描述为边缘主导者,但保留来源没有给出 SiMa.ai 所瞄准特定 Physical AI 芯片位中 Jetson、Hailo、Ambarella、Qualcomm 和 SiMa.ai 的精确、当前份额拆分。务实结论是,市场证据对需求方向最强,对 SiMa.ai 已捕获份额以及独立边缘芯片供应商在强势在位者面前可拿到的利润池最弱。投资者应把市场视为支撑性背景,把信心留给后文的设计导入、产品和财务证据。[CM011, CM019, CM036, CM038, CM039, CM040]
2.6 展示材料
03竞争对手
3.1 竞争格局与品类地图
SiMa.ai 的竞争集横跨五层。存量巨头层由 NVIDIA 主导,其 Jetson 模组(AGX Orin 最高 275 TOPS,以及更新的 Jetson Thor 一代)是默认边缘机器人计算平台,CUDA-X 和 JetPack 软件栈进一步加固了这一地位,大多数工程团队已熟悉这套工具;Qualcomm 是另一存量巨头,拥有巨大移动 / IoT 规模和 Robotics RB 系列平台。直接同类层是一批与 SiMa.ai 最相似、由风险资本支持的边缘 AI 芯片初创公司:Hailo(以色列,2017 年成立)、Blaize(现已在 Nasdaq 以 BZAI 上市)、EdgeCortix(日本)和 Axelera AI(欧洲)。Ambarella 位置不同,它是上市纯玩家,已把成熟的摄像头 SoC 业务转向边缘 AI,并正在放大真实收入。 除具体厂商外,SiMa.ai 也与替代方案和现状竞争:改造通用 SoC 和应用处理器、嵌入式 GPU、FPGA;对于尚未被迫迁移到边缘的团队,则还有继续依赖云端推理。对大型 OEM 和超大云厂商来说,内部自研是真实替代,因为它们能设计定制芯片(Tesla、Google、Amazon 和 Apple 在相邻领域已经这样做)。未来可能进入者包括超大云厂商的边缘推理芯片,以及 Mobileye 等汽车专用玩家向同一芯片位下探。实际含义是,SiMa.ai 不只是和几家初创公司打;它要在拥挤同类对手争抢相同设计导入的同时,替换一个软件加芯片都深度扎根的在位者。[CP001, CP002, CP003, CP004, CP005, CP006]
| 竞品 | 类别 | 规模 / 融资 | 目标细分市场 | 差异化 | 局限 |
|---|---|---|---|---|---|
| NVIDIA Jetson | 现有巨头 | 母公司万亿美元市值;边缘侧份额领先 | 机器人、自主机器、工业 | CUDA/JetPack 生态;最高 275 TOPS | 功耗高(15-60W);预算紧时成本偏高 |
| Qualcomm | 现有巨头 | 超大市值;移动 / IoT 规模庞大 | 5G 机器人、IoT、边缘设备 | 连接能力 + 规模;SoC 产品线广 | 边缘 AI 能效比落后于专精厂商 |
| Ambarella (AMBA) | 上市垂直厂商 | FY2026 收入 USD390.7M;约 80% 来自边缘 AI | AI 摄像头、汽车、安防、机器人 | 超低功耗视频 AI;出货 42M+ 颗 SoC | 摄像头基因较重;成本 / R&D 承压 |
| Hailo | 直接同业(私营) | 累计融资约 USD340M+;估值跌破 USD500M | 汽车、边缘 GenAI、安防、零售 | TOPS/W 高;边缘 GenAI(Hailo-10H) | SPAC 陷入困境;2026 年裁员约 10% |
| Blaize (BZAI) | 直接同业(上市) | 2025 年收入 USD38.6M;2025 年 1 月 SPAC | 智慧城市、公共安全、主权 AI | 可编程图流式架构 | 收入基数小;需求集中 |
| EdgeCortix | 直接同业(私营) | 累计融资 USD110M+(B 轮 + 补助 / 债务) | 机器人、电信、工业、国防 | SAKURA-II;软件优先基因 + MERA 编译器 | 商业规模仍早;相较现有巨头体量小 |
| SiMa.ai(研究对象) | 直接同业(私营) | 累计融资约 USD355M;据报估值约 USD1.4B | 机器人、工业、汽车、国防 | 单芯片 MLSoC + Palette 软件优先堆栈 | 未披露收入或出货量 |
| 通用 SoC / FPGA / 云 | 替代方案 / 现状 | 供应商基础广 | 还没被迫迁到边缘侧的团队 | 熟悉、灵活,不需要新芯片导入 | 能效比更差;延迟和隐私受限 |
画像基于公司披露、财务业绩和分析师 / 媒体报道;SiMa.ai 未披露收入,相关字段来自公司口径。
[CP001, CP002, CP009, CP010, CP011, CP012]按能效、商业规模与生态成熟度映射边缘 AI 芯片厂商;NVIDIA 占据高规模角落,高效率挑战者则集中在分销能力较弱的区域。
分数是基于公开规格、财务披露和对比资料作出的有证据支撑的序数判断,并非厂商直接披露的指标。
[CP002, CP009, CP014, CP016, CP018, CP020]3.2 竞争对手画像、规模与融资
同行在规模和财务健康上差异很大,这一点重要,因为边缘 AI 芯片资本开支重、设计导入周期长。NVIDIA 几乎不受约束:Jetson 由一个万亿美元业务版图供血,并拥有主导开发者生态。Ambarella 是最健康的独立可比公司,披露 2026 财年收入创纪录达 USD390.7 million,其中约 80% 来自边缘 AI,累计出货超过 42 million 颗边缘 AI SoC,并与 Hanwha 签有多年协议,收入机会为 USD800 million。Blaize 于 2025 年 1 月通过与 BurTech 的 SPAC 合并上市,披露 2025 年收入 USD38.6 million(高于 2024 年的 USD1.6 million),这是其首个完整商业收入年度,收入集中在智慧城市、公共安全和主权 AI 部署。 私人同行中,Hailo 是警示案例:它 2021 年成为独角兽,2024 年 4 月通过 USD120 million Series C 延展轮达到 USD1.2 billion 估值;到 2026 年初,其估值腰斩至低于 USD500 million,裁员约 10%,并寻求困境 SPAC 合并以获得生存资本。EdgeCortix 已融资超过 USD110 million(Series B 加补助和债务),正在放大 SAKURA-II 加速器,并规划下一代 SAKURA-X 小芯片路线图。面对这一阵容,SiMa.ai 累计融资约 USD355 million,据报在 2025 年 8 月 Series C 后估值约 USD1.4 billion;但与 Ambarella 和 Blaize 不同,它不披露收入,因此规模只能由融资额和产品里程碑衡量,而非出货或销售。[CP009, CP010, CP011, CP012, CP013, CP014]
这些规模与财务健康度对比信号,用来判断 SiMa.ai 相对于上市与私营同业的竞争就绪度。
[CP010, CP011, CP013, CP014, CP015, CP035]3.3 能力、定价与商业化对比
原始能力上,NVIDIA Jetson AGX Orin 以最高 275 TOPS 的峰值吞吐领先,但功耗为 15-60 瓦,更适合性能关键型机器人,而不是严格功耗预算。SiMa.ai 的 Modalix 和 Hailo 芯片在相反轴线上竞争:效率。Hailo-10H 以约 2.5 瓦实现约 40 INT4 TOPS(~16 TOPS/W),主打生成式 AI;SiMa.ai 则宣传 50+ TOPS、低于 10 瓦,并强调在单颗芯片上跑完整流程(视觉加生成式模型)。Ambarella 赢在超低功耗 AI 摄像头细分(高分辨率视频下低于 2W),Blaize 以可编程图流架构和易编程性区分自己,EdgeCortix 则靠软件优先的 Dynamic Neural Accelerator 加 MERA 编译器。Qualcomm 的 Robotics RB5 约 15 TOPS、功耗 5-15 瓦,并有强连接集成。 价格上,SiMa.ai 在多数同行不透明的地方给出清晰口径:1K 件量级下系统级模组为 USD349(8GB)和 USD599(32GB),开发套件为 USD1,499。NVIDIA Jetson 开发套件定价相近,但背后有更深、更免费、教学广泛的软件生态,这正是商业化竞争的核心。SiMa.ai 的回应是 Palette——一套软件优先、低代码工具链(包括 Edgematic 和端侧 LLM 框架 LLiMa),旨在降低集成摩擦——以及一个能缩短评估周期的伙伴生态(Arm、TSMC、Synopsys、L&T Technology Services、Enclustra、分销商 Macnica)。比较显示,SiMa.ai 在效率和单芯片集成上有竞争力,但必须顶住在位者的生态重力,也要面对多个同行已经批量出货的事实。[CP018, CP019, CP020, CP021, CP022, CP023]
| 采购标准 | SiMa.ai Modalix 芯片 | NVIDIA Jetson Orin | Hailo-10H | Ambarella | Blaize |
|---|---|---|---|---|---|
| 峰值吞吐(TOPS) | 50+ TOPS | 最高 275 TOPS | 约 40 INT4 TOPS | 按场景而定(视频) | 约 16 TOPS(Pathfinder) |
| 功耗 / 能效 | 10W 以下,TOPS/W 高 | 15-60W,效率较低 | 约 2.5W,约 16 TOPS/W | 8K 视频低于 2W | 约 7W,中等 |
| 软件生态成熟度 | Palette(新兴) | CUDA/JetPack(主导) | Dataflow SDK(成长中) | 成熟摄像头 SDK | 图流式 SDK |
| 端侧 GenAI / LLM 支持 | 是(LLiMa 框架) | 是 | 是(LLM/VLM) | 新兴 | 新兴 |
| 商业规模 / 出货 | 未披露 | 装机基数领先 | 已部署,但现在承压 | 42M+ 颗边缘 AI SoC | 早期,2025 年收入 USD38.6M |
| 目标垂直行业宽度 | 广(6 个垂直行业) | 广 | 汽车、GenAI、安防 | 摄像头、汽车、安防 | 智慧城市、国防 |
单元格反映供应商声称和独立对比来源;TOPS 数字使用供应商最大化条件,不同精度之间不能直接比较。
[CP018, CP019, CP020, CP021, CP022, CP023]| 供应商 | 定价模式 | 公开价格 | 包含能力 | 含义 / 未知 |
|---|---|---|---|---|
| SiMa.ai | 系统模块 + 开发套件 | SoM USD349(8GB)/ USD599(32GB);开发套件 USD1,499 | MLSoC 模块 + Palette 软件工具链 | 透明定价有利评估 |
| NVIDIA Jetson | 模块 + 开发套件 | 开发套件约 USD1,500-2,000 | 模块 + 免费 CUDA/JetPack 生态 | 生态价值超过硬件价格 |
| Hailo | M.2 / 模块加速器 | 未持续披露 | 加速器 + Dataflow SDK | 困境中定价压力加大 |
| Ambarella | 面向 OEM 的 SoC | 未披露(按设计导入) | 摄像头 / AI SoC + SDK | 批量 OEM 定价,不是标价 |
| Blaize | 芯片、系统和 AI 服务 | 未披露;AI Services 2026 年推出 | 可编程平台 + 服务 | 转向 API / 经常性收入 |
只有 SiMa.ai 和 NVIDIA 公开可获取价格点;同业价格多按设计导入或 OEM 谈判,因此标为未披露的单元格是真实缺口。
[CP025, CP026, CP019, CP016, CP013]SiMa.ai 与主要竞争对手在能效、生态、GenAI 支持和商业规模上的能力广度对比。
定性单元格来自厂商规格和独立对比;能效数字采用厂商给出的最优条件。
[CP018, CP021, CP016, CP035, CP037]3.4 转换成本、锁定与分销力量
边缘 AI 芯片一旦被设计进产品,转换成本很高:硬件布局、散热设计、软件移植和安全认证都绑定特定芯片,因此一个芯片位胜出往往能延续整个产品世代。这种动态对 SiMa.ai 是双刃剑。公司赢下设计时会被奖励,但更多时候它利好 NVIDIA,因为多数工程团队已经投入多年 CUDA/JetPack 技能和可复用软件,使 Jetson 成为感知风险最低的默认选择。设备层面多归属有限——一台具体产品只出货一颗推理芯片——但评估阶段常见,买家会在承诺前对多个部件跑基准测试。 分销力量差距最明显。NVIDIA 和 Qualcomm 通过庞大成熟的渠道和参考设计触达客户;Ambarella 借力数十年的摄像头 OEM 关系;Blaize 触达公共部门和国防系统集成商。SiMa.ai 依赖较年轻的直销动作,并由伙伴和分销商(Macnica、Enclustra)及战略关系补充(Synopsys 做汽车工具,L&T Technology Services 做 Physical AI 方案交付,另有 Micron 存储投资)。供应获取是共同依赖:SiMa.ai 像多数无晶圆厂同行一样依赖 TSMC(Modalix 用 N6),因此代工产能分配和存储(LPDDR5)供应是共同约束,而非差异化来源。净图景是:SiMa.ai 的伙伴生态可信,但尚未形成可与在位者相比的分销护城河。[CP027, CP028, CP029, CP030, CP031, CP032]
3.5 护城河耐久性、替代风险与负面证据
SiMa.ai 的护城河论点建立在三项主张上:真实的每瓦性能领先,能降低集成成本的差异化软件优先体验,以及在机器人、工业、汽车和国防等 Physical AI 场景中的早期设计导入。每一项都可争议。效率领先很窄且容易过期——Hailo 已声称拥有相当或更好的 TOPS/W,NVIDIA 迭代激进(Jetson Thor),今天的规格优势未必能保住。软件优先叙事直接面对 CUDA;CUDA 是 AI 计算里最深、最根深蒂固的护城河,尚无挑战者在边缘把它撬开。设计导入主张也只是陈述,未量化:保留公开来源没有披露 SiMa.ai 的出货量、收入或芯片位数量。 负面证据有分量。CNBC 报道称 NVIDIA 主导 AI 芯片市场,竞争者即便获得投资人资金,也面临陡峭爬坡。Hailo 从 USD1.2 billion 独角兽跌到低于 USD500 million 的困境 SPAC,并伴随裁员,是一个直接且新近的证明:资金充足、技术可信的边缘 AI 同行,也可能无法把资本和产品转化为持久价值。Blaize 已上市但仍很小(2025 年收入 USD38.6 million);即便最健康的独立公司 Ambarella 也警告,如果设计导入低于预期,上升的研发和运营成本可能快过收入。对 SiMa.ai 的平衡判断是:产品和伙伴关系可信,但 ~USD1.4 billion 估值隐含了公开证据尚无法确认的份额获取;在这个市场里,在位者的护城河是每个挑战者面临的最大单一风险。[CP014, CP021, CP033, CP034, CP035, CP036]
| 护城河主张 | 威胁 | 严重性 | 缓解 / 尽调追问 |
|---|---|---|---|
| 能效比领先 | Hailo 可匹敌;NVIDIA 持续迭代(Thor) | 高 | 获取当前独立跨代基准测试 |
| 软件优先的 Palette 体验 | CUDA 是最深、最牢的护城河 | 高 | 量化相较 JetPack 节省的集成时间 |
| 物理 AI 设计导入 | 未披露规模;同业已有批量出货 | 高 | 索取按垂直行业划分的设计插槽数、出货量和收入 |
| 单芯片集成优势 | 对手也打包类似流程链 | 中 | 在真实设计中验证 BOM / 功耗节省 |
| 伙伴驱动分销 | 现有巨头渠道深得多 | 中 | 评估伙伴来源销售管线的转化 |
| 约 USD1.4B 估值下的资金跑道 | Hailo 说明融资充足的同业也会崩盘 | 高 | 压测现金消耗、资金跑道和下一轮条款 |
严重性反映护城河主张与公开证据之间的缺口;所有高严重性项都取决于未披露的商业数据。
[CP033, CP034, CP035, CP036, CP037, CP038]3.6 展示材料
04财务
4.1 收入模式、定价与商业化
SiMa.ai 主要靠硬件变现:销售 Modalix MLSoC 芯片、封装后的系统级模组和开发套件,并把 Palette 软件工具链(包括 Edgematic 低代码环境和 LLiMa 端侧 LLM 框架)叠加其上,而不是把软件作为独立经常性许可出售。对这个品类来说,公开定价异常透明——生产版 Modalix 系统级模组在 1,000 件量级下,8GB 版本标价 USD349、32GB 版本 USD599,开发套件 USD1,499——但标价与实际成交价、量折和 OEM 合同条款都未披露。商业化动作混合直销与分销 / 伙伴渠道(Macnica、Enclustra、ThinkRobotics),以及能缩短客户评估的生态关系(Synopsys 做汽车工具,L&T Technology Services 做方案交付)。SiMa.ai 把机器人、工业自动化、汽车、航空航天与国防、智能视觉和医疗作为收入垂直。仍不透明的是芯片、模组、套件和软件之间的收入组合,以及 Palette 是否会作为单独产品线变现;这两点会实质影响收入基底的耐久性和利润率。[CI001, CI002, CI003, CI004, CI005, CI028]
| 收入流 | 机制 | 单位 | 当前价值 / 状态 | 质量 | 尽调追问 |
|---|---|---|---|---|---|
| Modalix MLSoC 芯片 | 向 OEM / 模块厂商销售芯片 | 按芯片 | 2025 年 8 月起量产;价格未披露 | 硬件,可能复购属性低 | 获取按客户划分的单芯片 ASP 和销量 |
| 系统模块(SoM) | 直销 + 分销商硬件销售 | 按模块 | 1K 件标价 USD349(8GB)/ USD599(32GB) | 硬件,交易型 | 确认实际 ASP 与标价差异及销量 |
| 开发套件 | 直销 + 分销商销售 | 按套件 | 标价 USD1,499 | 低毛利评估 / 导入 | 确认转化为量产设计导入的比例 |
| Palette 软件套件 | 与硬件捆绑(Edgematic、LLiMa) | 捆绑 / 无单独标价 | 公开口径未作为独立许可证变现 | 当前非经常性;未来可能成为单独收入线 | 询问软件是否会单独定价 |
| 服务 / 支持 | 通过伙伴交付解决方案(LTTS) | 按项目 | 未披露;由伙伴交付 | 项目型,毛利率波动 | 索取服务收入和毛利拆分 |
| 汽车 / IP 合作 | 工具链和设计合作(Synopsys) | 按项目 | 已宣布合作;经济条款未披露 | 不确定;可能是成本分摊而非收入 | 澄清合作是否产生收入 |
收入流基于公司披露、分销商列表和独立报道整理;所有标为未披露的数值都是真实缺口,因为 SiMa.ai 不公布收入拆分。
[CI001, CI004, CI005, CI028, CI029, CI030]| 产品 | 价格 / 单位 / 合同 | 标价 / 实际 | 折扣 / 未知 | 来源 |
|---|---|---|---|---|
| Modalix SoM 8GB | 每个模块 USD349(1,000 件数量) | 仅标价;实际价格未披露 | 批量 / OEM 折扣未知 | 来源:edge-ai-vision;Analytics India Magazine |
| Modalix SoM 32GB | 每个模块 USD599(1,000 件数量) | 仅标价;实际价格未披露 | 批量 / OEM 折扣未知 | edge-ai-vision;ThinkRobotics |
| Modalix 开发套件 | 每套件 USD1,499 | 标价;分销商价格可能不同 | 套装内容不一;零售加价未知 | edge-ai-vision;ThinkRobotics |
| Modalix MLSoC 芯片 | 未公开 | 标价和实际价格均未披露 | 单芯片 ASP 和阶梯量价未知 | 公司披露(未披露) |
| Palette 软件 | 无单独价格(捆绑) | 公开口径未作为独立许可证销售 | 未来变现路径未知 | 公司披露 |
| 批量 / OEM 合同 | 谈判定价;未公开 | 仅实际价格;保密 | 条款、最低采购量和返利未知 | 公司披露(未披露) |
只有模块和开发套件标价公开;芯片和合同价格未披露,因此实际 ASP 和折扣结构仍是尽调缺口。
[CI002, CI003, CI028]SiMa.ai 的变现从芯片和模块出发,经渠道与捆绑软件流向产生收入的垂直领域;软件线仍未独立变现。
流程反映已披露的变现结构;收入结构和各收入流贡献未披露,因此以一个未解决节点呈现。
[CI001, CI002, CI003, CI030, CI038]4.2 成本结构、毛利率与资本强度
作为无晶圆厂设计公司,SiMa.ai 的成本结构主要由研发、TSMC 先进制程流片和掩膜成本(Modalix 采用 N6 工艺)以及库存驱动,而不是自有制造。每一代 MLSoC 都需要新的先进制程流片,并在产生收入前持续投入软件,因此该结构资本强度高。SiMa.ai 不披露财务,其可能的利润率路径只能用上市同行代理。Ambarella——最健康的无晶圆厂边缘 AI 可比公司——报告 2026 财年收入创纪录达 USD390.7 million(同比增长 37.2%),GAAP 毛利率 59.2%、non-GAAP 毛利率 60.7%,但仍录得 GAAP 净亏损 USD75.9 million,non-GAAP 运营费用因人力和 SoC 开发成本上升约 12.9%。另一家上市纯玩家 Blaize 披露 2025 财年收入约 USD38.6 million,经营亏损约 USD103.8 million。读数是:边缘 AI 无晶圆厂专门公司毛利率大致在 50% 中段到 60% 中段,明显低于 NVIDIA 70% 以上、由规模驱动的毛利率;达到 Ambarella 规模收入也不足以实现 GAAP 盈利。因此,即便需求环境有利,SiMa.ai 仍要走一条漫长且耗资的利润率成熟路径。[CI013, CI014, CI015, CI016, CI017, CI018]
| 指标 | 数值 / null | 置信度 | 重要性 | 尽调追问 |
|---|---|---|---|---|
| 平均销售价格(模块) | USD349-599 标价(1K 件) | 中 | 锚定每单位硬件收入 | 确认扣除折扣后的实际 ASP |
| 毛利率 | null(同业推算值约 55-65%) | 低 | 决定盈利路径 | 索取经审计或管理层口径毛利率 |
| 隐含收入 | null(未披露) | 低 | 所有估值和利润率指标的基础 | 索取收入和 YoY 增长 |
| 获客成本 / 回本周期 | null(未披露) | 低 | 销售效率和渠道经济性 | 索取 CAC、销售周期和回本参考指标 |
| 设计导入 / 销售周期 | 长(边缘芯片常态) | 低 | 决定营运资金和收入确认节奏 | 索取平均从设计导入到收入确认的时间 |
| 累计融资 | ~USD355M | 高 | 反映烧钱承受力和稀释 | 确认账上现金和净现金消耗 |
SiMa.ai 不公布单位经济;使用同业推算值(Ambarella、Blaize)框定毛利率,每个 null 都是明确尽调追问。
[CI002, CI014, CI023, CI034, CI037, CI028]用同业代理把平均售价、销货成本、毛利、运营费用串到经营结果,展示边缘 AI 芯片为什么即使规模放大仍会亏损。
该桥接采用同业代理(Ambarella FY2026),因为 SiMa.ai 未披露单位经济性;数字说明方向和量级,不是 SiMa.ai 实际值。
[CI013, CI015, CI016, CI026, CI035]4.3 公开牵引力与私人指标缺口
SiMa.ai 的公开牵引力几乎完全体现在产品里程碑、融资和伙伴关系,而非财务结果。公司不披露收入、ARR、出货量、毛利率、烧钱速度或现金跑道;第三方追踪机构只报告它经历约九到十轮融资,约十五家投资方支持。与上市可比公司相比,这一反差明显:Ambarella 披露季度收入、毛利率、现金头寸,以及约 80% 收入和超过 USD1 billion 累计收入来自边缘 AI;Blaize 提交完整 10-K,披露收入、亏损和风险因素。对 SiMa.ai 来说,缺失的私人指标——收入及增长率、毛利率、月度现金消耗和跑道、积压订单与预订订单、客户集中度——正是投资评估需要的输入,每一项也都有具体尽调路径(管理层资料室、审计报表、NDA 下的设计导入管线)。在这些数据提供前,SiMa.ai 的隐含收入只能作为代理推断,其竞争牵引力无法与上市同行逐项对比。披露缺口是本报告最大的投资判断不确定性来源。[CI011, CI012, CI019, CI021, CI023, CI029]
| 缺失的私有指标 | 对投资测算的影响 | 具体尽调路径 |
|---|---|---|
| 收入和增长率 | 无法估算业务规模或计算倍数 | 要求提供审计 / 管理口径收入和同比增长 |
| 毛利率 | 毛利率路径和盈利时间点未知 | 要求按产品线提供 COGS 和毛利率 |
| 月度烧钱和现金续航期 | 融资依赖和稀释风险无法量化 | 要求提供净烧钱、现金和现金续航月数 |
| 客户集中度 | 收入韧性和单一客户风险未知 | NDA 下要求提供前 10 大客户收入 |
| 在手订单 / 签约额 | 前瞻可见性和设计定点转化未知 | 要求提供已确认在手订单和设计定点管线 |
| 人员和运营开支结构 | R&D 强度和经营杠杆未知 | 要求提供人数、R&D 与 S&M opex 拆分 |
每行都是公开可比公司会披露、但 SiMa.ai 未披露的指标;尽调路径列给出补齐缺口的具体请求。
[CI011, CI012, CI023, CI029, CI037]毛利率基准为 SiMa.ai 可能的利润率区间定框架,对照上市边缘 AI 同业和 NVIDIA 由规模撑起的天花板。
数值为毛利率百分比;SiMa.ai 未披露毛利率,因此区间由同业推断。Blaize 区间依据其 FY2025 低收入、高亏损画像估算。
[CI015, CI019, CI022, CI035, CI016]4.4 资本充足性、融资依赖与资金用途
SiMa.ai 从种子轮到 2025 年 8 月 Series C 累计融资约 USD355 million。时间线包括 2020 年约 USD30 million Series A、2021 年约 USD80 million Series B 及多次延展、2024 年 4 月约 USD70 million 融资、2025 年 8 月 USD85 million Series C(Maverick Capital 领投,StepStone Group 加入,且被描述为超额认购),以及约 2026 年 4 月报道的一笔 Micron 战略投资;该投资独立于主轮 Series C。公司称 Series C 资金将用于放大 Modalix 生产、全球扩张,并深化汽车和 Physical AI 商业化。由于 SiMa.ai 不披露在手现金、月度现金消耗或跑道,资本充足性无法直接衡量;Series C 后约八个月即出现战略融资,说明融资更像事件驱动——绑定生产放量和伙伴关系——而非对应一个已披露的现金耗尽日期。融资依赖风险真实存在,上市同行 Blaize 的 2025 财年 10-K 中持续经营警示就是例证。投资评估应把未披露现金消耗和依赖周期性新资本视为资本结构核心问题。[CI006, CI007, CI008, CI009, CI010, CI020]
| 维度 | 已披露数值 / 状态 | 估算 / 参考值 | 置信度 | 尽调追问 |
|---|---|---|---|---|
| 累计融资 | ~USD355M(种子轮至 Series C) | 公司和数据追踪方已确认 | 高 | 逐轮与股权结构表核对 |
| 最新轮次(Series C) | USD85M,Aug 2025,Maverick 领投 | 公司称超额认购 | 高 | 确认投后估值和新投资人条款 |
| 后续融资 | Micron 战略投资(约 Apr 2026) | 金额未披露 | 中 | 确认规模、结构和稀释 |
| 账面现金 | 未披露 | Unknown | 低 | 要求提供当前现金及等价物 |
| 月度烧钱 / 现金续航期 | 未披露 | 未知;同业烧钱较重 | 低 | 要求提供月度净烧钱和现金续航月数 |
| 计划资金用途 | 扩大 Modalix 量产;全球扩张;汽车 GTM | 公司披露 | 中 | 确认 capex 与 opex 分配 |
融资金额和轮次领投方有充分交叉印证,但现金、烧钱和现金续航期完全未披露,因此资本充足性只能推断, 不能直接测算。
[CI007, CI008, CI009, CI010, CI027, CI036]SiMa.ai 的资本消耗位置、可能资金来源,以及每项资本需求的披露状态。
资本需求行根据无晶圆厂模式和已披露的资金用途表述推断;所有美元金额仍未披露,因此图中展示结构而非规模。
[CI007, CI013, CI027, CI032, CI036]4.5 财务结论与尽调阻塞项
基于现有证据,SiMa.ai 的财务画像是一家资本充足但尚未披露关键数据的后期硬件初创公司;其收入质量、利润率路径和资本强度无法独立核验。正面因素真实存在:累计融资约 USD355 million,投资者需求仍在(超额认购 Series C 和后续 Micron 战略投资),产品定价透明,第二代芯片已进入量产阶段。负面因素则是结构性且可量化的。上市可比公司显示,即便收入达到 USD390.7 million、毛利率约 ~60%,无晶圆厂边缘 AI 同行 Ambarella 仍出现 GAAP 净亏损;更小的 Blaize 则有持续经营警示——说明这一品类资本需求大、利润率受限。SiMa.ai 据报 USD1.4 billion 估值未获公司确认;在没有披露收入的情况下,隐含收入倍数无法计算。因此尽调阻塞项具体且可关闭:审计或管理口径收入及增长、毛利率、月度现金消耗和跑道、积压订单、客户集中度。在这些材料提供前,任何投资判断都必须依赖同行代理,并把该估值视为风险投资账面标记,而不是经财务验证的数字。[CI022, CI023, CI024, CI025, CI034, CI035]
4.6 展示材料
05产品与技术
5.1 产品定义和产品族图谱
SiMa.ai 的产品是 Modalix,一款第二代 MLSoC,可在单颗芯片上运行完整 Physical AI 流程——计算机视觉、传统 CNN、Transformer、大语言模型和多模态生成式 AI——而不是把工作负载分散到 CPU 加加速器板卡上。放到客户工作流里,它用一个高能效设备替代离散 GPU 或加速器加主处理器:设备摄取摄像头和传感器数据,本地推理,并在机器人、车辆、无人机和工业机器中驱动实时决策。商业产品族覆盖出售给 OEM 和模组厂的裸 MLSoC 芯片、8GB 和 32GB 版本封装 Modalix 系统级模组、用于评估和原型开发的 Modalix DevKit 3.0,以及软件层:Palette Neat SDK(Model Compiler、Neat Library 和 sima-cli 工具)、Edgematic 基于浏览器的低代码流程构建器、LLiMa GenAI 运行时,以及公开 GitHub 模型库,内含预优化 PyTorch 和 ONNX 模型。模组在引脚和软件上兼容 SiMa.ai 第一代 MLSoC,也兼容流行 NVIDIA SoM,让买家能在现有载板设计中评估 Modalix。最终产品是一套硬件加软件方案,卖点是单芯片多模态集成,以及从模型到部署式边缘应用的低摩擦开发路径。[CE001, CE012, CE013, CE014, CE015, CE017]
| 模块 / 产品线 | 用户 | 状态 / 成熟度 | 差异化 | 尽调追问 |
|---|---|---|---|---|
| Modalix MLSoC 芯片 | OEM、模块厂商 | 已量产(Aug 2025) | 单芯片多模态边缘 AI | 确认良率、分档和供货条款 |
| Modalix SoM(8GB / 32GB) | 机器人 / 工业集成商 | 已量产;价格已列出 | 引脚 / 软件兼容 NVIDIA SoM | 确认出货量和载板 |
| Modalix DevKit 3.0 | 开发者、评估方 | 可通过分销商获取 | 快速评估和基准测试 | 确认转化到量产项目的附着率 |
| Palette Neat SDK | ML / 嵌入式开发者 | 已可用;Python/C++ API、sima-cli | Model Compiler + Neat Library 工具链 | 评估 SDK 稳定性和采用度 |
| Edgematic 低代码平台 | 应用开发者 | 托管在 AWS 云上;较新的层 | 浏览器管线构建器 | 评估成熟度、宣传主张和实际使用 |
| LLiMa GenAI 运行时 | GenAI / LLM 开发者 | 最新层;端侧 LLM/VLM | 端侧生成式推理 | 验证支持模型和性能 |
| GitHub 模型库 | 开发者 | 公开(SiMa-ai/models) | 预优化 PyTorch/ONNX 模型 | 评估覆盖广度和外部贡献 |
产品族来自公司文档、开发者门户、分销商上架信息和独立报道;成熟度标签区分量产芯片和较新的软件层。
[CE012, CE013, CE014, CE015, CE018, CE036]| 用户任务 | 当前工作流 | SiMa.ai 方案 | 可量化收益 | 限制 |
|---|---|---|---|---|
| 在机器人上部署视觉 AI | 独立 GPU + 主机 CPU 板 | 单颗 Modalix MLSoC 管线 | 更低功耗(<10W)和更简单的板卡设计 | Modalix 没有公开 MLPerf 验证 |
| 运行 PyTorch/ONNX 模型 | 手工移植和量化 | Palette Model Compiler + 模型库 | 模型上设备路径更快 | 模型覆盖深度未披露 |
| 可视化构建边缘管线 | 手写 GStreamer 管线 | Edgematic 低代码构建器 | 几分钟内组装并部署 | Edgematic 较新,验证不足 |
| 在边缘运行 GenAI/LLM | 云端推理,或没有端侧选项 | LLiMa 端侧 GenAI 运行时 | 端侧隐私和低延迟 | 最新层;支持模型不清楚 |
| 购买前基准测试 | 只能用实体评估板 | 通过 Edgematic 使用云端托管 MLSoC | 远程获取吞吐、延迟、功耗 KPI | 云端结果可能不同于现场 |
| 复用现有载板设计 | 按供应商重做模块设计 | 引脚 / 软件兼容的 SoM | 相对 NVIDIA SoM 可插拔评估 | 仍需要软件移植投入 |
用例综合自开发者门户、Edgematic 文档和产品报道;收益来自供应商定位,限制指出公开验证缺口。
[CE013, CE015, CE016, CE017, CE019, CE028]Modalix 芯片与软件层的成熟度、差异化和依赖风险:硬件达到量产级,GenAI 与低代码工具更新、更缺少验证。
评级是基于开发者门户、文档和产品报道作出的证据判断;软件层成熟度从发布时间推断,不来自已披露使用指标。
[CE011, CE014, CE021, CE031, CE033]5.2 Modalix 架构和运行模式
Modalix 是一颗异构系统级芯片,采用 TSMC N6 工艺,封装尺寸为 25mm x 25mm。计算部分围绕专用机器学习加速器展开,提供 25、50、100、200 INT8 TOPS 档位,支持 INT8、INT16、BFLOAT16 精度,覆盖 CNN、Transformer、LLM 和生成式推理;应用处理器复合体则由 8 个 Arm Cortex-A65 核心组成,运行频率 1.4GHz。视觉和信号处理由四核 Synopsys ARC EV74 计算机视觉单元和 Arm Mali-C71AE 图像信号处理器承担,安全片上网络 以高带宽连接各模块。内存包括模组上最高 32GB LPDDR5 加 8MB 片上 SRAM,连接能力包括 PCIe Gen5 x8、4 个 10GbE 端口、4 路 MIPI CSI-2 摄像头接口,以及 H.264、H.265、AV1 最高 4Kp60 的硬件视频编解码。安全由带信任根和 OTP 密钥支持的硬件安全启动兜底。SiMa.ai 称,50-TOPS 配置在低于 10W 的功耗包络内,单位功耗性能超过替代方案 10 倍以上——Modalix 追求效率优先,而不是峰值吞吐,这正是它相对高功耗主流模组的定位核心。[CE002, CE003, CE004, CE005, CE006, CE007]
| 层 / 组件 | 角色 | 依赖 | 风险 |
|---|---|---|---|
| 制程 / 封装 | TSMC N6,25mm x 25mm 裸片 | TSMC 晶圆代工产能分配 | 晶圆厂产能和节点成本 |
| ML 加速器 | 25-200 INT8 TOPS;支持 INT8/INT16/BFLOAT16 | SiMa.ai 自研设计 | Modalix 代际尚无基准测试 |
| CPU 集群 | 8x Arm Cortex-A65,频率 1.4GHz | Arm IP 授权 | 第三方 IP 依赖 |
| 视觉 / ISP | Synopsys ARC EV74 CVU;Arm Mali-C71AE ISP 视觉 / ISP 模块 | Synopsys 与 Arm IP | 第三方 IP 依赖 |
| 内存 | 最高 32GB LPDDR5;8MB SRAM | LPDDR5 内存供应商 | 内存供应和定价 |
| I/O 和连接 | PCIe Gen5 x8;4x 10GbE;4x MIPI CSI-2 接口 | 标准接口 IP | 集成复杂度 |
| 安全 | 硬件安全启动、信任根、OTP 密钥 | 片上 | 现场验证未公开 |
| 互连 | 安全高带宽片上网络 | 自研 | 性能未经独立验证 |
架构来自 Macnica 产品数据表和独立产品报道;风险集中在第三方 IP、晶圆代工依赖,以及 Modalix 缺少独立基准测试。
[CE003, CE004, CE005, CE006, CE007, CE008]Modalix 的产品架构是一套四层堆栈:从 TSMC 制造的芯片与异构计算起步,向上到 Palette 软件工具链和面向开发者的工具。
层级结构综合自 Macnica 数据表、开发者门户和 Edgematic 文档;公开规格之外的内部微架构细节未披露。
[CE003, CE006, CE012, CE015]5.3 部署、集成与路线图
部署围绕 Palette Neat SDK 和 Edgematic 平台展开。开发者带来 PyTorch 或 ONNX 模型(也可以从 GitHub 模型库选择),用 Palette Model Compiler 编译,再通过 Edgematic 组装流水线;Edgematic 是基于浏览器的低代码构建器,后端采用 GStreamer,可在云托管的 MLSoC 板卡上测量真实吞吐、延迟和功耗,并部署到 Modalix 设备。Edgematic 托管在 AWS 上,集成 Amazon SageMaker;开发者门户(developer.sima.ai)发布了由 sima-cli 驱动的、从安装到 GenAI 的自包含上手路径。上一代模组和 NVIDIA 模组的引脚、软件兼容性,以及解决方案和交付伙伴生态,降低了集成难度。路线图已经从第一代 MLSoC(2024 年 3 月列入 MLCommons MLPerf Inference 4.0)推进到 2024 年 9 月发布 Modalix 家族,再到 2025 年 8 月生产硅片和模组落地;LLiMa GenAI 运行时和 Edgematic 云工具是最新加入的层。尚未公开的是 Modalix 代际的 MLPerf 结果或独立可靠性数据,因此买方仍需直接在硬件上验证集成风险和真实吞吐。[CE015, CE016, CE019, CE020, CE028, CE030]
| 日期 / 阶段 | 功能 / 里程碑 | 状态 | 含义 | 来源 |
|---|---|---|---|---|
| 2022 | 第一代 MLSoC | 已出货 | 搭起平台 | EE Times;SiliconANGLE |
| March 2024 | MLPerf Inference 4.0 上榜(一代) | 已完成 | 一代基准可信度 | SiliconANGLE;Macnica |
| September 2024 | Modalix 产品族发布 | 已发布 | 多模态、可承载 GenAI 的路线图 | IOT Insider;TMCnet |
| August 2025 | Modalix 芯片量产 | 量产 | 商业化可用 | EE Times;PR Newswire |
| September 2025 | Modalix SoM(8GB / 32GB)发布 | 已可用 | 可部署模块,价格已列出 | 来源:Edge AI and Vision Alliance |
