DEEPX
一家边缘 AI 半导体初创公司,已有扎实商业验证,但估值仍跑在公开披露质量前面。
DEEPX 已进入真实 physical-AI 商业化阶段,但当前私募估值仍跑在公开披露质量前面,尤其是利润率、客户耐久性和融资确定性。
封面要素
公司概况
DEEPX 是一家韩国边缘 AI 半导体公司,2018 年成立,总部位于 Seongnam-si 的 Pangyo。公司面向机器人、摄像头、工业计算机和其他物理 AI 终端,打造低功耗商用 AI 加速器、模组和支撑部署的软件。DX-M1 已进入量产,伙伴栈正快速扩张,Series D 首关据报把估值推到 $2.2 billion。商业叙事可信,但公开的财务、治理和客户持久性披露仍跟不上价格。
- 官网
- deepx.ai
- 成立时间
- 2018-01-01
- 创始人
- Lokwon Kim
- 创立地点
- Seongnam-si, Gyeonggi-do, South Korea
- 总部
- Pangyo, Seongnam-si, Gyeonggi-do, South Korea
- 产品
- DX-M1、DX-M1M、DX-H1 产品和 DXNN 软件栈,瞄准机器人、智能摄像头、工业系统和边缘服务器中的低功耗推理。
- 客户
- 机器人 OEM、工业与嵌入式计算厂商、智能摄像头和 VMS 厂商、分销商以及系统集成商。
- 商业模式
- 作为 Fabless 半导体公司销售芯片、模组和板卡,并用软件工具和伙伴赋能把设计导入转成量产部署。
- 阶段
- series-d
- 融资情况
- Series D 首关已于 2026 年 8 月完成,据报估值 $2.2 billion;更大一轮融资仍在进行,公开资料对最终规模和累计融资额说法不一。
执行摘要
主要优势
- DX-M1 已经向更宽的伙伴和渠道生态出货,而不是停在实验室产品。
- DEEPX 卡在真实 physical-AI 楔口上,低功耗本地推理能解决具体问题。
- 公开定价、订单、验证和具名伙伴,让商业化故事比平均水平具体得多。
- 公司在搭硅片、软件和生态层,不只是讲一颗芯片。
主要风险
- 若缺少更好的公开利润率和客户质量披露,$2.2 billion 首关估值很难完全支撑。
- Samsung 集中度和 DX-M2 路线图野心,带来真实执行风险。
- 融资仍未关闭、且据称 pre-IPO 轮缩量,说明后期投资人已经有疑虑。
- 即便 DEEPX 在功耗和形态上胜出,既有生态仍是结构性竞争威胁。
未决问题
- 毛利率、现金跑道和烧钱的审计级或招股书级财务披露。
- 按客户和渠道拆分的复购、留存和集中度指标。
- 单一权威融资历史和当前轮目标。
- 更广路线图的独立 benchmark 和长周期可靠性证据。
目录
01公司概览
1.1 身份、布局和商业模式