| 2025-2026 | Edgematic AWS + LLiMa GenAI 运行时 | 最新层 | 端侧 GenAI 和低代码部署 | 来源:developer.sima.ai;AWS |
时间线来自新闻报道和开发者门户;最新层(Edgematic 云、LLiMa GenAI)的独立验证最少。
[CE001, CE002, CE013, CE016, CE020, CE038]开发者从训练好的模型出发,经 Palette 编译和 Edgematic 评估,最终部署为运行在 Modalix 芯片上的应用。
流程反映已记录的 Palette/Edgematic 工作流;真实集成工作量和移植时间尚无独立量化。
[CE013, CE015, CE017, CE019, CE028]5.4 差异化、IP 与供应依赖
SiMa.ai 的差异化有三层:单芯片多模态集成(视觉与生成式模型在同一颗芯片上运行)、效率优先的设计点(10W 以下实现 50+ TOPS,并宣称单位功耗性能优势达 10x),以及通过 Palette 和 Edgematic 打造的软件优先开发体验,用来降低拼装离散流水线的集成摩擦。关键在于,这种差异化锚定在授权和外包模块上,而不是完全自有硅资产:CPU 复合体使用 Arm Cortex-A65 核心,计算机视觉单元使用 Synopsys ARC EV74 IP,芯片由 TSMC 以 N6 工艺制造,内存依赖 LPDDR5 供应商。SiMa.ai 还在 2024 年 12 月与 Synopsys 达成汽车边缘 AI 开发战略合作,与 L&T Technology Services 合作交付物理 AI 解决方案,并借助 AWS 做云托管开发。因此,护城河不太来自专有制程或 CPU IP,而更多来自集成、编译器和运行时软件,以及开发者生态——这些资产真实存在,但可复制,也依赖第三方。SiMa.ai 公开维护 GitHub 组织(模型库和 Palette Neat),可作为开发者采用信号,只是外部社区贡献深度尚未建立。[CE011, CE023, CE024, CE025, CE026, CE029]
Modalix 的交付依赖第三方 IP(Arm、Synopsys)、单一晶圆代工厂(TSMC)、内存供应商,以及用于云端开发的 AWS,这些依赖最终汇入软件平台和客户应用。
依赖项从公开规格和合作伙伴披露推断;确切内存供应商和云承诺并未完全披露。
[CE008, CE016, CE023, CE024, CE032]5.5 信任、安全、质量与验证缺口
在信任和安全上,Modalix 实现带信任根和 OTP 密钥配置的硬件安全启动,其片上网络也被描述为安全互连——这些控制适合汽车和国防部署。Edgematic 的云托管评估定位为企业级,强调数据隐私和控制,运行在 AWS 上;平台支持端侧推理,使敏感数据在运行时留在本地。ONNX 和主流框架支持降低了模型层锁定,2024 年 12 月与 Synopsys 的汽车合作也释放了满足汽车级工具链预期的意图。实质性的质量和验证缺口在基准测试:SiMa.ai 第一代 MLSoC 列入 MLPerf Inference 4.0,但截至 2026 年中,尚无 Modalix 代际的公开 MLPerf 结果,也没有独立的现场可靠性或功能安全认证数据。GenAI(LLiMa)和 Edgematic 是技术栈中最新、最少被验证的层。对承销方而言,安全架构纸面上可信,但量产 Modalix 代际的独立性能、可靠性和安全验证仍是未完成的尽调项。[CE010, CE016, CE020, CE021, CE028, CE032]
| 控制 / 指标 | 状态 | 范围 | 缺口 |
|---|---|---|---|
| 硬件安全启动 | 已实现(信任根、OTP 密钥) | 端侧固件完整性 | 无公开认证细节 |
| 数据隐私(Edgematic) | 定位为 AWS 上的企业级能力 | 云端评估 + 端侧推理 | 未披露第三方审计 |
| 汽车工具链 | 与 Synopsys 合作(Dec 2024) | 汽车边缘 AI 开发 | 无公开 ASIL/ISO 26262 认证 |
| 供应链完整性 | Fabless 模式,TSMC N6 | 依赖晶圆代工 | 单一晶圆厂集中 |
| 性能验证(MLPerf) | 一代已列入;Modalix 未公开 | 基准可信度 | 无 Modalix 代际 MLPerf 结果 |
| 模型标准支持 | ONNX 和主流框架 | 模型可移植性 | 覆盖广度未披露 |
控制项来自公司和合作伙伴披露;最大缺口是没有 Modalix 代际 MLPerf 结果,也没有公开功能安全认证。
[CE010, CE016, CE020, CE021, CE025, CE037]5.6 图表
06客户
6.1 客户基础细分、垂直领域与渠道
SiMa.ai 面向「嵌入式边缘」——介于云数据中心和低功耗个人设备之间的计算层——目标工作负载功耗大致在 5W–25W。公司公开声明的目标垂直领域包括机器人、汽车和出行、工业自动化、航空航天与国防、智能视觉和医疗健康;早期公司表述还把医疗健康、智能零售、自动驾驶汽车、政府和机器人列为优先应用。买方通常是把 Modalix MLSoC 或系统级模组嵌入机器人、摄像头、车辆或工业机器的 OEM 或系统集成商,最终用户则是运营这些机器的企业,因此付款方和使用方通常不同。商业化路径 混合了直接企业销售和分销渠道——Macnica(日本 / APAC)、Enclustra、ThinkRobotics 转售模组和开发套件——以及工程 / 解决方案伙伴,如 L&T Technology Services(出行、医疗健康、工业自动化、机器人)、VVDN(设计和制造)、Synopsys(汽车工具链)和 AWS(Edgematic 云集成)。地域上,公司总部在 San Jose,但明确将 2025 年 Series C 融资用于拓展韩国、日本、欧洲和美国。由于 SiMa 不披露收入区间、客户规模或垂直领域收入拆分,细分只能根据合作和产品证据定性描述,无法按收入权重衡量。[CU001, CU002, CU003, CU004, CU005, CU006]
| 垂直 / 细分 | 买方 / 用户 / 付款方 | 代表性用例 | 具名或报道证据 | 渠道 | 披露质量 |
|---|---|---|---|---|---|
| 机器人 | OEM 采购;企业运营方使用 | AMR、人形机器人、SLAM / 导航、割草机器人 | STIGA(割草机器人);AMR 方案简介 | 直销 + 分销商(ThinkRobotics) | 具名合作;无出货量 |
| 工业自动化 | 机器 / 设备 OEM | AI 驱动激光器、工厂感知 | TRUMPF(AI 驱动激光器,Oct 2024) | 直销 + 合作伙伴(LTTS) | 具名合作;经济条款未披露 |
| 汽车 / 出行 | Tier-1 / OEM | MLSoC ONE 上的 IVI、AD/ADAS | LTTS 合作;Synopsys 汽车工具链 | 合作伙伴 / 解决方案渠道 | 合作具名;无设计定点数量 |
| 航空航天与国防 | 政府 / 国防集成商 | 离线自主、无人机、战场边缘 | 创始人 / 国防定位(Defense Disruptors) | 直销 | 仅定位;无具名项目 |
| 智能视觉 | 摄像头 / 视觉系统 OEM | 智能摄像头端侧推理 | 公司垂直定位 + Modalix SoM | 直销 + 分销商(Macnica) | 声称覆盖该垂直;无具名客户 |
| 医疗健康 | 医疗设备 OEM | 医疗设备边缘推理 | LTTS 医疗健康范围;公司垂直定位 | 合作伙伴渠道 | 垂直和伙伴具名;无具名客户 |
细分来自公司披露、合作伙伴新闻稿和独立报道;每个“无出货量 / 未披露”单元格都是真实缺口, 因为 SiMa 未披露分垂直收入或客户数量。
[CU001, CU013, CU014, CU015, CU016, CU033]SiMa.ai 面向嵌入式边缘 OEM 买家的五阶段客户旅程,从垂直认知到拥护,反映 STIGA、TRUMPF 与 LTTS 合作中可见的设计导入关系模式。
[CU002, CU005, CU013, CU024, CU027]6.2 采用与部署轨迹
SiMa.ai 的采用故事几乎完全通过产品里程碑、合作关系和设计导入动能来呈现,而不是通过已披露的出货或客户数量。独立报道(DatacenterDynamics)称,公司客户数在 2023 到 2024 年间大致翻倍;到 2024 年中,第一代 MLSoC 硅片已交到客户手中,第二代 Modalix 样片随后在 2024 年 Q4 跟进。Modalix 于 2025 年 8 月进入量产供应,Modalix 系统级模组也在 2025 年公开发布,并在 2026 年 4 月获得 Edge AI + Vision Alliance 2026 Product of the Year 的「最佳 Edge AI 板卡」奖——这是成熟度的外部验证,但不是出货量指标。部署漏斗偏早期:许多关系是宣布型战略合作(STIGA、TRUMPF、L&T Technology Services、VVDN),或通过开发套件和系统级模组进行评估播种;独立可验证的大批量生产项目较少。公司把 Modalix 描述为与现有 GPU 模组引脚和软件兼容,以缩短评估、降低切换摩擦,意在加快从试点到量产的迁移。缺失的是已披露活跃部署数、出货量、地点或利用率——因此轨迹看起来在改善,但仍未量化。[CU007, CU008, CU009, CU010, CU011, CU012]
| 里程碑 / 时期 | 采用信号 | 阶段 | 来源依据 | 是否量化? |
|---|---|---|---|---|
| 2023 至 2024 | 据报道客户数大约翻倍 | 早期商业化 | DatacenterDynamics 报道 | 仅方向性(无绝对数量) |
| Mid-2024 | 一代 MLSoC 已到客户手中;二代样品 Q4 2024 | 送样 / 评估 | DatacenterDynamics;公司 | 否;仍处送样阶段 |
| August 2025 | Modalix 进入量产可用 | 量产爬坡 | 公司 / PR Newswire / SiliconAngle | 未披露出货量 |
| September 2025 | LTTS 战略伙伴关系发布 | 生态扩张 | LTTS 新闻稿;Robotics Business News | 否;合作关系 |
| 2026 年 2–3 月 | STIGA 机器人割草机战略合作 | 旗舰设计导入 | STIGA 公司;行业媒体 | 未披露出货量 |
| 2026 年 4 月 | Modalix SoM 获 Edge AI + Vision 2026 年度产品奖 | 外部验证 | PR Newswire;BriefGlance | 奖项,不是出货量 |
这条轨迹按里程碑推进;没有单元格披露出货量、活跃账户数或利用率,因为 SiMa 未披露这些指标, 因此漏斗只能看方向,不能量化。
[CU007, CU008, CU009, CU010, CU011, CU017]截至 2026 年中,按部署成熟度阶段梳理 SiMa.ai 公开已知客户与合作伙伴关系。计数为近似值,且仅限公开宣布或记录的关系,因为 SiMa 未披露账户总数。
[CU008, CU010, CU011, CU014, CU016]6.3 具名客户证明与参考质量
最强的具名证明是 STIGA S.p.A.,这家欧洲领先的园林和户外设备制造商宣布战略合作(日期为 2026 年 3 月,最早于 2026 年 2 月报道),将 SiMa.ai 的 Modalix MLSoC 平台嵌入下一代家用和商用机器人割草机,STIGA 管理层也公开背书该合作。在工业自动化领域,SiMa 披露了 2024 年 10 月与 TRUMPF 的合作,使用其芯片打造 AI 驱动的激光器。在航空航天与国防领域,SiMa 通过创始人访谈(如 Defense Disruptors 系列)围绕断连、战场和自主无人机环境定位,但具体国防项目名称和规模并未公开。L&T Technology Services(2025 年 9 月)是覆盖车载信息娱乐、AD/ADAS、工业自动化、机器人和医疗健康的解决方案 / 工程伙伴,基于 MLSoC ONE 平台;VVDN 则是设计 / 制造伙伴。参考质量参差不齐:STIGA 和 TRUMPF 的关系具名、有日期,并有第三方报道佐证;但多数仍是合作或 设计导入公告,而不是带有量化结果(出货、ROI、部署规模)的量产案例。SiMa 自身也指出,并非所有客户都已公开,这限制了独立验证深度。[CU013, CU014, CU015, CU016, CU017, CU018]
| 客户 / 伙伴 | 垂直领域 | 关系类型 | 日期 | 量产还是试点 | 引用 / 结果质量 |
|---|---|---|---|---|---|
| STIGA S.p.A.(客户) | 机器人(机器人割草机) | 战略产品合作 | 2026 年 2–3 月 | 下一代产品设计导入(量产前) | 具名 + 有日期 + 高管背书;无出货量 |
| TRUMPF | 工业自动化 | 芯片供应 / 共同开发 | 2024 年 10 月 | 开发阶段(AI 驱动激光器) | 具名 + 有日期;无出货量或结果指标 |
| L&T Technology Services(LTTS,工程伙伴) | 出行、医疗、工业、机器人 | 解决方案 / 工程伙伴 | 2025 年 9 月 | 围绕 MLSoC ONE 共同开发 | 具名伙伴;无终端客户出货量 |
| VVDN Technologies | 跨垂直领域 | 设计 / 制造伙伴 | 2025 年报道 | 支撑生产规模化 | 具名伙伴;按角色证明,不是部署 |
| 航空航天 / 国防(未具名) | 航空航天与国防 | 定位 / 目标垂直领域 | 2024-2026 | 无具名项目 | 仅定位;无可验证项目 |
| Synopsys | 汽车 | 工具链 / IP 协作 | 2024 年 12 月 | 开发协作 | 具名;成本共担不是终端客户证明 |
证明材料偏向战略伙伴关系和设计导入公告,而不是带量化结果的量产案例;SiMa 也表示并非所有客户都会公开, 因此独立验证受限。
[CU013, CU014, CU015, CU016, CU018, CU019]针对 SiMa.ai 五个最具体、公开有记录的客户与合作伙伴关系,按证据维度给出的证据质量矩阵。评分是对公开证据质量的序数判断,不是审计数字。
语气反映公开证据质量的 0-10 定性读取(正向 = 强,中性 = 中等,负向 = 弱)。未经审计。
[CU013, CU014, CU015, CU018, CU020, CU037]6.4 留存、耐久性与满意度信号
SiMa.ai 不披露正式留存指标——没有净收入留存(NRR)、总收入留存(GRR)、流失、续约率、合同期限或客户分组数据——因此耐久性只能从业务结构和间接信号推断。两个结构性因素说明,一旦拿下 设计赢单,粘性会较强:边缘硅片设计导入周期长、成本高;客户把 Modalix 认证进机器人、摄像头或车辆平台后,切换会面临高昂再工程成本,这在嵌入式半导体中通常带来多年平台生命周期。SiMa 又通过全栈软件层(Palette、Edgematic、LLiMa 端侧 LLM 运行时)和模型库扩大集成面,使切换更具破坏性。正面外部信号包括 2026 Product of the Year 奖,以及 2023 到 2024 年客户数翻倍的报道,暗示至少存在一定复购和扩张需求。反向信号同样重要:没有公开续约,没有带量化满意度的参考客户群,没有披露客户留存; 第二代产品仍很年轻(Modalix 仅自 2025 年 8 月量产),尚未证明完整的多年续约客户分组。因此,耐久性在结构上可信,但经验数据尚未证明。[CU021, CU022, CU023, CU024, CU025, CU026]
| 留存维度 | 披露值 | 结构性 / 间接信号 | 为什么重要 | 尽调要求 |
|---|---|---|---|---|
| 净收入留存(NRR) | 空值(未披露) | 无公开数据 | 衡量扩张耐久性的核心指标 | 在 NDA 下要求提供分 cohort 的 NRR/GRR |
| Logo / 客户流失 | 空值(未披露) | 报告的 2023–2024 年客户数约 2x,意味着净增 | 区分真实牵引力和一次性赢单 | 要求按 cohort 提供 logo 留存和流失 |
| 合同期限 / 续约 | 空值(未披露) | 嵌入式设计导入意味着多年平台生命周期 | 决定收入可预测性 | 要求提供平均合同期限和续约率 |
| 切换成本 / 锁定 | 未量化 | 全栈 Palette 软件抬高重新工程化成本 | 高切换成本支撑留存 | 确认使用 Palette/Edgematic 的客户占比 |
| 满意度 / 推荐客户 | 奖项 + STIGA 背书 | 2026 年度产品奖;STIGA 管理层引述 | 说明质量感知,不等于留存 | 要求提供可联系的推荐客户 |
SiMa 未披露 NRR、GRR、流失或续约数据;这里所有留存判断都来自业务结构和间接信号, 不是公司披露指标,应视为尚未证实。
[CU021, CU022, CU023, CU024, CU025]SiMa.ai 主要客户分组的留存信号矩阵。由于 SiMa 不披露队列留存率,单元格给出定性信号强度;公开记录无法判断之处留空,而不是填入已披露的 NRR/GRR。
单元格把嵌入式平台生命周期预期与公开合作证据合并判断;属于定性信号解读,不是已披露留存率。
[CU021, CU022, CU024, CU026]6.5 扩张、集中度与渠道依赖
SiMa.ai 的扩张动作是典型的先落地再扩张,横跨垂直领域和地域:先拿下初始设计导入(如 STIGA 机器人割草机、TRUMPF 工业激光器),再把同一套硅片加软件平台扩展到相邻用例(AMR、人形机器人、ADAS、智能视觉、医疗健康)和新地区(韩国、日本、欧洲、美国),资金来自 2025 年 8 月 Series C。集中度风险无法直接衡量,因为 SiMa 不披露单客户收入;但定性图景显示,公司高度依赖早期旗舰合作和伙伴 / 分销渠道:L&T Technology Services 和 VVDN 承担解决方案交付与制造,Macnica、Enclustra、ThinkRobotics 承担分销,因此相当一部分商业化路径经过中介,增加利润分成,也削弱直接客户控制。SiMa 保守的目标市场采购摩擦显著——汽车、国防和工业买方认证周期长,在承诺放量前要求功能安全和可靠性证据。公司通过让 Modalix 与既有模组引脚和软件兼容来缓解切换和采用摩擦;但反过来,同样的兼容性也允许客户双源采购,或回到 NVIDIA 等根深蒂固的供应商。结论:扩张可选性很宽,但集中度和渠道依赖仍是未量化的尽调缺口。[CU027, CU028, CU029, CU030, CU031, CU032]
| 风险 / 扩张维度 | 当前判断 | 严重性 | 依据 | 尽调要求 |
|---|---|---|---|---|
| 先落地再扩张 | 先设计导入,再进入相邻用例 + 新区域 | 正向 | STIGA/TRUMPF 赢单;Series C 用于地理扩张 | 要求提供单账户随时间扩张收入 |
| 头部客户集中度 | 无法衡量;可能依赖少数旗舰客户 | 重大 | 未披露单客户收入 | 确认是否有客户收入占比 >20% |
| 渠道 / 伙伴依赖 | LTTS、VVDN、Macnica、Enclustra 承担 GTM | 中等 | 伙伴 / 分销商新闻稿和列表 | 确认直销与渠道收入拆分及利润率 |
| 采购摩擦 | 汽车 / 国防 / 工业认证周期长 | 中等 | 目标垂直领域属性;安全要求 | 要求提供从设计导入到收入的平均周期 |
| 双供 / 回退风险 | 引脚 / 软件兼容性是双刃剑 | 中等 | 兼容性让买方可回退到 NVIDIA | 确认独供与双供设计赢单占比 |
集中度和渠道依赖只能定性判断,因为 SiMa 未披露单客户或渠道收入;同一套 GPU 引脚 / 软件兼容性 一方面降低采用门槛,另一方面也降低客户切回去的门槛。
[CU027, CU028, CU029, CU030, CU031]07风险
7.1 按严重程度排序的风险概览
SiMa.ai 是一家尚未盈利、硬件驱动的半导体公司,面对根深蒂固的既有龙头,因此风险更多压在市场、资本和执行暴露上,而不是短期法律责任。按剩余暴露排序(可能性乘以影响,扣除缓释):最高严重度风险是竞争替代——NVIDIA Jetson 平台和 CUDA/JetPack 软件护城河,再加上资本充足的挑战者,可能把 SiMa 限制在小众市场。第二是融资和资本强度风险:无晶圆厂边缘 AI 玩家烧钱很重(Blaize 经营亏损接近 USD103.8 million,并发出持续经营警告;Ambarella 即便收入接近 USD391 million,仍有 GAAP 净亏损),而 SiMa 不披露财务,因此资金续航和盈利路径无法验证,并依赖反复融资。第三是供应链集中:Modalix 依赖 TSMC N6 工艺,把产能集中在台湾。第四是监管暴露:BIS 出口管制限制中国市场准入,汽车、国防和医疗健康认证关卡拖慢收入。第五是创始人主导团队的执行和关键人风险,公开材料对更广泛创始团队的佐证有限。第六是技术验证风险:没有公开的 Modalix 代际 MLPerf 基准来独立确认单位功耗性能主张。下方表格和热力图按可能性、影响、缓释成熟度和剩余暴露对每项风险排序。[CR001, CR002, CR003, CR004, CR005, CR006]
| 风险簇 | 缓释措施 | 监测指标 | 论点失效 / 终止标准 |
|---|---|---|---|
| 竞争替代 | 加深 Palette 软件护城河;锁定多年设计赢单 | 设计赢单到量产的转化;独立基准 | 旗舰设计导入输给 NVIDIA,或下一轮融资前仍无量产收入 |
| 融资 / 资本强度 | Series C 与 Micron 支持;烧钱克制 | 现金跑道;烧钱倍数;融资条款 | 降轮,或现金跑道耗尽前无法再融资 |
| 供应链集中 | 认证备选晶圆厂 / OSAT;建立库存缓冲 | 单一来源敞口;交期 | TSMC 产能配额丢失,或台海扰动 |
| 监管 / 资质 | 遵守出口管制;推进 ISO 26262 / FDA | 认证里程碑;BIS 执法行动 | 因缺少认证被目标垂直行业挡在门外 |
| 执行 / 关键人物 | 补强 CEO 之外的管理梯队 | 核心高管留任;团队交叉验证 | 创始人兼 CEO 离任,且没有可信接班人 |
| 技术验证 | 发布独立 Modalix 基准测试 | MLPerf 或第三方审计结果 | 独立测试显著低于公司口径 |
多数缓释措施还在早期,取决于 SiMa 尚未披露的信息;每条否决条件都是投资后可以跟踪的可观察事件。
[CR035, CR036, CR037, CR038, CR039]SiMa.ai 六大主要风险集群的严重度热力图,从发生可能性、影响、当前缓释成熟度和残余敞口打分。分数是基于公开证据的定性序位判断,不是经审计概率。
色调表示定性严重度(负向 = 高风险 / 弱缓释,中性 = 中等,正向 = 较低风险 / 较强缓释)。不是定量模型。
[CR001, CR002, CR014, CR019, CR028, CR037]7.2 监管与法律风险
SiMa.ai 所处的是硬件中监管最重的角落之一。美国商务部工业与安全局(BIS)自 2022 年以来不断升级先进计算半导体和制造设备出口管制,并在 2026 年成文化许可证审查流程,使中国 / 澳门准入保持严格受限,增加客户尽调、出货比例和报告义务,并且(据法律分析)对受管制芯片叠加新关税和认证要求。对于面向全球机器人、工业和汽车客户的边缘 AI 设计商,这些管制既缩小中国可服务市场,也抬高合规成本和转移责任风险。在汽车领域,ISO 26262 功能安全认证(ASIL 等级,其中 ASIL C/D 要求锁步 CPU、ECC、FMEDA 和可追溯安全架构)是设计导入的事实门槛;AI 特定指南(ISO/PAS 8800、SOTIF/ISO 21448)仍在成熟中——SiMa 尚未公开证明自己跨过该门槛。医疗健康(FDA)和航空航天 / 国防(ITAR、采购安全)又施加进一步认证。法律层面,SiMa 发布了隐私政策(最后更新于 2025 年 11 月)管理数据处理;作为无晶圆厂设计商,它也面对半导体行业标准的专利侵权和 IP 诉讼暴露,尽管目前未公开发现针对 SiMa 的重大诉讼。因此,主导监管风险是市场准入和认证摩擦,而不是当下的主动执法。[CR007, CR008, CR009, CR010, CR011, CR012]
| 风险 | 制度 / 依据 | 发生可能性 | 影响 | 缓释成熟度 | 剩余敞口 |
|---|---|---|---|---|---|
| 中国市场准入受损 | 美国 BIS 先进计算出口管制 | 高 | 中 | 早期(未披露合规计划) | 重大 — 可服务市场收缩 |
| 出口合规 / 转移责任 | BIS 2026 许可证审查、KYC、报告 | 中 | 中 | 早期 | 重大 — 罚款和报告负担 |
| 汽车功能安全门槛 | ISO 26262 / ASIL、SOTIF、ISO/PAS 8800 认证门槛 | 高 | 高 | 未证实(无公开认证) | 重大 — 阻断汽车设计导入 |
| 医疗 / 国防资质 | FDA;ITAR;采购安全 | 中 | 中 | 早期 | 中等 — 放慢垂直收入 |
| 数据隐私合规 | SiMa 隐私政策(2025 年 11 月);GDPR/CCPA | 低 | 低 | 已建立(已发布政策) | 轻微 — 行业标准风险 |
| 专利 / IP 诉讼 | 半导体 IP 侵权敞口 | 低 | 中 | 未知(无公开诉讼) | 中等 — 潜在风险,一旦发生成本高 |
监管敞口集中在市场准入和资质摩擦(出口管制、功能安全),而不是正在发生的执法; 截至本次报告日期,公开信息未发现针对 SiMa 的重大诉讼。
[CR007, CR008, CR009, CR010, CR011, CR012]7.3 运营、供应链与质量风险
SiMa.ai 的 Modalix MLSoC 采用 TSMC N6(6nm 级)工艺制造,把最关键依赖集中在台湾——全球领先制程、汽车级制造的枢纽。TSMC 若出现任何中断(自然灾害、台海地缘政治升级、产能分配转向大客户,或贸易限制),都会直接威胁 SiMa 造出产品的能力;在约 2027 年前,Samsung、Intel 或美国 / 欧洲晶圆厂很难为先进汽车级节点提供有意义的多元化。作为无晶圆厂公司,SiMa 还依赖 OSAT 伙伴做封装 / 测试,依赖 IP 供应商(Arm 核心、Synopsys ARC / 工具链)支撑架构,依赖分销商和合同制造商(Macnica、Enclustra、VVDN)向客户供货,因此任一层失效都会传导到出货。质量和可靠性风险更高,因为 SiMa 的目标垂直领域——汽车、工业、国防、医疗健康——要求严格可靠性;安全关键部署中一旦现场失效或召回,声誉后果会很重。一个具体验证缺口又放大了风险:没有公开的 Modalix 代际 MLPerf Inference 结果(SiMa 第一代 MLSoC 曾出现在 2024 年 MLPerf Inference 4.0 上,但当前产品线没有可比的独立基准),因此醒目的单位功耗性能优势主要依赖公司自报数字,而不是第三方审计结果。[CR014, CR015, CR016, CR017, CR018, CR019]
| 风险 | 敞口来源 | 发生可能性 | 影响 | 缓释成熟度 | 剩余敞口 |
|---|---|---|---|---|---|
| 单一晶圆厂依赖 | Modalix 基于 TSMC N6(台湾)制造 | 中 | 高 | 低(无公开第二来源) | 重大 — 若受扰动,生产停摆 |
| 台湾地缘政治扰动 | 台海紧张;产能集中在台湾 | 低 | 严重 | 低(行业共性,难对冲) | 重大 — 系统性、低概率 / 高影响 |
| OSAT / 封装依赖 | 外包组装与测试伙伴 | 中 | 中 | Unknown | 中等 — 传导至出货 |
| 安全关键场景现场故障 / 召回 | 汽车 / 工业 / 国防可靠性要求 | 低 | 高 | 早期(认证推进中) | 中等 — 声誉和责任风险 |
| Modalix 缺少独立基准 | 未公开 Modalix 代 MLPerf 结果 | 高 | 中 | 低(依赖公司数据) | 中等 — 性能主张未验证 |
| 渠道 / 合同制造依赖 | Macnica、Enclustra、VVDN 位于供应路径 | 中 | 中 | 中等(多个伙伴) | 中等 — 中间环节控制力损失 |
运营风险主要来自 TSMC 单一晶圆厂集中,以及缺少独立的当前代基准; 这两项都是结构性问题,SiMa 单靠自身只能部分缓释。
[CR014, CR015, CR016, CR017, CR018, CR019]有向图展示 SiMa.ai 根部风险如何级联为业务结果——供应、竞争和验证风险如何通过收入与融资传导,最终演变为关乎存续的结果。
[CR001, CR014, CR019, CR029, CR033]7.4 伙伴、客户与依赖风险
SiMa.ai 的商业化路径和技术栈编织在外部依赖中,每一层也都是风险向量。在竞争轴上,NVIDIA 同时是 SiMa 营销对标的基准(Modalix 模组被定位为与 NVIDIA 模组引脚和软件兼容),也是生态重力井——同样的兼容性降低采用难度,也让客户在 SiMa 失手时回到 NVIDIA。供给侧,SiMa 依赖 TSMC(晶圆厂)、Arm 和 Synopsys(IP 与汽车工具链),以及分销 / 解决方案伙伴渠道(L&T Technology Services、VVDN、Macnica、Enclustra、ThinkRobotics),这些渠道承担了相当一部分客户交付,增加中介利润,也降低直接控制。需求侧,由于 SiMa 不披露单客户收入,客户集中度无法衡量;但公开图景显示,公司依赖少数仍大多处于放量前的旗舰设计导入(STIGA、TRUMPF)。资本侧,SiMa 依赖一组风险投资和战略投资方(Maverick Capital、Fidelity、Dell、Micron Ventures、StepStone)及其持续出资意愿;2026 年 4 月 Micron 战略投资是正面信号,但也凸显对战略支持方的依赖。同行信号是警告:Hailo 报道估值跌破 USD500 million、走向困境 SPAC 路径并裁员,Blaize 发出持续经营警告,都说明这个行业里伙伴 / 资本信心可以多快蒸发。[CR021, CR022, CR023, CR024, CR025, CR026]
| 依赖对象 | 角色 | 不利事件发生可能性 | 影响 | 缓释成熟度 | 剩余敞口 |
|---|---|---|---|---|---|
| NVIDIA 生态 | 既是现有竞争者,也是兼容性锚点 | 高 | 高 | 早期(软件护城河建设中) | 重大 — 客户可回退到 NVIDIA |
| TSMC | 唯一披露的晶圆厂 | 中 | 高 | 低 | 重大 — 无公开替代来源 |
| Arm / Synopsys | CPU IP 与汽车工具链 | 低 | 中 | 已建立(已授权) | 中等 — 依赖授权和路线图 |
| 解决方案 / 分销伙伴 | LTTS、VVDN、Macnica、Enclustra、ThinkRobotics 等渠道伙伴 | 中 | 中 | 中等(多个伙伴) | 中等 — GTM 控制力和利润率 |
| 旗舰客户 | STIGA、TRUMPF(大多仍在量产前) | 中 | 高 | 早期 | 重大 — 集中度无法衡量 |
| 资本提供方 | Maverick、Fidelity、Dell、Micron、StepStone 等资本方 | 中 | 高 | 中等(近期 Series C + Micron) | 重大 — 依赖持续融资 |
依赖关系本身也是风险载体;NVIDIA 这条最尖锐,因为引脚 / 软件兼容性有助采用, 也降低了客户切回 NVIDIA 的门槛。
[CR021, CR022, CR023, CR024, CR025, CR026]有向图展示 SiMa.ai 关键外部依赖如何支撑其出货并服务客户,覆盖晶圆代工、IP、渠道、资本和需求。
[CR014, CR022, CR023, CR024, CR026]7.5 财务与商业模式风险
SiMa.ai 的财务风险被不透明放大。公司不披露收入、毛利率、月度现金消耗、现金余额或资金续航,因此承销方无法验证资本充足性、单位经济性或盈利路径,只能完全借用公开同行做代理。同行读数并不轻松:Blaize 报告 FY2025 收入约 USD38.6 million,经营亏损接近 USD103.8 million,并发出持续经营警告;Ambarella——最健康的无晶圆厂边缘 AI 可比公司——收入创纪录、接近 USD390.7 million,毛利率约 60%,但 GAAP 净亏损仍约 USD75.9 million。由此可读出,边缘 AI 无晶圆厂公司毛利率明显低于 NVIDIA 的 70%+,即便达到 Ambarella 规模的收入,也不足以实现 GAAP 盈利;这意味着 SiMa 面对的是漫长、吃资本、反复稀释的路径。资本强度是结构性的:每一代 MLSoC 都需要在 TSMC 先进节点重新流片,并在产生收入前持续投入软件。估值风险很尖锐:SiMa 有一个据报道但未经公司确认的 Series C 后估值,接近 USD1.4 billion,却没有披露收入锚定倍数,因此隐含收入倍数无法计算,独角兽标签更多建立在私人市场情绪上,而不是基本面。整合后的财务论点失效触发点,是资金续航 到期前无法以非惩罚性条款完成下一轮融资。[CR028, CR029, CR030, CR031, CR032, CR033]
7.6 缓释、监测指标与终止标准
每项重大风险都有可用缓释和可观察监测指标,但多数缓释仍处早期,并依赖 SiMa 尚未做出的披露。竞争风险的缓释,是加深 Palette 软件护城河并锁定多年设计赢单; 监测指标是设计赢单到量产的转化,以及任何已发布的独立基准;终止标准是旗舰设计导入输给 NVIDIA,或到下一轮融资时仍无法显示规模化收入。融资风险的缓释,是近期 Series C 和 Micron 背书;指标是资金续航和现金消耗倍数;终止标准是降估值融资,或资金续航到期前无法融资。供应链风险的缓释,是认证备选晶圆厂 / OSAT 产能并建立库存缓冲;指标是单一来源暴露和交期;终止标准是 TSMC 配额丢失或台海中断。监管风险的缓释,是出口管制合规计划,以及推进 ISO 26262 / ASIL 和 FDA 认证;指标是认证里程碑和任何 BIS 执法行动;终止标准是因缺少认证被目标垂直领域排除。执行风险的缓释,是在 CEO 之外扩展领导层梯队;指标是关键高管留任,以及对创始团队的独立佐证。核心尽调问题包括管理层资料室(财务、现金消耗、资金续航、在手订单)、独立 Modalix 基准、晶圆厂 / OSAT 合同、出口管制合规审查,以及功能安全认证证据。[CR035, CR036, CR037, CR038, CR039]
| 风险 | 依据 | 发生可能性 | 影响 | 缓释成熟度 | 剩余敞口 |
|---|---|---|---|---|---|
| 创始人 CEO 关键人依赖 | Krishna Rangasayee 是愿景和融资的核心 | 中 | 高 | 早期(未披露梯队建设) | 重大 — 离开会冲击信心 |
| 更广泛创始团队未验证 | 联合创始人未被公开信息独立佐证 | 中 | 中 | Unknown | 中等 — 治理 / 尽调缺口 |
| 从设计赢单爬坡到量产 | 硬件创业公司历史上的“量产惯性” | 高 | 高 | 早期 | 重大 — 核心执行挑战 |
| 人才竞争 | 边缘 AI 芯片和 ML 编译器人才稀缺 | 中 | 中 | 中等 | 中等 — 留才和招聘风险 |
| 治理 / 披露纪律 | 无公开财务数据;私营公司不透明 | 高 | 中 | 早期 | 中等 — 限制外部验证 |
执行风险集中在把设计赢单转成量产,以及叙事和融资都集中于创始人 CEO; 两者都是硬件创业公司的常见失败模式。
[CR035, CR036, CR037, CR038]08估值
8.1 投资论点与反论点
SiMa.ai 站在两条长期顺风的交汇处:AI 推理从云端迁移到边缘,以及机器人、汽车、工业自动化、航空航天与国防、智能视觉和医疗健康中的「物理 AI」兴起。多头论点有四根支柱。第一,节能边缘推理的可服务市场大且在增长,而 NVIDIA 的数据中心经济性无法干净套进嵌入式功耗和热预算。第二,Modalix MLSoC(基于 TSMC N6)与 Palette 软件套件形成差异化产品;SiMa 将其定位为单芯片、低功耗平台,覆盖传统计算机视觉、端侧生成式和推理模型。第三,真实的商业和战略验证:标志性合作(STIGA 机器人割草机、TRUMPF AI 激光器、LTTS、VVDN、Synopsys 汽车)、Modalix SOM 获得 2026 Edge AI and Vision Alliance Product of the Year 奖,以及 2026 年 4 月 Micron 战略投资叠加超额认购的 Series C。第四,由 CEO Krishna Rangasayee(前 Xilinx SVP、前 Groq COO)带领、具备半导体原生能力的可信创始团队,并获得顶级投资方支持(Maverick Capital、Fidelity、Point72、Dell Technologies Capital、StepStone)。反论点同样具体。SiMa 不披露收入、毛利率、现金消耗或资金续航,因此报道中约 USD1.4 billion 的估值没有公开财务锚;第三方收入估算约 USD35.5 million,意味着相对公开纯可比公司 Ambarella 约 5x–8x 市销率,SiMa 前瞻倍数约 28x–39x。NVIDIA 的 Jetson 和 CUDA / JetPack 生态根深蒂固;Hailo 近乎崩塌(估值据报低于 USD500 million)和 Blaize 的持续经营警告,展示了边缘 AI 硅片经济性会多快破裂。没有公开的 Modalix 代际 MLPerf 结果独立验证单位功耗性能主张,汽车垂直又被 SiMa 尚未公开证明的 ISO 26262 认证卡住。因此,当前估值定价的是尚未独立可见的执行。[CV001, CV002, CV003, CV004, CV005, CV006]
| 支柱 | 投资论点 | 反论点 | 证据质量 |
|---|---|---|---|
| 市场 | 边缘 AI / 物理 AI TAM 庞大且增长;云端 GPU 经济性不适合边缘侧功耗和散热预算 | 市场碎片化且由 NVIDIA 主导;SiMa 可拿到的份额尚未证明 | 中 |
| 产品与技术 | Modalix MLSoC 采用 TSMC N6,叠加 Palette 软件;单芯片、低功耗、从 CV 到 GenAI 的平台 | 没有公开的 Modalix 代 MLPerf 结果;每瓦性能仍靠公司口径支撑 | 中 |
| 商业牵引 | 标杆合作(STIGA、TRUMPF、LTTS、VVDN、Synopsys);2026 年度产品;Micron 战略投资 | 未披露收入;合作还处早期设计导入,并非已验证放量 | 低 |
| 资本与支持方 | 已融资约 USD355M;一线投资人;Series C 超额认购 | 无晶圆厂模式资本消耗高;Hailo、Blaize 等同业困境显示脆弱性 | 中 |
| 估值 | 独角兽背书与战略方兴趣 | 隐含约 28-39x,对比上市可比约 5-8x;USD1.4B 未确认且报道不一致 | 低 |
| 退出 | 战略并购需求强(NXP/Kinara、onsemi/Synaptics) | 没有公开进程;短期 IPO 可能性低;降轮风险 | 低 |
证据质量只反映公开来源的交叉验证;经审计财务和独立基准测试会提高「商业牵引」和「产品」两项的置信度。
[CV003, CV004, CV005, CV007, CV008, CV030]决策逻辑从三条正向投资支柱(市场、产品、资本)出发,穿过三道估值闸门(未披露收入、NVIDIA 主导叠加同业承压、USD1.4B 未确认且对比上市公司 5-8x),落到 research-more 建议。三道闸门全部通过后,才能上调为 buy。
[CV001, CV007, CV010, CV030]8.2 建议、信心与评级
建议是在报道的约 USD1.4 billion 入场价下继续研究,且带条件;对定性论点的信心为中等,对价格的信心为低。SiMa.ai 的战略故事在结构上有吸引力,但没有任何披露的收入、利润率、现金消耗 或资金续航,用公开证据无法支撑出手决定:入场倍数无法计算,可用锚点只有第三方收入估算(约 USD35.5 million)和远低于隐含私人倍数的公开同行倍数。估值立场偏贵,且很大程度缺少已披露基本面支撑:按估算 USD35.5 million 收入,报道估值隐含约 28x–39x 前瞻收入;而最接近的公开纯可比公司 Ambarella 交易在约 5x–8x 市销率,困境同行(Hailo、Blaize)则展示了增长或经济性不及预期时的下行。风险评级为高,主因包括 NVIDIA 带来的竞争替代风险、资本密集型无晶圆厂 模式中的融资和现金消耗风险、对 TSMC 的单一晶圆厂依赖,以及监管 / 认证摩擦(BIS 出口管制、ISO 26262)。最可能退出路径是战略 M&A,而不是近期 IPO;边缘 AI 收购市场活跃(NXP / Kinara、onsemi / Synaptics),Micron 也有战略持股。不过由于没有公开进程,退出路径信心较低。三情景模型给出的概率加权价值约 USD965 million,低于报道的 USD1.4 billion;因此在报道价格下,除非多头情景兑现,否则预期回报为负。只有满足最后一节的尽调条件,建议才会上调为买入。[CV010, CV011, CV012, CV013, CV014, CV015]
| 维度 | 评估 | 置信度 | 关键证据 |
|---|---|---|---|
| 总体建议 | 继续研究(有条件) | 中 | 未披露收入,估值缺少锚点;有吸引力的投资论点还需数据室验证 |
| 估值立场 | 偏贵 / 基本面不支撑 | 中 | 报道的 USD1.4B 估值,在约 USD35.5M 估算收入上对应约 28-39x;Ambarella 约 5-8x P/S |
| 风险评级 | 高 | 中 | NVIDIA 挤压、无晶圆厂烧钱、TSMC 单一晶圆厂、ISO 26262 与 BIS 摩擦 |
| 退出路径 | 战略并购最可能 | 低 | 边缘 AI 并购活跃(NXP/Kinara USD307M;onsemi/Synaptics 约 USD7B);Micron 持有战略股权 |
| 概率加权价值 | 约 USD965M,对比 USD1.4B 入场估值 | 低 | 牛 / 基准 / 熊 25/45/30 加权后,价值低于报道的入场价 |
置信度只反映公开证据;经审计财务与 Series C 优先条款可能改变总体建议。
[CV010, CV011, CV012, CV023]九项头部指标框定入场决策。报道估值和隐含倍数显著高于上市可比倍数,而收入、烧钱和现金续航均未披露——这正是建议为 research-more 而非 buy 的核心原因。
KPI 数值综合报道数据、第三方估算和上市同业数据;未披露项需要数据室访问权限。
[CV001, CV002, CV017, CV023, CV029, CV042]8.3 融资、估值背景与入场纪律
SiMa.ai 自 2018 年以来约九轮合计融资约 USD355 million,最高到 2025 年 8 月由 Maverick Capital 领投的 USD85 million Series C,StepStone Group 加入,Micron Ventures 参与;该轮被描述为超额认购。公司没有确认融资后估值数字;约 USD1.4 billion 的估值来自归因于 The Information 的报道,并被二级市场追踪平台重复引用,而其他私募市场追踪平台给出显著不同的锚(例如 2025 年中时间戳的 USD960 million 估值,以及约 2.7x 的估值 / 融资资本效率比),说明该数字既非官方,也没有被一致报道。Micron 2026 年 4 月战略投资(金额未披露)是定性验证,但不能形成可验证标记。披露缺口约束了入场纪律。由于 SiMa 是私营且未审计,投资人看不到约 USD355 million 优先股资本所附的清算优先权堆栈、参与权或优先级;在下行情景退出中,优先权悬置可能在普通股和后期投资人回本前吸收大部分甚至全部收益,在温和收入基数上叠加庞大累计优先权,是典型的后期风险。稀释很可能继续:一家不披露盈利能力、资本密集的 无晶圆厂半导体公司仍需后续融资,每一轮都会稀释现有持有人,并可能把新优先权堆栈置于早期资金之上。因此,有纪律的入场姿态应把报道的 USD1.4 billion 视为未经验证的上限,要求 Series C 条款清单和股权结构表,并在获得经审计财务和优先权条款前,按概率加权价值(约 USD965 million)而非报道头条数字定价。[CV001, CV017, CV018, CV019, CV020, CV021]
8.4 情景分析——多头、基准与空头
由于 SiMa 不披露收入,下方情景基于第三方收入估算(约 USD35.5 million)和边缘 AI 行业基准构建,前瞻退出估值则从公开同行和 M&A 倍数推断;任何情景都无法用 SiMa 财务独立验证。多头情景(赋予约 25%):Modalix 设计导入在机器人、工业和汽车中转化为量产,收入到 2028 年逼近约 USD150 million,物理 AI 定位吸引战略收购方,退出以约 10x–17x 的溢价倍数打印,产生约 USD1.5 billion–USD2.5 billion。基准情景(约 45%):设计导入稳定但转化较慢,收入增长到约 USD80 million,市场给予高个位数到低双位数的混合边缘 AI 倍数(以 Ambarella 的 5x–8x 为锚,叠加适度增长溢价),产生约 USD0.6 billion–USD1.0 billion,等于或低于报道入场价。空头情景(约 30%):NVIDIA 生态把 SiMa 限制在小众市场,汽车认证滑坡,关键客户或晶圆厂中断咬伤业务,收入停滞在约 USD40 million 以下,SiMa 被迫进入稀释性降估值融资或类似 Hailo 的困境流程,产生约 USD0.2 billion–USD0.5 billion。概率加权价值约 USD965 million,低于报道的 USD1.4 billion 入场价。敏感性分析显示,只有高收入和持续溢价倍数同时成立,才能支撑报道价格;按公开同行倍数,报道估值在合理收入区间内都缺少支撑。会把 SiMa 推向空头情景的下行触发点,列在论点失效表中。[CV023, CV024, CV025, CV026, CV027, CV028]
| 情景 | 概率 | 2028E 收入 | 退出倍数 | 退出估值区间 | 关键假设 |
|---|---|---|---|---|---|
| 牛 | ~25% | ~USD150M | ~10-17x | USD1.5B-USD2.5B | 量产设计导入转化;战略收购方为物理 AI 稀缺性支付溢价 |
| 基准 | ~45% | ~USD80M | ~8-12x | USD0.6B-USD1.0B | 转化稳定但放慢;混合边缘 AI 倍数接近 Ambarella,并叠加增长溢价 |
| 熊 | ~30% | <USD40M | n/m(困境) | USD0.2B-USD0.5B | NVIDIA 把 SiMa 挤到小众市场;出现类似 Hailo 的降轮或困境流程 |
| 概率加权价值 | 100% | 混合 | 混合 | ~USD0.97B | 加权价值低于报道的 USD1.4B 入场价 |
收入估算来自第三方跟踪器和行业基准;概率权重是定性判断,有待经审计财务和按金额加权的管线验证。
[CV023, CV024, CV025, CV026, CV027, CV028]三种收入情景(USD40M、USD80M、USD150M)与三档倍数(上市可比 8x、增长溢价 15x、战略 25x)下的退出估值。只有高收入、高倍数组合才撑得起报道的 USD1.4B 入场估值;按上市可比倍数,整个合理收入区间都无法支撑这一报道估值。
收入数值是分析师情景,锚定第三方估算(约 USD35.5M)和行业基准;SiMa 不披露收入。倍数档位锚定 Ambarella(约 5-8x P/S)、增长溢价和边缘 AI 并购先例。所有数值单位为百万美元。
[CV002, CV023, CV024, CV025, CV029, CV034]各情景退出估值区间,加上概率加权价值和报道入场价。牛市情景区间高于入场价;基准情景持平或更低;熊市情景意味着显著亏损。概率加权价值(约 USD965M)低于报道的 USD1.4B 入场价,说明按报道价格进入的预期回报为负。
情景区间为分析师估算;收入与倍数假设来自可比公司和行业并购推断。概率权重(25/45/30)是在取得经审计财务前的定性判断。数值单位为百万美元。
[CV023, CV024, CV025, CV026, CV028]8.5 可比估值组
最有信息量的公开可比公司是 Ambarella(NASDAQ AMBA),这是一家无晶圆厂边缘 AI 视觉处理器公司,报告 FY2026 收入 USD390.7 million(增长 37.2%)、毛利率约 60%、GAAP 净亏损 USD75.9 million、非 GAAP 盈利、现金 USD312.6 million;在约 USD2 billion–USD3 billion 市值区间,它的市销率约 5x–8x,比 SiMa 隐含的约 28x–39x 低一个数量级。Blaize(NASDAQ BZAI)是警示性公开可比:FY2025 收入接近 USD38.6 million,经营亏损接近 USD103.8 million,并发出持续经营警告,说明市场会如何否定亏损扩张快于收入的边缘 AI 公司。私营同行中,Hailo 融资超过 USD340 million,但据报估值跌到 USD500 million 以下,并走向困境 SPAC 路径、伴随裁员;EdgeCortix 完成超额认购的 Series B,总融资超过 USD110 million。M&A 可比交易框定了退出:NXP 以全现金 USD307 million 收购边缘 AI NPU 初创 Kinara(2025 年 2 月),onsemi 同意以约 USD7 billion 全股票交易收购 Synaptics,以扩展边缘 AI,证明战略买方意愿强,并愿为差异化边缘 AI IP 支付溢价。行业数据表明,具吸引力增长标的的 AI M&A 收入倍数大致聚集在 25x–30x 区间,部分异常值更高;不过无晶圆厂 硬件通常低于软件定价。总体看,可比组在已披露基本面上支持远低于报道 USD1.4 billion 的估值,同时承认战略收购方可能为物理 AI 硅片支付稀缺性溢价。[CV002, CV029, CV030, CV031, CV032, CV033]
| 公司 | 类型 | 估值 / 对价 | 收入 | 隐含倍数 | 备注 |
|---|---|---|---|---|---|
| SiMa.ai | 私营(Series C) | 约 USD1.4B(未确认) | 约 USD35.5M(估算;未披露) | ~28-39x | 第三方跟踪器转引 The Information 报道;公司未确认 |
| Ambarella (AMBA) | 上市 | 约 USD2-3B 市值 | USD390.7M(FY2026) | 约 5-8x P/S | 最接近的上市纯边缘 AI 可比公司;毛利率约 60%,non-GAAP 盈利 |
| Blaize (BZAI) | 上市 | 困境 | 约 USD38.6M(FY2025) | n/m | 经营亏损约 USD103.8M;持续经营警告 |
| Hailo | 私营 | <USD500M(下降约 50%) | 未披露 | n/m | 已融资 >USD340M;走困境 SPAC 路径并裁员 |
| EdgeCortix | 私营 | 未披露 | 未披露 | n/m | Series B 超额认购;总融资超过 USD110M |
| Kinara(被 NXP 收购) | 并购 | USD307M(全现金,2025 年 2 月) | 未披露 | n/m | 边缘 AI NPU 收购;战略买家先例 |
| Synaptics(被 onsemi 收购) | 并购 | 约 USD7B(全股票,2026) | 未披露 | n/m | 边缘 AI 扩张交易;战略溢价高 |
标记 n/m 的倍数,原因是收入未披露或公司处于困境;从已披露基本面看,可比组支撑低于报道 USD1.4B 的估值。
[CV002, CV029, CV030, CV031, CV032, CV033]8.6 退出准备度、论点失效触发点与最终尽调要求
退出准备度仍早。SiMa 是一家私营、披露前公司,没有公开收入、没有经审计财务,也没有宣布 IPO 流程;现实退出是战略 M&A,买方群体(NXP、onsemi、Qualcomm、Renesas 等大型模拟 / 嵌入式厂商,以及作为战略伙伴的 Micron)活跃,并愿为边缘 AI IP 支付溢价,Kinara 和 Synaptics 交易已经说明这一点。鉴于缺少披露规模,且 Blaize 在公开市场反响不佳,近期 IPO 不太可能。主要论点失效触发点包括:降估值融资或平价融资否定报道估值(Hailo 先例);设计丢单输给 NVIDIA Jetson,使 SiMa 被限制在小众市场;收入持续低于约 USD40 million,使任何溢价倍数都站不住;到约 2027 年仍未取得 ISO 26262 汽车认证,关闭既定增长垂直;TSMC N6 配额丢失或台湾中断导致停产;以及现金耗尽、被迫困境融资。每一项都是离散、可监测的终止标准。最终尽调要求按优先级排序为:(1)经审计收入、毛利率、经营 现金消耗和资金续航;(2)Series C 条款清单,包括清算优先权、参与权和优先级,加完整股权结构表和稀释历史;(3)按金额加权的设计赢单管线、 量产时间表和客户集中度;(4)晶圆厂和 OSAT 第二来源认证路线图;(5)独立 Modalix 代际基准。满足要求(1)和(2)是门槛条件;没有它们,报道估值无法验证,建议仍是继续研究而不是买入。[CV012, CV030, CV036, CV037, CV038, CV039]
| 触发因素 | 可观察信号 | 投资含义 |
|---|---|---|
| 降轮或平轮融资 | Series D 定价等于或低于报道的 USD1.4B | 证实估值过高;存在 Hailo 式重定价风险 |
| 被 NVIDIA 生态锁出 | 目标槽位反复输给 Jetson/CUDA | 被困小众市场;基准 / 熊情景 |
| 收入持续不及预期 | 收入低于约 USD40M 且增速放缓 | 任何溢价倍数都站不住 |
| 汽车认证失败 | 到约 2027 年仍无 ISO 26262 / ASIL 认证 | 汽车垂直行业关闭 |
| 晶圆厂扰动 | TSMC N6 产能配额丢失,或台湾供应冲击 | 生产停摆;收入与交付风险 |
| 现金耗尽 | 无确定融资时,现金跑道低于约 12 个月 | 被迫困境融资;严重稀释 |
每个触发因素都是离散且可监控的否决条件;任意两个发生,都会显著提高熊情景的概率权重。
[CV007, CV023, CV036, CV037, CV038]| 索取项 | 理由 | 优先级 |
|---|---|---|
| 经审计收入、毛利率、经营性烧钱、现金跑道 | 锚定入场倍数和判断融资风险时点的唯一办法 | 关键(门槛) |
| Series C 条款清单——清算优先权、参与权、优先顺位 | 量化下行退出时压在普通股之上的优先权包袱 | 关键(门槛) |
| 完整股权结构表与稀释历史 | 揭示持股、期权池和未来稀释路径 | 高 |
| 按金额加权的设计赢单管线及量产时间表 | 检验收入韧性,以及早期合作转化 | 高 |
| 客户集中度(前五大收入占比) | 暴露单一客户依赖风险 | 高 |
| 晶圆厂和 OSAT 第二来源认证路线图 | 评估 TSMC 单一晶圆厂供给韧性 | 中 |
| 独立 Modalix 代基准测试(MLPerf 级别) | 由第三方验证每瓦性能主张 | 中 |
两项门槛问题得到满足,才可从继续研究上调为买入;其余问题用于细化价格和风险评级。
[CV012, CV018, CV039, CV040, CV041, CV042]8.7 图表
免责声明
本报告仅用于尽调和信息参考,基于截至 2026-07-23 的公开资料,不构成投资、法律、会计或税务建议。SiMa.ai 是未上市公司,不披露财务数据;报道估值未经公司确认,第三方口径也不一致。读者在作出投资决策前,应独立核验所有事实,并取得一手尽调材料。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | SiMa.ai was founded in 2018 and is headquartered in San Jose, California. | 高 | SO003, SO017, SO023 |
| CO002 | SiMa.ai is a fabless edge-AI semiconductor company that markets its category as "Physical AI." | 高 | SO001, SO003 |
| CO003 | SiMa.ai sells a full-stack platform branded "SiMa.ai ONE" combining purpose-built silicon with a software-centric approach. | 高 | SO002, SO001 |
| CO004 | The hardware layer is the MLSoC, whose second generation is branded Modalix. | 高 | SO008, SO010 |
| CO005 | The software layer, Palette, includes an SDK and a no-code/low-code Edgematic visual development tool. | 高 | SO002, SO006 |
| CO006 | SiMa.ai's software roadmap added the LLiMa on-device LLM framework in 2025 and the agentic Palette Neat environment in 2026. | 中 | SO008, SO003 |
| CO007 | SiMa.ai raised an $85 million oversubscribed Series C on August 1, 2025, bringing total disclosed funding to $355 million. | 高 | SO002, SO007, SO009 |
| CO008 | The Modalix MLSoC entered production and began shipping in August 2025, supporting LLMs, transformers, CNNs, and GenAI under 10 watts. | 高 | SO008, SO010, SO012, SO013 |
| CO009 | Modalix is built on TSMC's N6 process with an Arm-based architecture. | 高 | SO008, SO010, SO014 |
| CO010 | Modalix SoM commercial 1K-unit pricing starts at $349 for an 8GB module and $599 for a 32GB module, with a developer kit at $1,499. | 中 | SO008 |
| CO011 | Krishna Rangasayee is the founder and CEO of SiMa.ai. | 高 | SO004, SO015, SO017 |
| CO012 | Rangasayee spent about 18 years at Xilinx (rising to SVP/GM and EVP Global Sales), was COO of Groq, held roles at Altera and Cypress, and holds 25+ patents. | 高 | SO004, SO015 |
| CO013 | SiMa.ai's board is chaired by Moshe Gavrielov (retired Xilinx CEO; TSMC, Cadence, NXP boards) and includes Intel CEO Lip-Bu Tan. | 中 | SO003 |
| CO014 | The board also includes Scott Darling (Dell Technologies Capital), Andrew Homan (Maverick Capital), Jake Flomenberg (Wing Ventures), and Mike Dauber (Amplify Partners). | 中 | SO003 |
| CO015 | Harry Kroeger serves as President of Automotive, bringing Bosch-Daimler management-board experience and prior board roles at Tesla and Rivian. | 中 | SO003 |
| CO016 | Public records consistently name only Rangasayee as founder, leaving the additional co-founders (Manish Garg, Yann Lepenant, Manu Prasad) uncorroborated in retained sources. | 中 | SO017, SO003 |
| CO017 | SiMa.ai carries very high key-person dependence concentrated on its founder-CEO. | 中 | SO004, SO015 |
| CO018 | SiMa.ai raised capital across roughly nine disclosed rounds from a $30M Series A (2020, Dell TC) through the $85M Series C (2025, Maverick). | 高 | SO019, SO018, SO002 |
| CO019 | Fidelity led an $80M Series B (2021) and a $30M extension (2022); MSD Partners led a $37M Series B1 extension (2022). | 中 | SO019, SO028 |
| CO020 | StepStone Group joined as a new investor in the August 2025 Series C. | 高 | SO002, SO007 |
| CO021 | Dell Technologies Capital and Fidelity are repeat investors continuing across multiple SiMa.ai rounds. | 中 | SO019, SO028 |
| CO022 | Micron Technology made a strategic investment in SiMa.ai of undisclosed size dated to around April 2026. | 中 | SO016, SO019, SO026 |
| CO023 | SiMa.ai declined to disclose its post-Series C valuation. | 中 | SO009 |
| CO024 | The Information reported that SiMa.ai was raising more than $100 million at a valuation of around $1.4 billion. | 中 | SO027 |
| CO025 | The reported ">$100M" round size conflicts with the company's official "$85M" Series C headline. | 中 | SO027, SO002 |
| CO026 | The reported ~$1.4B valuation represents a premium of more than 45% over a roughly $960M valuation from August 2024 per PitchBook. | 中 | SO027 |
| CO027 | SiMa.ai operates from its San Jose headquarters with offices claimed across eight countries. | 中 | SO003 |
| CO028 | SiMa.ai's named office locations include the US (San Jose), Germany (Stuttgart), India (Bengaluru), Israel (Modiin), Japan (Tokyo), and Korea (Seoul). | 中 | SO003 |
| CO029 | SiMa.ai targets robotics, automotive, industrial automation, aerospace & defense, smart vision, and healthcare. | 高 | SO002, SO005 |
| CO030 | SiMa.ai announced a strategic alliance with L&T Technology Services in September 2025 to co-develop Physical AI solutions. | 中 | SO021, SO029 |
| CO031 | SiMa.ai has a strategic collaboration with Synopsys (December 2024) for automotive edge-AI development. | 高 | SO020, SO008 |
| CO032 | SiMa.ai has not publicly disclosed a 2026 revenue, run-rate, gross margin, or unit-shipment figure. | 中 | SO009, SO019 |
| CO033 | No retained public source pins a current SiMa.ai headcount as of the run date. | 中 | SO003, SO019 |
| CO034 | SiMa.ai says it shipped the industry's first purpose-built MLSoC with a full software stack around 2022. | 中 | SO003 |
| CO035 | SiMa.ai's first-generation MLSoC moved into production deployments across robotics, industrial automation, and smart vision in 2023. | 中 | SO003 |
| CO036 | SiMa.ai competes in a market where NVIDIA holds a dominant share, a structural headwind for edge-AI challengers. | 中 | SO030 |
| CO037 | Even well-funded edge-AI rivals such as Hailo are reported to struggle to convert pilots into scale against NVIDIA's ecosystem. | 中 | SO030 |
| CO038 | Widely cited ~$1.4B unicorn status rests on unnamed sources rather than a company statement, an adverse disclosure signal. | 中 | SO027, SO009 |
| CO039 | The incomplete public co-founder record is a disclosure gap affecting the leadership picture. | 中 | SO017 |
| CO040 | Cap-table economics, ownership percentages, and control rights across SiMa.ai's rounds are not publicly disclosed. | 中 | SO019 |
| CO041 | SiMa.ai's silicon relies on ecosystem partners including TSMC, Arm, Synopsys, and Enclustra. | 高 | SO008, SO020 |
| CM001 | SiMa.ai's relevant market is the edge-AI inference silicon layer, not cloud training silicon or end-user AI applications. | 高 | SM016, SM019 |
| CM002 | SiMa.ai's included spend covers purpose-built MLSoCs, system-on-modules, developer kits, and the Palette software toolchain. | 高 | SM016, SM019 |
| CM003 | Cloud/training GPUs, general-purpose CPUs, discrete sensors, and end-user AI SaaS seats sit outside SiMa.ai's directly served market. | 中 | SM016, SM001 |
| CM004 | The primary status-quo substitute for SiMa.ai is NVIDIA's Jetson platform, alongside Hailo, Ambarella, Blaize, Qualcomm SoCs, FPGAs, and continued cloud inference. | 高 | SM014, SM020 |
| CM005 | SiMa.ai differentiates on performance-per-watt and a single-chip, software-first experience versus incumbent and general-purpose alternatives. | 中 | SM016, SM023 |
| CM006 | SiMa.ai's market spans robotics, industrial automation, automotive/ADAS, aerospace and defense, smart vision, and healthcare. | 高 | SM017, SM018 |
| CM007 | Buyers evaluate SiMa.ai because cloud round-trips are too slow, power-hungry, or exposed for real-time on-device inference. | 中 | SM013, SM023 |
| CM008 | The edge-AI hardware market, the closest SAM proxy for SiMa.ai's chips, is sized at USD26.14 billion in 2025 rising to USD58.90 billion by 2030 at a 17.6% CAGR. | 高 | SM001, SM015 |
| CM009 | Fortune Business Insights sizes the broad edge-AI market at USD35.60 billion in 2025 growing to USD445.75 billion by 2034 at a 32.5% CAGR. | 中 | SM002 |
| CM010 | Global Market Insights sizes the edge-AI market at USD25.2 billion in 2025 rising to USD225.5 billion by 2035 at a 24.7% CAGR. | 中 | SM003 |
| CM011 | Grand View Research (via Axis Intelligence) sizes the edge-AI market at USD24.9 billion in 2025 and USD30.0 billion in 2026, reaching USD118.7 billion by 2033 at a 21.7% CAGR. | 中 | SM004 |
| CM012 | Published edge-AI market estimates cluster on a USD24.9-35.6 billion 2025 base but diverge widely on end-year forecasts due to differing scope and windows. | 高 | SM001, SM002, SM003, SM004 |
| CM013 | MarketsandMarkets sizes a narrow physical-AI compute/software layer reaching USD15.24 billion by 2032 at a 47.2% CAGR from 2026. | 中 | SM005, SM006 |
| CM014 | Broader physical-AI ecosystem estimates reach roughly USD383 billion in 2026 and into the trillions by 2040, far too broad for a direct SiMa.ai SAM. | 低 | SM006, SM007 |
| CM015 | The AI-robotics market is projected to grow from USD6.11 billion in 2025 to USD33.39 billion by 2030. | 低 | SM006 |
| CM016 | MarketsandMarkets sizes automotive AI at USD18.83 billion in 2025 rising to USD38.45 billion by 2030 at a 15.3% CAGR. | 中 | SM008 |
| CM017 | The automotive AI SoC segment is projected to reach roughly USD13.0 billion by 2034 at a 15.6% CAGR. | 中 | SM009 |
| CM018 | The ADAS market is projected to grow from USD20.73 billion in 2021 to USD74.57 billion by 2030 at a 14.2% CAGR. | 中 | SM010 |
| CM019 | No retained public source discloses SiMa.ai's revenue, unit shipments, design-win count, or served-market share, so the sizing lenses bound the opportunity rather than the company's SAM/SOM. | 高 | SM016, SM020 |
| CM020 | SiMa.ai organizes go-to-market around verticals, with users being embedded/ML engineering teams and budget owners being hardware product P&L or program budgets. | 中 | SM017, SM016 |
| CM021 | Adoption triggers differ by vertical - thermal/power limits in robotics, functional safety and latency in automotive, ruggedization and supply assurance in defense, and privacy/regulation in healthcare. | 中 | SM017, SM022, SM013 |
| CM022 | The adoption path runs from evaluation on a developer kit and Palette software, through design-in on a system-on-module, to volume production. | 中 | SM019, SM016 |
| CM023 | SiMa.ai offers system-on-modules at published 1K-unit pricing of USD349 (8GB) and USD599 (32GB), with a USD1,499 developer kit. | 中 | SM019 |
| CM024 | Hardware design-ins create long sales cycles and high post-design-in switching costs, which reward socket wins but also favor the known incumbent. | 中 | SM014, SM020 |
| CM025 | Ecosystem partners such as Arm, TSMC, Synopsys, L&T Technology Services, and Enclustra shorten SiMa.ai's evaluation and design-in cycles. | 中 | SM019, SM016 |
| CM026 | The edge-AI value chain runs from IP and foundry (Arm, TSMC) through SiMa.ai's MLSoC and modules to vertical OEMs and end deployments. | 中 | SM019, SM016 |
| CM027 | Reference customers and ecosystem partners materially shorten evaluation because they reduce perceived integration risk. | 中 | SM025, SM016 |
| CM028 | Low-latency local inference is the dominant demand driver for edge AI because cloud round-trips are too slow for robotics and safety-critical control. | 中 | SM013, SM003 |
| CM029 | Data privacy and sovereignty push sensitive video, industrial, and healthcare data to stay on-device to meet regulatory requirements. | 中 | SM013, SM012 |
| CM030 | Power efficiency (performance-per-watt) is a core edge-AI driver enabling AI on battery- and thermally-constrained products. | 中 | SM001, SM004 |
| CM031 | Global 5G connections rose to 1.76 billion in 2023 and are forecast to reach 7.9 billion by 2028, expanding the base of devices that can host edge intelligence. | 中 | SM012 |
| CM032 | On-device generative and agentic AI is raising the per-device compute intensity edge silicon must handle. | 中 | SM005, SM013 |
| CM033 | Edge-AI adoption is constrained by security/attack surface, ecosystem fragmentation, skills gaps, regulatory requirements, and high hardware cost with long design-in cycles. | 中 | SM013, SM003 |
| CM034 | NVIDIA holds a dominant share of edge robotics compute, making its Jetson platform the default option engineering teams evaluate first. | 高 | SM020, SM014 |
| CM035 | Most edge-AI chip challengers have struggled to scale against the incumbent, so market growth is easier to prove than durable share capture. | 中 | SM020 |
| CM036 | The large edge-AI TAM does not neutralize NVIDIA's incumbency, so SiMa.ai's reachable revenue is gated by design-win capture rather than market size. | 中 | SM020, SM014 |
| CM037 | Ambarella, a public pure-play edge-AI chipmaker, reported USD390.7 million in fiscal 2026 revenue with 80% attributable to edge AI, evidencing a real but modest independent-vendor revenue pool. | 中 | SM004 |
| CM038 | Published edge-AI forecasts differ materially in scope (hardware vs software/services), geography, and horizon, so no single forecast should be treated as canonical. | 高 | SM001, SM002, SM003 |
| CM039 | The physical-AI ecosystem estimate is usable only as context, not as a SiMa.ai SAM, because it aggregates all robots, autonomous machines, and infrastructure. | 中 | SM006, SM007 |
| CM040 | Retained sources do not provide a precise current share split among Jetson, Hailo, Ambarella, Qualcomm, and SiMa.ai for the specific physical-AI sockets SiMa.ai targets. | 中 | SM014, SM020 |
| CP001 | SiMa.ai's competitive set spans incumbents, direct venture-backed peers, a public pure-play, substitutes, and likely future entrants. | 高 | SP004, SP007, SP002 |
| CP002 | NVIDIA's Jetson platform is the dominant edge-robotics compute incumbent, reinforced by the CUDA-X and JetPack software ecosystem. | 高 | SP014, SP002 |
| CP003 | NVIDIA Jetson AGX Orin delivers up to 275 TOPS, and NVIDIA has introduced the newer Jetson Thor generation. | 中 | SP014, SP007 |
| CP004 | Qualcomm is a second incumbent bringing large mobile/IoT scale and Robotics RB-series platforms at roughly 15 TOPS and 5-15 watts. | 中 | SP007, SP004 |
| CP005 | SiMa.ai's direct peers are venture-backed edge-AI silicon startups Hailo, Blaize, EdgeCortix, and Axelera AI. | 中 | SP004, SP005 |
| CP006 | Ambarella is a public pure-play that pivoted its camera-SoC franchise into edge AI and is scaling real revenue. | 中 | SP012, SP013 |
| CP007 | Substitutes and status-quo alternatives include general-purpose SoCs, embedded GPUs, FPGAs, and continued cloud inference. | 中 | SP007, SP004 |
| CP008 | Likely future entrants include hyperscaler edge-inference silicon, OEM custom chips, and automotive-specific players such as Mobileye. | 低 | SP002, SP007 |
| CP009 | NVIDIA funds Jetson from a trillion-dollar franchise and owns the dominant edge developer ecosystem, giving it effectively unconstrained resources. | 高 | SP002, SP014 |
| CP010 | Ambarella reported record fiscal-2026 revenue of USD390.7 million with roughly 80% from edge AI and over 42 million edge-AI SoCs shipped. | 中 | SP012, SP013 |
| CP011 | Ambarella has a multi-year agreement with Hanwha carrying an USD800 million revenue opportunity across security, robotics, and industrial automation. | 中 | SP012 |
| CP012 | Blaize went public via a SPAC merger with BurTech in January 2025 and trades on Nasdaq as BZAI. | 高 | SP010, SP017 |
| CP013 | Blaize reported 2025 revenue of USD38.6 million, up from USD1.6 million in 2024, its first full year of commercial revenue. | 高 | SP010, SP011 |
| CP014 | Hailo's valuation fell from a USD1.2 billion peak to under USD500 million by early 2026, prompting layoffs and a distressed SPAC merger. | 高 | SP005, SP016 |
| CP015 | Hailo became a unicorn in 2021 and raised a USD120 million Series C extension in April 2024 to reach a USD1.2 billion valuation, with total funding across rounds of roughly USD340 million or more. | 中 | SP005, SP008, SP015 |
| CP016 | EdgeCortix has raised over USD110 million (Series B plus grants and debt) and is scaling its SAKURA-II accelerator with a SAKURA-X chiplet roadmap. | 高 | SP018, SP019 |
| CP017 | SiMa.ai has raised roughly USD355 million in total and is reportedly valued at about USD1.4 billion after its August 2025 Series C. | 中 | SP021, SP025 |
| CP018 | SiMa.ai markets its Modalix MLSoC at 50+ TOPS under ten watts, emphasizing running vision and generative pipelines on a single chip. | 中 | SP003, SP022 |
| CP019 | Hailo-10H delivers about 40 INT4 TOPS at roughly 2.5 watts (~16 TOPS/W) with a generative-AI focus. | 中 | SP009, SP008 |
| CP020 | NVIDIA Jetson leads on peak throughput but at 15-60 watts, positioning it for performance-critical rather than power-constrained designs. | 中 | SP007, SP014 |
| CP021 | SiMa.ai and Hailo compete on efficiency (TOPS/W) while NVIDIA competes on peak performance, so no single vendor leads on all axes. | 中 | SP004, SP007 |
| CP022 | Ambarella wins the ultra-low-power AI-camera niche with sub-2W high-resolution video inference. | 中 | SP012, SP004 |
| CP023 | Blaize differentiates on a programmable graph-streaming architecture and EdgeCortix on a software-first Dynamic Neural Accelerator plus MERA compiler. | 中 | SP010, SP018 |
| CP024 | NVIDIA's CUDA/JetPack ecosystem is the most mature software stack in the field, which no edge challenger has yet dislodged. | 高 | SP014, SP002 |
| CP025 | SiMa.ai publishes system-on-module pricing of USD349 (8GB) and USD599 (32GB) and a USD1,499 developer kit, unusually transparent versus peers. | 中 | SP020, SP003 |
| CP026 | Most peers price on a design-win or OEM-negotiated basis and do not publish list prices, so pricing comparison has genuine gaps. | 中 | SP010, SP012 |
| CP027 | Edge-AI silicon carries high switching costs after design-in because layout, thermal design, software porting, and safety qualification are chip-specific. | 中 | SP007, SP020 |
| CP028 | Multi-homing is limited at the device level but common at the evaluation stage, where buyers benchmark several parts before committing. | 中 | SP007, SP004 |
| CP029 | Prior CUDA/JetPack investment makes NVIDIA Jetson the lowest-perceived-risk default for most engineering teams. | 中 | SP014, SP002 |
| CP030 | NVIDIA, Qualcomm, Ambarella, and Blaize have deeper or more established distribution channels than SiMa.ai's younger partner-led motion. | 中 | SP002, SP012 |
| CP031 | SiMa.ai relies on direct sales plus partners and distributors (Macnica, Enclustra) and strategic relationships (Synopsys, L&T Technology Services, Micron). | 中 | SP020, SP023 |
| CP032 | SiMa.ai, like most fabless peers, depends on TSMC (Modalix on N6) and LPDDR5 memory supply, a shared constraint rather than a differentiator. | 中 | SP022, SP023 |
| CP033 | SiMa.ai's performance-per-watt lead is narrow and perishable because Hailo claims comparable efficiency and NVIDIA iterates aggressively. | 中 | SP009, SP014 |
| CP034 | A software-first pitch must displace CUDA, the deepest and most entrenched moat in AI compute, which is a high-severity risk to SiMa.ai's thesis. | 中 | SP002, SP014 |
| CP035 | Hailo's collapse from a USD1.2 billion unicorn to a distressed sub-USD500 million SPAC demonstrates that a well-funded, credible edge-AI peer can fail to convert capital into durable value. | 高 | SP005, SP016 |
| CP036 | Even Ambarella, the healthiest independent, warns that rising R&D and operating costs could outpace revenue if design wins disappoint. | 中 | SP012 |
| CP037 | No retained public source discloses SiMa.ai's shipment volumes, revenue, or design-win count, so its competitive standing rests on product and partnerships rather than proven share. | 高 | SP021, SP002 |
| CP038 | SiMa.ai's reported ~USD1.4 billion valuation implies share capture that public evidence cannot yet confirm in a market where the incumbent's moat is the largest challenger risk. | 中 | SP002, SP005 |
| CI001 | SiMa.ai monetizes through a hardware-led model, selling Modalix MLSoC silicon, system-on-modules, and developer kits with the Palette software toolchain layered on top rather than sold separately. | 高 | SI001, SI011, SI013 |
| CI002 | SiMa.ai sells production Modalix system-on-modules at USD349 for the 8GB variant and USD599 for the 32GB variant in 1,000-unit quantities. | 中 | SI011, SI013 |
| CI003 | SiMa.ai prices its Modalix developer kit at USD1,499. | 中 | SI011, SI012 |
| CI004 | SiMa.ai distributes through partners and distributors including Macnica, Enclustra, and ThinkRobotics alongside a direct sales motion. | 中 | SI025, SI012 |
| CI005 | SiMa.ai's Palette software suite (Edgematic and the LLiMa on-device LLM framework) is bundled with hardware rather than sold as a standalone recurring license per public disclosures. | 中 | SI001, SI013 |
| CI006 | SiMa.ai states its Series C proceeds will scale Modalix production, expand globally, and deepen automotive and physical-AI go-to-market. | 高 | SI001, SI002 |
| CI007 | SiMa.ai has raised approximately USD355 million in total across seed through its August 2025 Series C. | 高 | SI001, SI002, SI016 |
| CI008 | SiMa.ai's August 2025 Series C was USD85 million, led by Maverick Capital with StepStone Group joining, and was described as oversubscribed. | 高 | SI001, SI017 |
| CI009 | SiMa.ai's funding chronology spans a 2020 Series A of about USD30 million, a 2021 Series B of about USD80 million with multiple extensions, a roughly USD70 million round in April 2024, and the 2025 Series C. | 中 | SI009, SI016 |
| CI010 | SiMa.ai received a strategic Micron investment reported around April 2026, separate from and following the headline Series C. | 中 | SI015, SI014 |
| CI011 | SiMa.ai discloses no revenue, ARR, gross margin, burn rate, or runway figures publicly. | 中 | SI009, SI010 |
| CI012 | Third-party trackers report SiMa.ai has raised across roughly nine to ten rounds backed by about fifteen investors. | 中 | SI009, SI010 |
| CI013 | As a fabless designer, SiMa.ai's cost structure is dominated by R&D, advanced-node tape-out and mask costs at TSMC N6, and inventory rather than owned fabrication. | 中 | SI011, SI013 |
| CI014 | Comparable fabless edge-AI vendor Ambarella reported record fiscal-2026 revenue of USD390.7 million, up 37.2% year over year. | 高 | SI003, SI004, SI005 |
| CI015 | Ambarella's fiscal-2026 GAAP gross margin was 59.2% and its non-GAAP gross margin was 60.7%. | 高 | SI003, SI004 |
| CI016 | Ambarella posted a fiscal-2026 GAAP net loss of USD75.9 million despite a non-GAAP profit of USD26.9 million. | 高 | SI003, SI005 |
| CI017 | Ambarella ended fiscal-2026 with about USD312.6 million in cash and marketable securities. | 中 | SI003, SI018 |
| CI018 | Ambarella derives roughly 80% of revenue from edge AI and has surpassed USD1 billion in cumulative edge-AI revenue. | 中 | SI003, SI005 |
| CI019 | Comparable public peer Blaize reported fiscal-2025 revenue of about USD38.6 million and an operating loss of roughly USD103.8 million. | 中 | SI021, SI019, SI008 |
| CI020 | Blaize's fiscal-2025 10-K disclosed substantial doubt about its ability to continue as a going concern absent additional financing. | 高 | SI006, SI008 |
| CI021 | Blaize was described as having heavy cash burn while scaling its edge-AI business in fiscal 2025. | 中 | SI008, SI020 |
| CI022 | Edge-AI fabless specialists typically run gross margins in the mid-50% to mid-60% range, below NVIDIA's 70%-plus scale-driven margins. | 中 | SI003, SI022 |
| CI023 | SiMa.ai's implied revenue is not publicly quantified, so its margin and unit economics can only be proxied from public peers. | 中 | SI009, SI021 |
| CI024 | SiMa.ai's reported post-Series C valuation of about USD1.4 billion was not confirmed by the company. | 中 | SI023, SI017 |
| CI025 | SiMa.ai crossed into unicorn status in 2025 amid a broader wave of newly minted AI and semiconductor unicorns. | 中 | SI024, SI023 |
| CI026 | Ambarella's fiscal-2026 non-GAAP operating expense rose about 12.9%, driven by higher labor and SoC development costs, a proxy for edge-AI R&D intensity. | 中 | SI005, SI003 |
| CI027 | With roughly USD355 million raised and no disclosed revenue, SiMa.ai's runway depends on undisclosed burn and periodic new financing, evidenced by a strategic Micron round soon after the Series C. | 中 | SI014, SI015 |
| CI028 | SiMa.ai's pricing is transparent for modules and kits but list-versus-realized pricing, volume discounts, and OEM contract terms are undisclosed. | 中 | SI011, SI012 |
| CI029 | SiMa.ai's revenue mix across chips, modules, kits, and software is not publicly broken out. | 中 | SI009, SI013 |
| CI030 | SiMa.ai targets robotics, industrial automation, automotive, aerospace and defense, smart vision, and healthcare as revenue verticals. | 中 | SI001, SI013 |
| CI031 | SiMa.ai's go-to-market leans on ecosystem partners such as Synopsys automotive tooling and L&T Technology Services delivery whose revenue contribution is undisclosed. | 中 | SI025, SI001 |
| CI032 | SiMa.ai's capital intensity is structurally high because each MLSoC generation requires a new advanced-node tape-out and sustained software investment before revenue. | 中 | SI013, SI011 |
| CI033 | The going-concern warning at public peer Blaize illustrates the financing-dependency risk facing undisclosed-revenue edge-AI silicon firms like SiMa.ai. | 中 | SI008, SI006 |
| CI034 | SiMa.ai's total capital raised of about USD355 million is comparable to Hailo's roughly USD340 million-plus, but SiMa.ai discloses neither revenue nor shipment volumes. | 中 | SI009, SI016 |
| CI035 | Ambarella's roughly 60% gross margin and continued GAAP net losses at USD390.7 million revenue signal the scale SiMa.ai would need before profitability. | 高 | SI004, SI003 |
| CI036 | SiMa.ai's next external financing appears event-driven, tied to production scale-up and strategic partnerships rather than a disclosed cash-out date. | 中 | SI014, SI001 |
| CI037 | Key private metrics including revenue, gross margin, burn, runway, backlog, and customer concentration remain undisclosed, constituting the primary diligence blockers for underwriting SiMa.ai. | 中 | SI009, SI010 |
| CI038 | SiMa.ai's Series C was described as oversubscribed, indicating continued investor demand despite no public revenue. | 中 | SI001, SI017 |
| CE001 | Modalix is SiMa.ai's second-generation MLSoC, a single-chip multi-modal edge-AI platform supporting CNNs, Transformers, LLMs, LMMs, and generative AI. | 中 | SE024, SE025, SE018 |
| CE002 | Modalix entered production in August 2025 following a September 2024 product announcement. | 高 | SE018, SE023, SE001 |
| CE003 | Modalix is built on TSMC's N6 process in a 25mm-by-25mm package. | 中 | SE007, SE022 |
| CE004 | Modalix delivers 50-plus TOPS under ten watts, with accelerator options at 25, 50, 100, and 200 INT8 TOPS. | 高 | SE007, SE024, SE020 |
| CE005 | Modalix's ML accelerator supports INT8, INT16, and BFLOAT16 for CNN, Transformer, LLM, and generative inference. | 中 | SE007, SE024 |
| CE006 | Modalix's application-processor complex is eight Arm Cortex-A65 cores running at 1.4GHz. | 中 | SE007, SE022 |
| CE007 | Modalix integrates a quad-core Synopsys ARC EV74 computer-vision unit and an Arm Mali-C71AE image-signal processor. | 中 | SE007, SE022 |
| CE008 | Modalix supports up to 32GB of LPDDR5 memory on the module and 8MB of on-chip SRAM. | 中 | SE007, SE010 |
| CE009 | Modalix I/O includes PCIe Gen5 by eight lanes, four 10-gigabit Ethernet ports, and four MIPI CSI-2 camera interfaces. | 中 | SE007, SE008 |
| CE010 | Modalix includes hardware secure boot with a root of trust and OTP key support. | 中 | SE007, SE008 |
| CE011 | SiMa.ai claims Modalix delivers more than ten times the performance per watt of alternatives. | 中 | SE024, SE025 |
| CE012 | SiMa.ai's software platform is Palette (Palette Neat), including a Model Compiler, Neat Library, and the sima-cli tool. | 中 | SE012, SE013, SE002 |
| CE013 | LLiMa is SiMa.ai's GenAI runtime for running LLMs and vision-language models on Modalix within the Palette environment. | 中 | SE012, SE002 |
| CE014 | SiMa.ai publishes a model zoo on GitHub (SiMa-ai/models) with PyTorch and ONNX models plus compilation and conversion scripts. | 中 | SE013, SE012 |
| CE015 | Edgematic is SiMa.ai's browser-based low-code pipeline builder with a GStreamer backend and a catalog of pre-optimized models. | 中 | SE009, SE005, SE006 |
| CE016 | Edgematic is cloud-hosted on AWS and integrates with Amazon SageMaker for edge model development. | 中 | SE015, SE005 |
| CE017 | Modalix system-on-modules are pin- and software-compatible with SiMa.ai's first-generation MLSoC and with popular NVIDIA SoMs. | 中 | SE024, SE011 |
| CE018 | SiMa.ai's product family spans the MLSoC chip, Modalix SoMs in 8GB and 32GB variants, the DevKit 3.0, and the Palette and Edgematic software. | 中 | SE010, SE007 |
| CE019 | Modalix supports ONNX models and multiple frameworks via the Palette ModelSDK. | 中 | SE013, SE012 |
| CE020 | SiMa.ai's first-generation MLSoC was listed on MLCommons MLPerf Inference 4.0 in March 2024. | 中 | SE019, SE007 |
| CE021 | Public MLPerf benchmark results specific to the Modalix generation have not been detailed as of mid-2026. | 中 | SE007, SE024 |
| CE022 | Modalix targets robotics, automotive, smart vision, drones, healthcare, retail, defense, and industrial applications. | 中 | SE024, SE003 |
| CE023 | SiMa.ai depends on TSMC for fabrication of Modalix on the N6 process. | 中 | SE007, SE020 |
| CE024 | SiMa.ai licenses core CPU IP from Arm (Cortex-A65) and vision IP from Synopsys (ARC EV74) for Modalix. | 中 | SE007, SE022 |
| CE025 | SiMa.ai and Synopsys announced a strategic collaboration in December 2024 to accelerate automotive edge-AI development. | 中 | SE016, SE001 |
| CE026 | SiMa.ai partners with L&T Technology Services to deliver physical-AI solutions across mobility, healthcare, industrial automation, and robotics. | 中 | SE017, SE003 |
| CE027 | Modalix includes hardware video encode and decode for H.264, H.265, and AV1 up to 4Kp60. | 中 | SE007, SE008 |
| CE028 | Edgematic lets developers upload ONNX models and measure real throughput, latency, and power KPIs on cloud-hosted MLSoC boards. | 中 | SE005, SE009 |
| CE029 | SiMa.ai positions Modalix as software-centric, prioritizing the developer toolchain to reduce integration friction. | 中 | SE025, SE019 |
| CE030 | The developer onboarding path at developer.sima.ai provides a self-contained install-to-GenAI workflow driven by the sima-cli tool. | 中 | SE012, SE013 |
| CE031 | SiMa.ai's differentiation rests on single-chip multi-modal integration and a software-first stack rather than raw peak TOPS. | 中 | SE019, SE024 |
| CE032 | Modalix uses a secure high-bandwidth network-on-chip to interconnect its functional blocks. | 中 | SE007, SE008 |
| CE033 | SiMa.ai's product maturity is production-stage for silicon and modules but its GenAI/LLiMa and Edgematic tooling are earlier-stage layers. | 中 | SE012, SE009 |
| CE034 | SiMa.ai maintains multiple public GitHub organizations (SiMa-ai/models and Palette Neat) for developer assets. | 中 | SE013, SE014 |
| CE035 | Modalix is available through distributors including Macnica for industrial and robotics customers. | 中 | SE007, SE010 |
| CE036 | SiMa.ai's Palette Neat SDK exposes Python and C++ APIs plus containers and CLI tooling. | 中 | SE012, SE013 |
| CE037 | SiMa.ai emphasizes on-device data privacy and enterprise-grade security in Edgematic's cloud-hosted evaluation. | 中 | SE005, SE015 |
| CE038 | SiMa.ai's roadmap has progressed from the first-generation MLSoC (MLPerf 4.0 in 2024) to Modalix production in 2025, with GenAI/LLM support as the newest capability. | 中 | SE018, SE012 |
| CU001 | SiMa.ai targets six customer verticals - robotics, automotive/mobility, industrial automation, aerospace and defense, smart vision, and healthcare - in the roughly 5W-to-25W embedded-edge power segment. | 高 | SU012, SU019, SU009 |
| CU002 | SiMa.ai's typical buyer is an OEM or system integrator embedding Modalix into a machine, while the ultimate user/operator is a separate enterprise, so buyer, user, and payer are often distinct. | 中 | SU019, SU012 |
| CU003 | SiMa.ai reaches customers through direct sales plus a distributor channel including Macnica, Enclustra, and ThinkRobotics. | 中 | SU018, SU025 |
| CU004 | SiMa.ai uses engineering and solution partners including L&T Technology Services, VVDN, Synopsys, and AWS to reach and enable customers. | 高 | SU013, SU015, SU017 |
| CU005 | SiMa.ai is investing its August 2025 Series C proceeds to expand its customer footprint in Korea, Japan, Europe, and the United States. | 中 | SU006, SU019 |
| CU006 | SiMa.ai earlier named healthcare, smart retail, autonomous vehicles, government, and robotics as priority applications within a target market it sized at roughly $40 billion. | 中 | SU009 |
| CU007 | SiMa.ai's customer count reportedly roughly doubled between 2023 and 2024 according to independent coverage. | 中 | SU009 |
| CU008 | By mid-2024 SiMa.ai had first-generation MLSoC silicon in customers' hands with second-generation Modalix samples following in Q4 2024. | 中 | SU009, SU021 |
| CU009 | SiMa.ai's Modalix MLSoC reached production availability in August 2025. | 高 | SU020, SU019 |
| CU010 | SiMa.ai's Modalix system-on-module won the Edge AI + Vision Alliance's 2026 Product of the Year award for "Best Edge AI Board" in April 2026. | 中 | SU005, SU011 |
| CU011 | SiMa.ai markets Modalix as pin- and software-compatible with incumbent GPU-based modules to shorten evaluation and lower switching friction from pilot to production. | 中 | SU019, SU020 |
| CU012 | SiMa.ai discloses no count of active deployments, units shipped, locations, or utilization for its customer base. | 中 | SU009, SU019 |
| CU013 | STIGA S.p.A., a leading European garden-machinery manufacturer, announced a strategic partnership (first reported February 2026, dated March 2026) to embed SiMa.ai's Modalix MLSoC into its next-generation robotic lawn mowers. | 高 | SU001, SU002, SU008 |
| CU014 | SiMa.ai announced an October 2024 partnership with TRUMPF to develop AI-powered lasers using its chips. | 中 | SU009 |
| CU015 | L&T Technology Services entered a September 2025 strategic partnership with SiMa.ai spanning in-vehicle infotainment, AD/ADAS, industrial automation, robotics, and healthcare on the MLSoC ONE platform. | 中 | SU013, SU014 |
| CU016 | SiMa.ai positions itself for aerospace and defense customers around disconnected, autonomous-drone, and battlefield edge environments, but no specific defense program or volume is public. | 低 | SU016, SU019 |
| CU017 | SiMa.ai's Modalix system-on-module launched publicly in 2025 before the 2026 Product of the Year recognition. | 中 | SU005, SU020 |
| CU018 | VVDN Technologies serves as a design and manufacturing partner helping SiMa.ai customers scale edge AI applications into production. | 低 | SU007 |
| CU019 | Synopsys and SiMa.ai announced a December 2024 strategic collaboration to accelerate automotive edge AI tooling and development. | 中 | SU015 |
| CU020 | Most of SiMa.ai's named customer relationships are strategic partnerships or design-in announcements rather than production case studies with quantified outcomes. | 中 | SU009, SU001, SU013 |
| CU021 | SiMa.ai discloses no net revenue retention, gross revenue retention, churn, renewal rate, contract length, or cohort-retention data. | 中 | SU019, SU009 |
| CU022 | Edge-silicon design-in cycles are long and costly, so a qualified Modalix design win historically implies a multi-year platform lifetime and high switching cost. | 中 | SU020, SU019 |
| CU023 | SiMa.ai's full-stack Palette software, Edgematic, LLiMa runtime, and model zoo raise the integration surface and increase customer switching cost. | 中 | SU017, SU012 |
| CU024 | Positive satisfaction signals include the 2026 Product of the Year award and public endorsement of the partnership by STIGA leadership. | 中 | SU005, SU001 |
| CU025 | SiMa.ai provides no public reference base with quantified satisfaction, no disclosed logo retention, and its second-generation product has been in production only since August 2025, so a full multi-year renewal cohort has not yet been demonstrated. | 中 | SU009, SU020 |
| CU026 | SiMa.ai's customer durability is structurally plausible from embedded-platform economics but empirically unproven given the absence of renewal and retention disclosure. | 中 | SU009, SU019 |
| CU027 | SiMa.ai runs a land-and-expand motion, winning an initial design-in then extending the same silicon-plus-software platform into adjacent use cases and new regions funded by the August 2025 Series C. | 中 | SU006, SU013, SU019 |
| CU028 | SiMa.ai's top-customer concentration cannot be measured because it discloses no per-customer revenue, but the qualitative picture is heavy dependence on a few early flagship partnerships. | 低 | SU009, SU001 |
| CU029 | A meaningful share of SiMa.ai's go-to-market runs through solution and distribution partners (LTTS, VVDN, Macnica, Enclustra, ThinkRobotics), adding intermediary margin and reducing direct customer control. | 中 | SU013, SU018, SU025 |
| CU030 | SiMa.ai's conservative target markets - automotive, defense, and industrial - impose long qualification cycles and functional-safety demands that create significant procurement friction. | 中 | SU015, SU019 |
| CU031 | The GPU pin- and software-compatibility that eases SiMa.ai adoption also lets customers dual-source or revert to entrenched suppliers such as NVIDIA. | 中 | SU020, SU011 |
| CU032 | SiMa.ai's customer proof remains far thinner than an incumbent like NVIDIA's, whose CUDA/JetPack ecosystem and installed base give it a documented, large customer footprint. | 低 | SU011, SU009 |
| CU033 | SiMa.ai is headquartered in San Jose, California, and sells the Modalix MLSoC, system-on-modules, and Palette software into its target verticals. | 高 | SU009, SU019 |
| CU034 | SiMa.ai publishes no revenue-band, account-size, or per-vertical revenue split, so its customer segmentation can only be described qualitatively. | 中 | SU019, SU009 |
| CU035 | SiMa.ai's Modalix robotics use cases include AMRs, humanoids, SLAM/3D navigation, conversational HMI with on-device small language models, and sensor fusion, all under 10W on-device. | 中 | SU012, SU004 |
| CU036 | SiMa.ai's deployment funnel is weighted toward early stages - announced partnerships and evaluation seeding - with fewer independently verifiable high-volume production programs. | 中 | SU009, SU001 |
| CU037 | SiMa.ai states that not all of its customers have been made public, which limits the depth of independent verification of its customer base. | 中 | SU009 |
| CU038 | SiMa.ai's AMR solution brief positions Modalix for deterministic, low-latency robotics autonomy running perception, localization, and control concurrently on a single device without cloud reliance. | 中 | SU004 |
| CR001 | SiMa.ai's highest-severity risk is competitive displacement by NVIDIA, whose Jetson platform and CUDA/JetPack software moat could confine SiMa to a niche. | 高 | SR010, SR020 |
| CR002 | Well-funded challengers (EdgeCortix over USD110 million, Hailo over USD340 million) crowd the edge-AI silicon market alongside NVIDIA. | 中 | SR024, SR030 |
| CR003 | Peer distress - Hailo's reported valuation collapse to under USD500 million with a distressed SPAC route and layoffs, and Blaize's going-concern warning - signals fragility in edge-AI silicon economics. | 中 | SR009, SR011, SR027 |
| CR004 | Fabless edge-AI players burn heavily, with Blaize posting an operating loss near USD103.8 million and Ambarella still running GAAP net losses at nearly USD391 million revenue. | 高 | SR011, SR013 |
| CR005 | SiMa.ai's risks are weighted toward market, capital, and execution exposure rather than near-term legal liability, given its pre-profitability hardware-led profile. | 中 | SR015, SR013 |
| CR006 | SiMa.ai's risk clusters rank by residual exposure as competitive displacement, financing/capital intensity, supply-chain concentration, regulatory exposure, execution/key-person, and technical validation. | 中 | SR010, SR015 |
| CR007 | US BIS export controls on advanced computing semiconductors restrict access to China/Macau and shrink SiMa.ai's addressable market there. | 高 | SR001, SR002 |
| CR008 | The 2026 BIS license-review process imposes know-your-customer, shipment-ratio caps, third-party testing, tariffs, and reporting obligations on covered advanced chips. | 高 | SR003, SR004, SR008 |
| CR009 | Automotive functional-safety certification (ISO 26262 ASIL C/D, requiring lock-step CPUs, ECC, FMEDA, and traceable safety architecture) is a de facto gate for automotive design-in that SiMa.ai has not publicly demonstrated it clears. | 中 | SR005, SR006 |
| CR010 | AI-specific automotive safety guidance (ISO/PAS 8800, SOTIF/ISO 21448) is still maturing, adding uncertainty to SiMa.ai's automotive qualification path. | 中 | SR006, SR005 |
| CR011 | Healthcare (FDA) and aerospace/defense (ITAR, procurement security) impose further qualification regimes on SiMa.ai's stated verticals. | 低 | SR002, SR014 |
| CR012 | SiMa.ai publishes a privacy policy last updated November 2025 governing collection, use, processing, and sharing of personal information. | 中 | SR007 |
| CR013 | No material litigation or IP dispute involving SiMa.ai is publicly identified as of the run date, though semiconductor IP-infringement exposure is a latent risk. | 低 | SR007, SR015 |
| CR014 | SiMa.ai's Modalix MLSoC is fabricated on TSMC's N6 process, concentrating its most critical manufacturing dependency in Taiwan. | 高 | SR015, SR016 |
| CR015 | A disruption at TSMC (disaster, cross-strait escalation, capacity reallocation, or trade restriction) would directly threaten SiMa.ai's ability to build product. | 中 | SR015, SR001 |
| CR016 | Meaningful diversification to Samsung, Intel, or US/Europe fabs for advanced automotive-grade nodes is not expected to materialize before roughly 2027. | 低 | SR005, SR006 |
| CR017 | As a fabless company SiMa.ai also depends on OSAT packaging/test partners, on Arm and Synopsys IP, and on distributors/contract manufacturers, so a failure at any layer propagates to shipments. | 中 | SR029, SR014 |
| CR018 | SiMa.ai's target verticals - automotive, industrial, defense, healthcare - demand stringent reliability, so a field failure or recall in a safety-critical deployment would be reputationally severe. | 中 | SR005, SR014 |
| CR019 | There is no public Modalix-generation MLPerf Inference result; SiMa.ai's first-generation MLSoC appeared on MLPerf in 2024, but current performance-per-watt claims rest largely on company-reported figures. | 中 | SR016, SR015 |
| CR020 | SiMa.ai depends on OSAT partners for packaging and test, adding a dependency layer whose terms and capacity are not publicly disclosed. | 低 | SR029 |
| CR021 | NVIDIA is simultaneously SiMa.ai's benchmark competitor and an ecosystem gravity well; Modalix's pin/software compatibility with NVIDIA modules eases adoption but also lets customers revert to NVIDIA. | 中 | SR020, SR010 |
| CR022 | SiMa.ai depends on TSMC as its sole disclosed foundry, with no public alternate source for Modalix. | 中 | SR015, SR016 |
| CR023 | SiMa.ai depends on Arm CPU IP and Synopsys IP and automotive tooling for its architecture and design flow. | 中 | SR016, SR014 |
| CR024 | A distributor and solution-partner channel (LTTS, VVDN, Macnica, Enclustra, ThinkRobotics) carries a meaningful share of SiMa.ai's customer delivery, adding intermediary margin and reducing direct control. | 中 | SR014, SR029 |
| CR025 | SiMa.ai's customer concentration cannot be measured because it discloses no per-account revenue, but the public picture is dependence on a small number of largely pre-volume flagship design-ins. | 中 | SR015, SR014 |
| CR026 | SiMa.ai depends on a syndicate of venture and strategic investors (Maverick Capital, Fidelity, Dell, Micron Ventures, StepStone) and their continued willingness to fund. | 高 | SR014, SR022, SR029 |
| CR027 | The April 2026 strategic Micron investment is a positive validation signal but also underscores SiMa.ai's reliance on strategic backers. | 中 | SR022 |
| CR028 | SiMa.ai discloses no revenue, gross margin, monthly burn, cash balance, or runway, so capital adequacy and path to profitability cannot be verified. | 高 | SR015, SR014 |
| CR029 | Public peers imply SiMa.ai faces mid-50% to mid-60% gross margins (below NVIDIA's 70%-plus) and sustained losses; Ambarella still lost about USD75.9 million on record revenue while Blaize lost near USD103.8 million operating. | 高 | SR013, SR011, SR025 |
| CR030 | SiMa.ai's fabless model is capital-intensive because each MLSoC generation requires a fresh advanced-node tape-out at TSMC plus sustained software investment before generating revenue. | 中 | SR015, SR016 |
| CR031 | SiMa.ai carries a reported but company-unconfirmed post-Series-C valuation near USD1.4 billion. | 中 | SR018 |
| CR032 | With no disclosed revenue to anchor a multiple, SiMa.ai's implied revenue multiple is uncomputable and the unicorn label rests on private-market sentiment rather than fundamentals. | 中 | SR018, SR015 |
| CR033 | The consolidated financial thesis-break trigger for SiMa.ai is a failure to raise the next round on non-punitive terms before runway expires. | 中 | SR009, SR011 |
| CR034 | SiMa.ai has raised roughly USD355 million across about nine to ten rounds backed by around fifteen investors, implying recurring dilution to fund its long path to profitability. | 高 | SR029, SR014, SR023 |
| CR035 | The mitigation for competitive risk is deepening the Palette software moat and locking in multi-year design wins, monitored via design-win-to-production conversion. | 低 | SR010, SR014 |
| CR036 | The mitigation for financing risk is the recent Series C and Micron backing, monitored via cash runway and burn multiple, with a down-round as the kill criterion. | 中 | SR014, SR022 |
| CR037 | The mitigation for technical-validation risk is publishing an independent Modalix benchmark, with results materially below company claims as the kill criterion. | 低 | SR016 |
| CR038 | The mitigation for execution risk is broadening the leadership bench beyond founder-CEO Krishna Rangasayee, whose departure without a credible successor is a kill criterion. | 低 | SR017 |
| CR039 | Priority diligence asks are a management data room (financials, burn, runway, backlog), an independent Modalix benchmark, foundry/OSAT contracts, an export-control compliance review, and functional-safety certification evidence. | 中 | SR015, SR003 |
| CR040 | SiMa.ai relies heavily on founder-CEO Krishna Rangasayee (ex-Xilinx, ex-Groq COO) for vision and fundraising, a key-person concentration risk. | 中 | SR017, SR015 |
| CR041 | SiMa.ai's other named co-founders are not independently corroborated in public filings, a governance and diligence gap. | 低 | SR017 |
| CR042 | For an edge-AI designer targeting global robotics, industrial, and automotive customers, BIS export controls both shrink the China market and raise compliance cost and diversion-liability risk. | 中 | SR003, SR008 |
| CR043 | Translating design wins into volume production is a core execution challenge, echoing the "production inertia" that has historically stalled hardware startups. | 中 | SR015, SR013 |
| CR044 | The pin- and software-compatibility with NVIDIA modules that aids SiMa.ai adoption also lowers the barrier for customers to dual-source or revert to NVIDIA. | 中 | SR020, SR010 |
| CR045 | Even Ambarella-scale revenue near USD390.7 million is insufficient for GAAP profitability in edge-AI fabless, implying SiMa.ai faces a long, capital-hungry road. | 高 | SR013, SR021 |
| CV001 | SiMa.ai carries a reported-but-company-unconfirmed post-Series-C valuation of roughly USD1.4 billion, attributed to The Information and repeated by secondary trackers. | 中 | SV010, SV003 |
| CV002 | Ambarella reported FY2026 revenue of USD390.7 million (up 37.2 percent) with a GAAP net loss of USD75.9 million and non-GAAP profitability, trading at roughly 5x to 8x price-to-sales. | 高 | SV014, SV020 |
| CV003 | SiMa.ai's Modalix MLSoC is built on TSMC N6 and paired with the Palette software suite as a single-chip low-power platform spanning computer vision through on-device generative and reasoning models. | 中 | SV024, SV003 |
| CV004 | SiMa.ai has marquee engagements including STIGA, TRUMPF, LTTS, VVDN, and Synopsys, and won the 2026 Edge AI and Vision Alliance Product of the Year for the Modalix SOM. | 中 | SV024, SV003 |
| CV005 | SiMa.ai is backed by top-tier investors including Maverick Capital, Fidelity, Point72, and StepStone Group. | 中 | SV011, SV025 |
| CV006 | SiMa.ai does not disclose revenue, gross margin, operating burn, or runway, so its reported valuation lacks a public financial anchor. | 中 | SV001, SV010 |
| CV007 | At an estimated USD35.5 million revenue, SiMa.ai's reported valuation implies a stretched roughly 28x to 39x forward revenue multiple, far above Ambarella's roughly 5x to 8x price-to-sales. | 中 | SV002, SV014 |
| CV008 | There is no public Modalix-generation MLPerf result to validate SiMa.ai's performance-per-watt claim independently, and the automotive vertical is gated by ISO 26262 certification SiMa has not publicly demonstrated. | 中 | SV024, SV022 |
| CV009 | NVIDIA's Jetson and CUDA/JetPack ecosystem is entrenched, and challengers continue to raise capital but face the incumbent's ecosystem dominance as the central obstacle. | 中 | SV022 |
| CV010 | The investment recommendation for SiMa.ai is research-more, conditional, at the reported roughly USD1.4 billion entry, with medium confidence in the thesis and low confidence in the price. | 中 | SV010, SV001 |
| CV011 | SiMa.ai's valuation stance is rich and largely unsupported by disclosed fundamentals given the roughly 28-39x implied multiple against public peers near 5-8x. | 中 | SV002, SV014 |
| CV012 | The most probable exit for SiMa.ai is strategic M&A rather than a near-term IPO, given active edge-AI acquisition activity and Micron's strategic stake. | 中 | SV004, SV005, SV026 |
| CV013 | SiMa.ai's risk rating is high, driven by NVIDIA displacement risk, capital-intensive fabless burn, single-foundry dependence on TSMC, and BIS export-control and ISO 26262 qualification friction. | 中 | SV022, SV021 |
| CV014 | A three-scenario model produces a probability-weighted value near USD965 million, below the reported USD1.4 billion entry price. | 低 | SV002, SV014 |
| CV015 | At the reported entry price the expected return is negative unless the bull scenario materializes. | 低 | SV002 |
| CV016 | The recommendation upgrades to buy only on satisfaction of gating diligence conditions, principally audited financials and Series C preference terms. | 中 | SV001, SV010 |
| CV017 | SiMa.ai has raised approximately USD355 million across roughly nine rounds since 2018, culminating in an August 2025 Series C of USD85 million. | 高 | SV012, SV013, SV011 |
| CV018 | The Series C was led by Maverick Capital, with StepStone Group joining and Micron Ventures participating, and was described as oversubscribed. | 高 | SV013, SV028 |
| CV019 | Micron made a strategic investment in SiMa.ai in April 2026 (amount undisclosed), a qualitative validation that does not establish a verifiable valuation mark. | 中 | SV026, SV027 |
| CV020 | Private-market trackers disagree on SiMa.ai's valuation, with one timestamping USD960 million in mid-2025 and a roughly 2.7x valuation-to-funding capital-efficiency ratio. | 中 | SV001, SV010 |
| CV021 | Because SiMa.ai is private and unaudited, an investor cannot see the liquidation-preference stack or participation rights attached to roughly USD355 million of preferred capital, creating preference-overhang risk in a downside exit. | 中 | SV001, SV011 |
| CV022 | A capital-intensive fabless company with no disclosed profitability will need further financing, so continued dilution and potential reset of the preference stack are likely. | 中 | SV018, SV021 |
| CV023 | The bull case (roughly 25 percent) assumes revenue scaling toward USD150 million by 2028 and a premium roughly 10x to 17x exit multiple, yielding roughly USD1.5 billion to USD2.5 billion. | 低 | SV007, SV005 |
| CV024 | The base case (roughly 45 percent) assumes revenue near USD80 million and a blended edge-AI multiple in the high-single to low-double digits, yielding roughly USD0.6 billion to USD1.0 billion, at or below the reported entry. | 低 | SV014, SV020 |
| CV025 | The bear case (roughly 30 percent) assumes niche confinement by NVIDIA, revenue stalling below USD40 million, and a dilutive down-round or distressed process akin to Hailo, yielding roughly USD0.2 billion to USD0.5 billion. | 低 | SV021, SV022 |
| CV026 | Because SiMa discloses no revenue, the scenario boundaries are inferred from a third-party revenue estimate and edge-AI sector benchmarks rather than verified against SiMa financials. | 中 | SV009, SV006 |
| CV027 | The sensitivity analysis shows that only the combination of high revenue and a sustained premium multiple justifies the reported price; at public-peer multiples the reported valuation is unsupported across the plausible revenue range. | 中 | SV002, SV014 |
| CV028 | AI M&A revenue multiples in 2025-2026 cluster in a roughly 25x to 30x range for attractive growth targets, with fabless hardware typically pricing below software. | 中 | SV007, SV006 |
| CV029 | Ambarella is the closest public pure-play edge-AI comparable, with gross margins near 60 percent and cash of USD312.6 million, and a market capitalization in the roughly USD2 billion to USD3 billion range. | 高 | SV014, SV020 |
| CV030 | Blaize reported FY2025 revenue near USD38.6 million, an operating loss near USD103.8 million, and a going-concern warning, illustrating how the market repudiates edge-AI names that scale losses faster than revenue. | 高 | SV016, SV018 |
| CV031 | Hailo raised more than USD340 million but saw its valuation reportedly fall below USD500 million into a distressed SPAC route with layoffs. | 中 | SV021, SV030 |
| CV032 | NXP acquired edge-AI NPU startup Kinara for USD307 million in an all-cash deal in February 2025, a strategic-buyer precedent for edge-AI IP. | 中 | SV004 |
| CV033 | Onsemi agreed to acquire Synaptics in a roughly USD7 billion all-stock transaction to expand in edge AI, evidencing strong strategic-buyer appetite and premiums for differentiated edge-AI IP. | 中 | SV005 |
| CV034 | EdgeCortix closed an oversubscribed Series B taking total funding over USD110 million, evidencing continued private capital availability for edge-AI silicon peers. | 中 | SV029 |
| CV035 | On disclosed fundamentals the comparable set supports a valuation well below the reported USD1.4 billion, while a strategic acquirer could pay a scarcity premium for physical-AI silicon. | 中 | SV014, SV005 |
| CV036 | A down-round or flat financing priced at or below the reported USD1.4 billion would confirm overvaluation and echo the Hailo repricing precedent. | 中 | SV021 |
| CV037 | A sustained revenue shortfall below roughly USD40 million or repeated design losses to NVIDIA Jetson would make any premium multiple unsupportable and move SiMa toward the bear scenario. | 中 | SV022 |
| CV038 | Failure to achieve ISO 26262 automotive certification by roughly 2027, a TSMC N6 allocation loss, or cash exhaustion forcing a distressed raise are discrete monitorable kill criteria. | 中 | SV024, SV018 |
| CV039 | The two gating final diligence asks are audited revenue, gross margin, burn, and runway, and the Series C term sheet including liquidation preferences, participation, and seniority. | 中 | SV001, SV010 |
| CV040 | Additional diligence asks include the full cap table and dilution history, a dollar-weighted design-win pipeline with production timelines, and customer concentration. | 中 | SV011, SV025 |
| CV041 | A foundry and OSAT second-source qualification roadmap and an independent Modalix-generation benchmark are needed to assess supply resilience and validate the performance claim. | 中 | SV024 |
| CV042 | The single metric that best explains why the recommendation is research-more rather than buy is the absence of any disclosed revenue against which to test the reported valuation. | 中 | SV001, SV010 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | SiMa.ai | SiMa.ai: Scaling Physical AI | Hardware and software that accelerates physical AI application development and deployment. |
| SO002 | SiMa.ai | SiMa.ai Raises $85M to Scale Physical AI, Bringing Total Funding to $355M | SiMa.ai today announced it has raised $85 million in an oversubscribed round, bringing total capital raised to $355 million. |
| SO003 | SiMa.ai | Our Story - SiMa.ai | That belief has driven every decision we've made since 2018. |
| SO004 | SiMa.ai | CEO Bio: Krishna Rangasayee | He was with Xilinx for 18 years... Previously, he was the COO of Groq. He holds 25+ international patents. |
| SO005 | SiMa.ai | Robotics - SiMa.ai | SiMa.ai powers physical AI where cloud latency is not an option. |
| SO006 | SiMa.ai | Palette Software Suite - SiMa.ai | Palette software suite with SDK and no-code Edgematic development tool. |
| SO007 | PR Newswire | SiMa.ai Raises $85M to Scale Physical AI, Bringing Total Funding to $355M | Maverick Capital Led the Oversubscribed Round with StepStone Group Joining as a New Investor. |
| SO008 | PR Newswire | SiMa.ai Next-Gen Platform for Physical AI in Production | Modalix... supporting LLMs, transformers, CNNs, and GenAI workloads under 10 watts... commercial-grade 1K units pricing starting at $349 for the 8GB SoM. |
| SO009 | Business Standard | Chip startup SiMa.ai lands $85 mn to power AI robots, autonomous cars | The company declined to disclose its valuation following this funding round. |
| SO010 | EE Times | SiMa.ai's Second-Gen Edge AI Chip Goes Multi-Modal | SiMa.ai's second-generation Modalix MLSoC targets multimodal edge AI. |
| SO011 | Data Center Dynamics | SiMa.ai launches MLSoC Modalix, a multi-modal platform for Edge AI | SiMa.ai launches its second-generation multimodal MLSoC Modalix for edge AI. |
| SO012 | SiliconANGLE | SiMa.ai launches its next-gen system-on-chip for physical AI into production | SiMa.ai moves its next-generation Modalix system-on-chip into production. |
| SO013 | Edge AI and Vision Alliance | SiMa.ai Next-Gen Platform for Physical AI in Production | Modalix now in production for Physical AI applications. |
| SO014 | Macnica | MLSoC Modalix - Second generation SoC for physical AI | Modalix delivers 25 to 50 TOPS INT8 ML performance with power consumption under 10 watts. |
| SO015 | Unite.AI | Krishna Rangasayee, Founder & CEO of SiMa.ai – Interview Series | Before founding SiMa.ai in 2018, he spent nearly 18 years at Xilinx. |
| SO016 | Barchart | Micron Is Investing in SiMa.ai. What Does That Mean for MU Stock? | Micron Ventures is identified as a strategic investor aligned with SiMa.ai. |
| SO017 | NextGen Defense | Defense Disruptors: Inside SiMa.ai's Push to Secure the AI Edge | SiMa.ai was founded in 2018 with Krishna Rangasayee as founder and CEO. |
| SO018 | Data Center Dynamics | Edge compute chip firm SiMa.ai raises $70m in funding round | SiMa.ai raises $70 million in a round led by Maverick Capital. |
| SO019 | Clay | How Much Did SiMa.ai Raise? Funding & Key Investors | SiMa.ai has raised funding across ten rounds... latest being a Series C... total at least $355M. |
| SO020 | Synopsys | Synopsys and SiMa.ai Announce Strategic Collaboration to Accelerate Development of Automotive Edge AI Solutions | Synopsys and SiMa.ai announce a strategic collaboration for automotive edge AI. |
| SO021 | L&T Technology Services | LTTS and SiMa.ai Forge Strategic Alliance to Advance Physical AI | LTTS and SiMa.ai forge a strategic alliance to advance Physical AI across mobility, healthcare, industrial automation, and robotics. |
| SO022 | CEO Insider | Interview with Krishna Rangasayee, CEO of SiMa.ai | Krishna Rangasayee, Founder & CEO of SiMa.ai. |
| SO023 | TechCrunch | More than 100 new tech unicorns were minted in 2025 — here they are | A running list of the tech companies that reached unicorn valuations in 2025. |
| SO024 | TheTopVoices | SiMa.ai Unveils Modalix MLSoC for On-Device Physical AI and GenAI | SiMa.ai unveils Modalix MLSoC for on-device Physical AI and GenAI. |
| SO025 | Digital Terminal | SiMa.ai Raises $85 Million to Accelerate Physical AI Innovation and Global Expansion | SiMa.ai raises $85 million to accelerate Physical AI innovation and global expansion. |
| SO026 | CoinUnited | SiMa.ai Closes $85M Series C with Micron Ventures Backing | SiMa.ai... closed an oversubscribed $85M Series C on August 1, 2025... Micron Ventures... has been identified as an aligned investor. |
| SO027 | The Information (via Megadose) | Edge Inference Chip Startup SiMa.ai Raising at $1.4 Billion Valuation | SiMa.ai... is in talks with investors to raise more than $100 million... at a valuation of around $1.4 billion... a premium of more than 45% to its $960 million valuation from last August, according to PitchBook. |
| SO028 | EPR News | SiMa.ai Secures $85 Million to Advance Physical AI Innovations | SiMa.ai secures $85 million... continuing investors including Dell Technologies Capital and Fidelity Management. |
| SO029 | Robotics Business News | L&T Technology Services Partners with SiMa.ai to Drive Next-Gen AI Solutions | L&T Technology Services partners with SiMa.ai to drive next-gen AI solutions. |
| SO030 | CNBC | Nvidia AI chip rivals attract record funding as competition heats up | Nvidia's established software stack and hardware ecosystem remain key hurdles to adoption... many smaller startups struggling or failing. |
| SM001 | MarketsandMarkets | Edge AI Chip / Edge AI Hardware Market Size, Share & Trends | The edge AI hardware market is projected to grow from USD 26.14 billion in 2025 to USD 58.90 billion by 2030, at a CAGR of 17.6%. |
| SM002 | Fortune Business Insights | Edge AI Market Size, Share & Growth Report | The global edge AI market size was valued at USD 35.60 billion in 2025... projected to grow from USD 46.96 billion in 2026 to USD 445.75 billion by 2034, exhibiting a CAGR of 32.5%. |
| SM003 | Global Market Insights | Edge AI Market Size, Forecasts Report 2026-2035 | Edge AI market was valued at USD 25.2 billion in 2025 and is estimated to grow to USD 225.5 billion by 2035, at a CAGR of 24.7%. |
| SM004 | Axis Intelligence (citing Grand View Research) | Edge AI Statistics 2026: Market Size, Chips, Adoption & Sector Data | The global edge AI market was valued at $24.9 billion in 2025 and is projected to reach $30.0 billion in 2026, growing at a 21.7% CAGR through 2033 to hit $118.7 billion... Hardware remains the dominant segment at 51.8%. |
| SM005 | MarketsandMarkets | Physical AI Market Size, Share, Technology & Trends | The physical AI market is projected to reach USD 15.24 billion by 2032, growing at a CAGR of 47.2% from 2026. |
| SM006 | TechRT | Physical AI Statistics 2026: Powerful Market Stats | Another estimate values the broader physical AI ecosystem at $383 billion in 2026, with projections reaching $3.26 trillion by 2040... AI robotics markets are expected to grow from $6.11 billion in 2025 to $33.39 billion by 2030. |
| SM007 | Boston Consulting Group | How Physical AI Is Reshaping Robotics Today | Physical AI is reshaping robotics as intelligence moves from software into machines that sense, move, and act in the real world. |
| SM008 | MarketsandMarkets | Automotive AI Market Size, Share & Trends | The automotive AI market is expected to reach USD 38.45 billion by 2030 from USD 18.83 billion in 2025, at a CAGR of 15.3%. |
| SM009 | Mobility Foresights | Global Automotive AI SoC Market Size and Forecast 2030 | The Automotive AI SoC Market is expected to grow at a CAGR of 15.6% from 2024 to 2034 with overall sales revenue estimated to reach US$ 13.0 billion by the end of 2034. |
| SM010 | Next Move Strategy Consulting | Advanced Driver Assistance Systems (ADAS) Market Growth Analysis 2030 | The global Advanced Driver Assistance Systems (ADAS) Market size was valued at $20.73 billion in 2021 and is predicted to reach USD 74.57 billion by 2030 with a CAGR of 14.2%. |
| SM011 | Mordor Intelligence | Autonomous Vehicles Semiconductor Market Forecasts to 2030 | The autonomous vehicles semiconductor market covers the chips, including AI SoCs and sensors, that enable autonomous-driving systems through 2030. |
| SM012 | Polaris Market Research | Edge AI Market Size, Share & Growth Forecast Report 2026-2034 | Global 5G connections rose sharply to 1.76 billion in 2023... forecast to grow to 7.9 billion by 2028, strengthening the viability of deploying AI algorithms directly on devices. |
| SM013 | Wevolver | The 2026 Edge AI Technology Report: Future of Edge AI | Edge AI is shifting from cloud offload to native, end-to-end decision-making, driven by latency, privacy, and power efficiency and constrained by security, fragmentation, and skills gaps. |
| SM014 | AIMultiple | Edge AI Chips Benchmark and Vendor Landscape | The edge AI chip landscape spans NVIDIA Jetson, Hailo, Ambarella, Qualcomm, and challengers such as SiMa.ai, with NVIDIA holding a leading position at the edge. |
| SM015 | Yahoo Finance (ResearchAndMarkets) | Edge AI Hardware Market worth $58.90 billion by 2030 | The edge AI hardware market is projected to reach $58.90 billion by 2030, up from $26.14 billion in 2025, growing at a CAGR of 17.6%. |
| SM016 | SiMa.ai | SiMa.ai homepage - Physical AI platform | SiMa.ai delivers the industry's first software-centric, purpose-built MLSoC platform for physical AI at the edge. |
| SM017 | SiMa.ai | SiMa.ai Robotics and physical AI solutions | SiMa.ai targets robotics, industrial automation, automotive, aerospace and defense, smart vision, and healthcare with its MLSoC. |
| SM018 | Edge AI and Vision Alliance | SiMa.ai Next-Gen Platform for Physical AI | SiMa.ai's Modalix MLSoC targets physical AI applications across robotics, industrial, automotive, and defense. |
| SM019 | Macnica | SiMa.ai MLSoC product and distribution overview | SiMa.ai's MLSoC integrates Arm cores with a dedicated ML accelerator for low-power edge inference across vision and generative workloads. |
| SM020 | CNBC | Nvidia AI chip rivals draw funding as edge challengers scale | Nvidia dominates the AI chip market, and rivals face a steep climb to win share even as investors fund challengers. |
| SM021 | Barchart | Micron is investing in SiMa.ai as edge AI demand grows | Micron's investment in SiMa.ai reflects growing demand for memory and compute in edge AI and physical AI systems. |
| SM022 | NextGen Defense | Defense Disruptors: SiMa.ai | SiMa.ai's low-power MLSoC is positioned for aerospace and defense edge applications requiring ruggedized, power-efficient inference. |
| SM023 | Unite.AI | Krishna Rangasayee, Founder and CEO of SiMa.ai - Interview | Rangasayee describes physical AI at the edge as a large, underserved market where power efficiency and a software-first experience are decisive. |
| SM024 | TechCrunch | At least 36 new tech unicorns were minted in 2026 | Semiconductor and AI-infrastructure startups featured prominently among newly minted unicorns as investors chased AI compute demand. |
| SM025 | Data Center Dynamics | SiMa.ai launches MLSoC Modalix for physical AI | SiMa.ai's Modalix MLSoC brings generative and computer-vision AI to edge devices at low power for physical AI use cases. |
| SP001 | SiMa.ai | SiMa.ai homepage - Physical AI platform | SiMa.ai delivers the industry's first software-centric, purpose-built MLSoC platform for physical AI at the edge. |
| SP002 | CNBC | Nvidia AI chip rivals draw funding as edge challengers scale | Nvidia dominates the AI chip market, and rivals face a steep climb to win share even as investors fund challengers. |
| SP003 | EE Times | SiMa.ai's Second-Gen Edge AI Chip Goes Multi-Modal | SiMa.ai's second-generation Modalix MLSoC targets multi-modal generative AI at the edge under a low power envelope. |
| SP004 | AIMultiple | Top 15 Edge AI Chip Makers with Use Cases | The edge AI chip field spans NVIDIA Jetson, Qualcomm, Hailo, Ambarella, Blaize, and SiMa.ai, differentiated by TOPS, power, and software. |
| SP005 | Calcalist | AI chip startup Hailo sees valuation halved to under $500 million ahead of SPAC | Hailo's valuation has fallen by more than half from its peak of $1.2 billion, now worth less than $500 million... it completed a Series C extension of $120 million in April 2024. |
| SP006 | Chip.computer | Best Edge AI Chips in 2026 | Chip.computer compares 550+ chips including edge AI parts from NVIDIA, SiMa.ai, Hailo, and Ambarella on performance and power. |
| SP007 | Benned | Edge AI Chips: NVIDIA Jetson, Qualcomm, Apple Neural Engine Compared (2026) | Edge AI chips span five orders of magnitude in power and compute; choosing one is about matching compute, power, memory, and software ecosystem to the deployment. |
| SP008 | EE Times | Hailo Debuts Edge GenAI Chip, Raises $120 Million | Hailo raised $120 million and debuted the Hailo-10, an edge AI accelerator bringing generative AI to edge devices at low power. |
| SP009 | Data Center Dynamics | Hailo releases Hailo-10H Edge AI chip | The Hailo-10H delivers up to 40 TOPS at low power for generative AI on edge devices such as PCs, automotive, and security. |
| SP010 | Blaize Holdings (Business Wire) | Blaize Announces Fourth Quarter and Full-Year 2025 Financial Results | In 2025, Blaize delivered $38.6 million in revenue, up from $1.6 million in 2024, marking its first full year of commercial revenue generation. |
| SP011 | Stock Analysis | Blaize Holdings (BZAI) Revenue 2022-2026 | Blaize (BZAI) reported 2025 revenue of $38.6 million, a sharp increase from $1.6 million in 2024. |
| SP012 | Sahm Capital | How Investors Are Reacting To Ambarella Record Edge AI Revenue Surge | Ambarella reported fiscal 2026 revenue of US$390.7 million, with Edge AI accounting for 80% and over 42 million Edge AI SoCs shipped; the Hanwha agreement carries an US$800 million opportunity. |
| SP013 | Yahoo Finance | Ambarella Inc (AMBA) Q4 2026 Earnings Call Highlights - Record Revenue | Ambarella reported record fiscal 2026 revenue with rapid edge-AI growth and a rising edge-AI revenue mix. |
| SP014 | NVIDIA | NVIDIA Jetson Modules | NVIDIA Jetson modules deliver accelerated AI performance at the edge with the JetPack SDK and unified CUDA-X software across the family. |
| SP015 | Tracxn | Hailo - 2026 Company Profile, Funding & Competitors | Hailo is a provider of neural learning processors for edge AI applications embedded into devices, with total funding across multiple rounds. |
| SP016 | SaaS Sentinel | AI Chip Startup Hailo Plans SPAC Merger After Valuation Falls 50% to Under $500M | Hailo plans a SPAC merger after its valuation fell 50% to under $500 million amid funding pressure and layoffs. |
| SP017 | Green Stock News | Blaize Announces Fourth Quarter and Full-Year 2025 Financial Results | Blaize reported full-year 2025 revenue of $38.6 million with fourth-quarter revenue more than doubling from the prior quarter. |
| SP018 | EdgeCortix (Business Wire) | EdgeCortix Closes Oversubscribed Series B, Bringing Total Funding Over $110M | EdgeCortix closed an oversubscribed Series B, bringing total funding to over $110 million amid growing edge AI demand. |
| SP019 | Forbes | Japanese Semiconductor Startup Secures $21 Million In Grants For Edge AI | EdgeCortix secured about $21 million in government-backed grants to develop next-generation edge AI chiplets. |
| SP020 | Macnica | SiMa.ai MLSoC product and distribution overview | SiMa.ai's MLSoC integrates Arm cores with a dedicated ML accelerator for low-power edge inference across vision and generative workloads. |
| SP021 | Business Standard | SiMa.ai lands $85M Series C to scale physical AI | SiMa.ai raised an $85 million Series C, bringing total funding to about $355 million to scale its physical-AI platform. |
| SP022 | Data Center Dynamics | SiMa.ai launches MLSoC Modalix for physical AI | SiMa.ai's Modalix MLSoC brings generative and computer-vision AI to edge devices at low power for physical AI use cases. |
| SP023 | Barchart | Micron is investing in SiMa.ai as edge AI demand grows | Micron's investment in SiMa.ai reflects growing demand for memory and compute in edge AI and physical AI systems. |
| SP024 | Unite.AI | Krishna Rangasayee, Founder and CEO of SiMa.ai - Interview | Rangasayee positions SiMa.ai on a software-first experience and performance-per-watt advantage against incumbent edge platforms. |
| SP025 | PR Newswire | SiMa.ai raises $85M to scale Physical AI | SiMa.ai raised $85 million in an oversubscribed Series C to scale its physical-AI MLSoC platform. |
| SI001 | SiMa.ai | SiMa.ai Raises $85M to Scale Physical AI, Bringing Total Funding to $355M | SiMa.ai raised $85 million in an oversubscribed Series C, bringing total funding to $355 million to scale physical AI at the edge. |
| SI002 | PR Newswire | SiMa.ai Raises $85M to Scale Physical AI, Bringing Total Funding to $355M | SiMa.ai raised $85 million in Series C funding led by Maverick Capital, bringing total funding to $355 million. |
| SI003 | StockTitan | Ambarella Announces Fourth Quarter and Fiscal Year 2026 Financial Results | Ambarella reported fiscal 2026 revenue of $390.7M, up 37.2% year-over-year, with GAAP gross margin of 59.2% and a GAAP net loss of $75.9M. |
| SI004 | U.S. Securities and Exchange Commission | Ambarella, Inc. Form 10-K for fiscal year ended January 31, 2026 | Annual report on Form 10-K for the fiscal year ended January 31, 2026, filed by Ambarella, Inc. with the SEC. |
| SI005 | The Motley Fool | Ambarella (AMBA) Q4 2026 Earnings Call Transcript | Annual revenue was $390.7 million, a 37.2% increase year over year, with HAI products contributing approximately 80% of total revenue. |
| SI006 | StockTitan | Blaize Holdings, Inc. Files Annual Report (Form 10-K) for Fiscal Year 2025 | Blaize Holdings, Inc. Form 10-K for the fiscal year ended December 31, 2025 discloses substantial net losses and going-concern considerations. |
| SI007 | Last10K | Blaize Holdings, Inc. (BZAI) 10-K Annual Report Cover Page | Blaize Holdings, Inc. (CIK 1871638, ticker BZAI) filed a Form 10-K annual report for the period ended December 31, 2025. |
| SI008 | Minichart | Blaize Holdings, Inc. 2025 Annual Report: Growth Strategy, Risks and Future Outlook | Blaize reported an operating loss of $103.8 million for 2025 and flagged substantial doubt about its ability to continue as a going concern. |
| SI009 | Tracxn | SiMa.ai - 2026 Funding Rounds & List of Investors | SiMa.ai has raised funding across multiple rounds from seed through Series C, backed by investors including Maverick Capital and StepStone Group. |
| SI010 | Clay | How Much Did SiMa.ai Raise? Funding & Key Investors | SiMa.ai has raised a total of $355M across roughly nine to ten rounds, backed by about 15 investors including StepStone Group and Maverick Capital. |
| SI011 | Edge AI and Vision Alliance | Introducing Modalix SoM: Power-Efficient SoM with Rich Peripherals | The Modalix SoM is offered at $349 for the 8GB variant and $599 for the 32GB variant in 1,000-unit quantities, delivering up to 50 TOPS under 10W. |
| SI012 | ThinkRobotics | SiMa.ai Modalix DevKit 3.0 - Development Kit for MLSoC Modalix | The SiMa.ai Modalix DevKit 3.0 is listed for sale through the distributor channel for edge-AI development and benchmarking. |
| SI013 | Analytics India Magazine | How Can SiMa.ai's New Chip Transform On-Device AI? | SiMa.ai's Modalix runs reasoning-based LLMs on-device in under 10 watts, with modules priced from $349 and a developer kit for prototyping. |
| SI014 | CoinUnited | SiMa.ai Closes $85M Series C with Micron Ventures Backing | SiMa.ai's $85M Series C drew Micron Ventures backing, a strategic signal for edge AI and the semiconductor supply chain. |
| SI015 | Barchart | Micron Is Investing in SiMa.ai - What Does That Mean for MU Stock? | Micron is investing in SiMa.ai, a strategic tie-up around edge-AI memory and compute separate from SiMa.ai's headline venture round. |
| SI016 | Data Center Dynamics | Edge compute chip firm SiMa.ai raises $70m in funding round | SiMa.ai raised $70 million in a round led by Maverick Capital, extending its total funding before the 2025 Series C. |
| SI017 | Business Standard | SiMa.ai lands $85 million for AI in robots and autonomous cars | SiMa.ai raised $85 million in a Series C led by Maverick Capital, reportedly valuing the startup at about $1.4 billion. |
| SI018 | Yahoo Finance | Ambarella, Inc. (AMBA) Q4 2026 Results | Ambarella ended fiscal 2026 with about $312.6 million in cash and marketable securities after record annual revenue of $390.7 million. |
| SI019 | GreenStockNews | Blaize Announces Fourth Quarter and Full Year 2025 Financial Results | Blaize reported full-year 2025 revenue with a substantial operating loss as it scaled its edge-AI hardware and software business. |
| SI020 | Business Wire | Blaize Announces Fourth Quarter and Full Year 2025 Financial Results | Blaize reported full-year 2025 financial results detailing revenue, operating losses, and its cash position as it pursued growth in edge AI. |
| SI021 | StockAnalysis | Blaize Holdings (BZAI) Revenue | Blaize reported 2025 revenue of about $38.6 million, up sharply from $1.6 million in 2024, its first full commercial year. |
| SI022 | Sahm Capital | How Investors Are Reacting to Ambarella's Record Edge-AI Revenue and Evolving Risk Profile | Despite record edge-AI revenue, Ambarella's gross margin slipped and it remained GAAP net-loss making, underscoring the category's margin pressure. |
| SI023 | Megadose | The Information - SiMa.ai valuation and edge-AI funding note | SiMa.ai's roughly $1.4 billion valuation was reported by The Information but not confirmed by the company, which discloses no revenue. |
| SI024 | TechCrunch | At least 36 new tech unicorns were minted in 2025 so far | A wave of new tech unicorns was minted in 2025, reflecting continued investor appetite for AI and semiconductor startups. |
| SI025 | Macnica | SiMa.ai products (distributor page) | Macnica distributes SiMa.ai's MLSoC and Modalix products, providing a channel to reach industrial and robotics customers. |
| SE001 | SiMa.ai | SiMa.ai - Physical AI at the edge | SiMa.ai delivers a software-centric, purpose-built MLSoC platform for physical AI at the edge. |
| SE002 | SiMa.ai | Palette software platform | Palette is SiMa.ai's software platform for developing, optimizing, and deploying AI models on the MLSoC. |
| SE003 | SiMa.ai | SiMa.ai for robotics | SiMa.ai's MLSoC targets robotics and physical-AI applications across industrial and mobility markets. |
| SE004 | SiMa.ai | SiMa.ai product family | The SiMa.ai family spans the MLSoC silicon, modules, and developer tools for edge AI. |
| SE005 | SiMa.ai | Edgematic - SiMa.ai | Instantly evaluate model performance - throughput, latency, power - on SiMa.ai's cloud-hosted MLSoC boards with Edgematic. |
| SE006 | SiMa.ai | Empowering Your AI Vision at the Edge with Palette Edgematic Software | Palette Edgematic lets developers build and deploy edge AI applications with a low-code, visual workflow. |
| SE007 | Macnica | MLSoC Modalix - Second generation SoC for physical AI | MLSoC Modalix is built on TSMC N6 with 8x Arm Cortex-A65, a Synopsys ARC EV74 CVU, up to 200 TOPS, LPDDR5, PCIe Gen5, and hardware secure boot. |
| SE008 | Macnica | SiMa.ai products (distributor catalog) | SiMa.ai's Modalix products offer PCIe Gen5, 10GbE, MIPI CSI-2, and video codecs for edge AI integration. |
| SE009 | SiMa.ai Documentation | Edgematic - SiMa.ai System Documentation | The Edgematic documentation covers uploading and compiling a model, building applications, and the model and application catalogs. |
| SE010 | Edge AI and Vision Alliance | SiMa.ai Next-Gen Platform for Physical AI in Production | SiMa.ai's Modalix platform for physical AI is in production, with system-on-modules supporting up to 50 TOPS under 10W. |
| SE011 | Edge AI and Vision Alliance | SiMa.ai Next-Gen Platform for Physical AI - MLSoC Modalix Now in Production | MLSoC Modalix is now in production and its modules are pin- and software-compatible with popular GPU SoMs. |
| SE012 | SiMa.ai Developer Portal | For AI Agents - SiMa.ai System Documentation | The developer page is a self-contained path to install sima-cli, install Palette Neat, connect a Modalix DevKit, compile a model, and run GenAI applications. |
| SE013 | GitHub | SiMa-ai/models (model zoo) | A large set of models across PyTorch and ONNX are supported on the SiMa.ai platform as part of Palette, with compilation and conversion scripts. |
| SE014 | GitHub | SiMa.ai Palette Neat organization | SiMa.ai maintains public GitHub organizations for its Palette Neat developer toolkit and assets. |
| SE015 | Amazon Web Services | Accelerate edge AI development with SiMa.ai Edgematic with a seamless AWS integration | With Amazon SageMaker AI and SiMa.ai's Palette Edgematic platform, you can build, train, and deploy optimized ML models at the edge on SiMa's MLSoC. |
| SE016 | Synopsys | Synopsys and SiMa.ai Announce Strategic Collaboration for Automotive Edge AI | Synopsys and SiMa.ai announced a strategic collaboration to accelerate development of automotive edge-AI solutions. |
| SE017 | L&T Technology Services | LTTS and SiMa.ai collaborate on product innovation across mobility, healthcare, industrial automation and robotics | L&T Technology Services and SiMa.ai are collaborating to deliver physical-AI solutions across mobility, healthcare, industrial automation, and robotics. |
| SE018 | EE Times | SiMa.ai's Second-Gen Edge AI Chip Goes Multi-Modal | SiMa.ai's second-generation Modalix MLSoC targets multi-modal generative AI at the edge under a low power envelope. |
| SE019 | SiliconANGLE | SiMa.ai launches next-gen system-on-chip for physical AI in production | SiMa.ai launched its next-generation Modalix MLSoC into production, emphasizing a software-centric approach to edge AI. |
| SE020 | Data Center Dynamics | SiMa.ai launches MLSoC Modalix | SiMa.ai launched the MLSoC Modalix, a multi-modal edge-AI platform scaling to 200 TOPS with a low-power design. |
| SE021 | Data Center Dynamics | SiMa.ai launches MLSoC Modalix, a multi-modal platform for Edge AI | The MLSoC Modalix is a multi-modal platform supporting CNNs, Transformers, and generative AI for the edge. |
| SE022 | The Top Voices | SiMa.ai Unveils Modalix MLSoC for On-Device Physical AI and GenAI | Modalix integrates eight Arm Cortex-A65 cores and a Synopsys ARC vision unit for on-device physical AI and generative AI. |
| SE023 | PR Newswire | SiMa.ai Next-Gen Platform for Physical AI in Production | SiMa.ai announced its next-generation Modalix platform for physical AI is in production. |
| SE024 | IOT Insider | SiMa.ai unveils new product family for Edge AI | SiMa.ai unveiled the MLSoC Modalix family, claiming more than 10x the performance per watt of alternatives. |
| SE025 | TMCnet | SiMa.ai Expands ONE Platform for Edge AI with MLSoC Modalix | SiMa.ai announced MLSoC Modalix, the industry's first multi-modal edge-AI product family, delivering more than 10x the performance per watt of alternatives. |
| SU001 | STIGA S.p.A. | SiMa.ai and STIGA Announce Strategic Partnership in Physical AI | STIGA and SiMa Technologies today announced a strategic partnership to bring AI-powered solutions to robotic lawn mowers using SiMa's low-power MLSoC platform. |
| SU002 | The AI Insider | SiMa.ai and STIGA S.p.A. Partner for AI-Powered Autonomous Robotic Lawn Mowers | STIGA S.p.A. has entered a strategic partnership with SiMa Technologies to integrate low-power Physical AI platforms into its next generation of robotic lawn mowers. |
| SU003 | ResearchMingle | STIGA & SiMa.ai Advance Robotic Lawn Mowers | STIGA and SiMa.ai are teaming up to advance autonomous robotic lawn mowers with edge AI. |
| SU004 | SiMa.ai | Autonomous Mobile Robots (AMR) Running on SiMa Modalix - Solution Brief | SiMa.ai's Modalix platform delivers an integrated edge-native architecture optimized for robotics workloads, enabling reliable autonomy with real-time perception, localization, and control on-device. |
| SU005 | PR Newswire | SiMa.ai Wins Edge AI + Vision Alliance 2026 Product of the Year for Modalix SoM | SiMa.ai was named winner of the "Best Edge AI Board" by the Edge AI + Vision Alliance's 2026 Product of the Year Awards. |
| SU006 | Entrepreneur News Network | SiMa.ai Secures $85 Million to Accelerate Global Growth and Physical AI Innovation | SiMa.ai will invest its Series C proceeds to accelerate go-to-market and global reach across Korea, Japan, Europe, and the US. |
| SU007 | AInvest | Chip Startup SiMa.ai Secures $85 Mn Funding to Power AI Robots and Autonomous Cars | SiMa.ai raised $85 million to power AI robots and autonomous cars, targeting robotics and automotive customers. |
| SU008 | News by Wire | SiMa.ai and STIGA S.p.A. Announce Strategic Partnership in Physical AI | The collaboration unlocks real-time, efficient AI solutions for STIGA's robotic lawn mowers on SiMa's platform. |
| SU009 | DatacenterDynamics | Chip startup SiMa is making waves at the low-power AI Edge | While not all of SiMa's customers have been made public, earlier this month it announced a partnership with manufacturing organization TRUMPF, which will use the chips to develop AI-powered lasers. |
| SU010 | AInvest | SiMa.ai's Modalix Solves the Edge AI Power Bottleneck - Enabling Cloud-Free Robotics at Scale | Modalix is built for the medium-performance 10-to-50 TOPS edge layer designed for mass deployment of physical AI in factories, warehouses, and cities. |
| SU011 | BriefGlance | SiMa.ai Wins Top Award, Challenges Edge AI Giants with Low-Power Chip | SiMa.ai's award-winning Modalix positions it as a low-power challenger to edge AI giants like NVIDIA. |
| SU012 | SiMa.ai | SiMa.ai in Robotics | SiMa.ai enables intelligent robots - from autonomous mobile platforms to humanoid systems - with real-time on-device AI inference under 10W. |
| SU013 | L&T Technology Services | LTTS and SiMa.ai Collaborate on Product Innovation across Mobility, Healthcare, Industrial Automation and Robotics | LTTS announced a strategic partnership with SiMa.ai integrating LTTS engineering expertise with SiMa.ai's MLSoC ONE platform across IVI, AD/ADAS, industrial automation, robotics, and healthcare. |
| SU014 | Robotics Business News | L&T Technology Services Partners with SiMa.ai to Drive Next-Gen AI Solutions | The collaboration combines LTTS engineering expertise with SiMa.ai's MLSoC ONE platform to accelerate AI adoption worldwide. |
| SU015 | Synopsys | Synopsys and SiMa.ai Announce Strategic Collaboration to Accelerate Automotive Edge AI Solutions | Synopsys and SiMa.ai announced a strategic collaboration to accelerate development of automotive edge AI solutions. |
| SU016 | NextGen Defense | Defense Disruptors - SiMa.ai | Rangasayee saw that AI systems were not designed to run where increasingly needed - in real-time, decentralized, and often disconnected environments from autonomous drones to battlefield systems. |
| SU017 | Amazon Web Services | Accelerate Edge AI Development with SiMa.ai Edgematic and a Seamless AWS Integration | SiMa.ai's Edgematic integrates with AWS to accelerate edge AI application development for customers. |
| SU018 | Macnica | SiMa.ai Products - Macnica Semiconductor Distribution | Macnica distributes SiMa.ai's MLSoC and Modalix products across its semiconductor channel. |
| SU019 | SiMa.ai | SiMa.ai Raises $85M to Scale Physical AI, Bringing Total Funding to $355M | SiMa.ai serves customers across robotics, automotive, industrial automation, aerospace and defense, smart vision, and healthcare. |
| SU020 | SiliconANGLE | SiMa.ai Launches Next-Gen System-on-Chip for Physical AI in Production | SiMa.ai's Modalix moved to production, targeting customers in robotics, automotive, and industrial markets. |
| SU021 | IoT Insider | SiMa.ai Unveils New Product Family for Edge AI | SiMa.ai unveiled its Modalix product family for edge AI aimed at multiple industrial and robotics customers. |
| SU022 | SiMa.ai | Our Family - SiMa.ai | SiMa.ai describes its ecosystem of partners and the markets it serves for physical AI at the edge. |
| SU023 | CEO Insider | Interview with Krishna Rangasayee, CEO of SiMa.ai | Rangasayee describes SiMa.ai's focus on serving embedded-edge customers across multiple physical-AI verticals. |
| SU024 | Unite.AI | Krishna Rangasayee, Founder & CEO of SiMa.ai - Interview Series | Rangasayee explains SiMa.ai's full-stack approach designed to make edge AI easier for customers to deploy. |
| SU025 | ThinkRobotics | SiMa.ai Modalix DevKit 3.0 | ThinkRobotics resells the SiMa.ai Modalix DevKit 3.0 for edge AI developers and customers. |
| SR001 | Bureau of Industry and Security | Commerce Strengthens Restrictions on Advanced Computing Semiconductors and Enhances Foundry Due Diligence | BIS released rules that update export controls on advanced computing semiconductors and add entities in the PRC to the Entity List to protect US national security. |
| SR002 | US Government Accountability Office | Export Controls - Commerce Implemented Advanced Semiconductor Rules | To protect advanced chip technology from foreign military use, the Department of Commerce issued rules in 2022 and 2023 to control its export and the equipment used to make it. |
| SR003 | Finnegan | BIS's New 2026 License Review Process for AI Chips | The 2026 BIS process requires exporters to certify sufficient US supply, cap China/Macau shipments, meet KYC and reporting obligations, and satisfy third-party testing before approval. |
| SR004 | Mayer Brown | Administration Policies on Advanced AI Chips Codified | The codified policy introduces case-by-case licensing, tariffs on covered chips, and strict end-use and destination checks for advanced AI semiconductors. |
| SR005 | Semiconductor Engineering | Achieving An ASIL-C Safety Architecture | Achieving ASIL-C requires lock-step CPUs, ECC, built-in self-test, FMEDA, and traceable safety architecture in the semiconductor design. |
| SR006 | Leadvent Group | Understanding ISO 26262 - Key Updates and Implementation Challenges | ISO 26262's upcoming third edition adds AI-specific guidance aligned with ISO/PAS 8800 and SOTIF, raising the bar for AI-driven automotive electronics. |
| SR007 | SiMa.ai | SiMa.ai Privacy Policy | This Privacy Policy describes how SiMa Technologies, Inc. collects, uses, processes, and shares personal information from users of its website and services. |
| SR008 | Baker McKenzie | BIS Revises License Review Policy for Advanced Computing Commodities and AI Semiconductors | BIS revised its license-review policy to case-by-case review for advanced computing commodities and AI semiconductors exported to China and Macau under strict conditions. |
| SR009 | SaaS Sentinel | AI Chip Startup Hailo Plans SPAC Merger After Valuation Falls 50% to Under $500M | Edge-AI chip startup Hailo plans a SPAC merger after its valuation fell roughly 50% to under $500 million, following layoffs. |
| SR010 | CNBC | Nvidia AI Chip Rivals Chase Funding as Startups Race to Challenge Its Dominance | NVIDIA's dominance in AI compute leaves rivals racing for funding and a defensible niche against its entrenched hardware and software ecosystem. |
| SR011 | GreenStockNews | Blaize Announces Fourth Quarter and Full Year 2025 Financial Results | Blaize reported full-year 2025 revenue of roughly $38.6 million against a large operating loss and disclosed a going-concern warning. |
| SR012 | StockTitan | Blaize Holdings Inc Files Annual Report (10-K) | Blaize's 10-K discloses substantial operating losses and going-concern risk factors typical of pre-scale edge-AI semiconductor companies. |
| SR013 | StockTitan | Ambarella Announces Fourth Quarter and Fiscal Year 2026 Financial Results | Ambarella reported record fiscal-2026 revenue of about $390.7 million yet still posted a GAAP net loss of roughly $75.9 million. |
| SR014 | SiMa.ai | SiMa.ai Raises $85M to Scale Physical AI, Bringing Total Funding to $355M | SiMa.ai raised $85 million in an oversubscribed Series C led by Maverick Capital, bringing total funding to $355 million. |
| SR015 | DatacenterDynamics | Chip startup SiMa is making waves at the low-power AI Edge | SiMa's second-generation chip is based on TSMC's 6nm process, and not all of the company's customers have been made public. |
| SR016 | EE Times | SiMa.ai's Second-Gen Edge AI Chip Goes Multi-Modal | SiMa.ai's first-generation MLSoC appeared on MLPerf, but its second-generation Modalix performance figures are largely company-reported. |
| SR017 | SiMa.ai | Krishna Rangasayee CEO Biography | Krishna Rangasayee is founder and CEO of SiMa.ai, previously a senior executive at Xilinx and COO at Groq. |
| SR018 | The Information (via Megadose) | SiMa.ai Valuation and Funding Context | SiMa.ai's Series C reportedly valued the company at about $1.4 billion, a figure the company has not officially confirmed. |
| SR019 | Tracxn | Hailo - Company Funding and Investors | Hailo raised over $340 million but faced a sharp valuation decline amid edge-AI market pressure. |
| SR020 | NVIDIA | Jetson Modules | NVIDIA's Jetson modules span a broad performance range backed by the CUDA and JetPack software ecosystem for edge AI. |
| SR021 | SEC (EDGAR) | Ambarella Inc Annual Report Filing Index | Ambarella's SEC filing details revenue, losses, and risk factors for a public fabless edge-AI semiconductor peer. |
| SR022 | Barchart | Micron Is Investing in SiMa.ai - What Does That Mean for MU Stock | Micron made a strategic investment in SiMa.ai, signaling memory-maker interest in edge-AI silicon. |
| SR023 | Business Standard | SiMa.ai Lands $85 Million for AI Robots and Autonomous Cars | SiMa.ai landed $85 million to fund AI chips for robots and autonomous cars. |
| SR024 | Businesswire | EdgeCortix Closes Oversubscribed Series B Bringing Total Funding Over $110M | EdgeCortix closed an oversubscribed Series B bringing total funding over $110 million amid growing edge-AI demand. |
| SR025 | Sahm Capital | How Investors Are Reacting to Ambarella's Record Edge AI Revenue and Evolving Risk Profile | Investors weigh Ambarella's record edge-AI revenue against its evolving risk profile and continued losses. |
| SR026 | DatacenterDynamics | Hailo Releases Hailo-10H Edge AI Chip | Hailo released its Hailo-10H edge AI chip as competition in the edge-AI silicon market intensifies. |
| SR027 | Calcalist | Hailo AI Chip Company Coverage | Hailo faced a difficult financing environment and restructuring as edge-AI valuations came under pressure. |
| SR028 | Last10K | Blaize Holdings SEC Filings | Blaize's SEC filings disclose operating losses and going-concern considerations for the edge-AI chipmaker. |
| SR029 | Tracxn | SiMa.ai - Funding and Investors | SiMa.ai raised across roughly nine to ten rounds backed by about fifteen investors, per third-party tracking. |
| SR030 | EE Times | Hailo Debuts Edge GenAI Chip, Raises $120 Million | Hailo debuted an edge GenAI chip and raised $120 million, underscoring intense competition in edge-AI silicon. |
| SV001 | Premier Alternatives | SiMa.ai Valuation 2026: USD960.0M | Private Company Worth | SiMa.ai is currently valued at USD960.0M with total funding raised of USD355.3M and a capital efficiency of 2.70x valuation to funding. |
| SV002 | Caplight | SiMa.ai | Valuation, Funding Rounds & Stock Price | Caplight | Caplight lists SiMa.ai comps including OpenAI (~40x revenue) and Cohere, framing SiMa against public and private peers by sector, stage, and business model. |
| SV003 | Parsers VC | SiMa.ai - Funding, Valuation, Investors, News | Parsers VC lists SiMa.ai's April 2026 strategic investment from Micron and a series of 2026 partnership and product announcements. |
| SV004 | NXP Semiconductors | NXP to Acquire Kinara to Expand Edge AI Portfolio | NXP's acquisition of Kinara enhances its processing portfolio with cutting-edge NPUs and AI software, establishing a scalable platform for AI-powered edge systems. |
| SV005 | SiliconReport | Onsemi to Acquire Synaptics for Nearly USD7 Billion in All-Stock Edge AI Deal | Onsemi agreed to acquire Synaptics in a nearly USD7 billion all-stock deal to expand in the edge AI market, reflecting a premium for AI-driven growth prospects. |
| SV006 | Aventis Advisors | AI Valuation Multiples in 2025 | AI valuation multiples vary widely by niche and stage, with some categories trading at 70x revenue while others struggle to justify double digits. |
| SV007 | Finro Financial Consulting | M&A in AI: 2025 Valuation Multiples and Key Trends | Median revenue multiples for private AI M&A transactions in 2025 cluster in the 25x to 30x range, higher than most traditional tech exits. |
| SV008 | ARC Group | Tech M&A Outlook 2025: AI, Chips, and Hardware | Strategic semiconductor consolidation and hardware-plus-AI integration are driving renewed dealmaking optimism entering 2025. |
| SV009 | Finro Financial Consulting | AI Startup Valuations in 2025: Benchmarks Across 400+ Companies | Analysis of over 400 AI companies across 15 categories shows valuation multiples ranging from double digits to 70x depending on niche, round, and status. |
| SV010 | Megadose (citing The Information) | SiMa.ai Reported Valuation and Series C Coverage | The reported roughly USD1.4 billion post-money valuation traces to The Information and is not confirmed by SiMa.ai. |
| SV011 | Tracxn | SiMa.ai - 2026 Funding Rounds & List of Investors | Tracxn records SiMa.ai's funding rounds and investor list, including Maverick Capital, StepStone Group, Fidelity, and Point72. |
| SV012 | SiMa.ai | SiMa.ai Raises USD85M to Scale Physical AI, Bringing Total Funding to USD355M | SiMa.ai raised USD85 million in Series C funding led by Maverick Capital, bringing total funding to USD355 million to scale Physical AI. |
| SV013 | PR Newswire | SiMa.ai Raises USD85M to Scale Physical AI, Bringing Total Funding to USD355M | The Series C round was led by Maverick Capital with participation from StepStone Group and Micron Ventures and was oversubscribed. |
| SV014 | StockTitan | Ambarella Inc. Announces Fourth Quarter and Fiscal Year 2026 Results | Ambarella reported fiscal 2026 revenue of USD390.7 million, up 37.2 percent, with a GAAP net loss of USD75.9 million and non-GAAP profitability. |
| SV015 | US Securities and Exchange Commission | Ambarella Inc. EDGAR Filing Index | Ambarella's SEC filing index on EDGAR provides the audited financial disclosures underpinning its reported fiscal 2026 results. |
| SV016 | StockTitan | Blaize Holdings Inc. Files Annual Report (10-K) | Blaize's 10-K discloses an operating loss near USD103.8 million and a going-concern warning against modest revenue. |
| SV017 | Last10K | Blaize Holdings (BZAI) SEC Filings | Last10K aggregates Blaize's SEC filings detailing its financial results, risk factors, and going-concern disclosure. |
| SV018 | Business Wire | Blaize Announces Fourth Quarter and Full Year 2025 Financial Results | Blaize reported full-year 2025 revenue near USD38.6 million with a substantial operating loss, underscoring edge-AI silicon loss-making economics. |
| SV019 | Green Stock News | Blaize Announces Fourth Quarter and Full Year 2025 Financial Results | Coverage of Blaize's 2025 results details its revenue, widening losses, and strategic outlook as a public edge-AI chip company. |
| SV020 | Sahm Capital | How Investors Are Reacting to Ambarella's Record Edge AI Revenue Surge | Investors reacted to Ambarella's record edge-AI revenue surge and evolving risk profile, reflected in its market capitalization and valuation multiples. |
| SV021 | SaaS Sentinel | AI Chip Startup Hailo Plans SPAC Merger After Valuation Falls 50% to Under USD500M | Hailo's valuation reportedly fell roughly 50 percent to under USD500 million as it pursued a distressed SPAC merger following layoffs. |
| SV022 | CNBC | NVIDIA AI Chip Rivals Attract Funding as Edge and Inference Startups Emerge | NVIDIA's AI chip rivals continue to attract funding, but the incumbent's ecosystem dominance remains the central challenge for challengers. |
| SV023 | TechCrunch | At Least 36 New Tech Unicorns Were Minted in 2025 So Far | Dozens of new tech unicorns were minted through 2025, reflecting continued late-stage venture appetite for AI-adjacent companies. |
| SV024 | DatacenterDynamics | SiMa - The Chip Semiconductor Firm Betting on Low-Power AI at the Edge | DatacenterDynamics analyzes SiMa's low-power edge-AI positioning and its differentiation against cloud-GPU economics. |
| SV025 | Clay | SiMa.ai Funding Dossier | Clay's dossier compiles SiMa.ai's funding history, round sequence, and investor roster. |
| SV026 | Barchart | Micron Is Investing in SiMa.ai - What Does That Mean for MU Stock? | Micron's strategic investment in SiMa.ai signals memory-and-edge-systems alignment with the physical-AI thesis. |
| SV027 | CoinUnited | SiMa.ai Closes USD85M Series C with Micron Ventures Backing | Coverage frames SiMa.ai's USD85 million Series C with Micron Ventures backing and its signal for edge-AI and semiconductor stocks. |
| SV028 | Business Standard | SiMa.ai Lands USD85 Million for AI Robots, Autonomous Cars | SiMa.ai raised USD85 million to target AI robots and autonomous cars, extending its Physical AI go-to-market. |
| SV029 | Business Wire | EdgeCortix Closes Oversubscribed Series B, Bringing Total Funding Over USD110M | EdgeCortix closed an oversubscribed Series B bringing total funding over USD110 million amid growing edge-AI demand. |
| SV030 | Tracxn | Hailo - 2026 Funding Rounds & Investors | Tracxn records Hailo's funding history exceeding USD340 million across multiple rounds as a leading edge-AI private peer. |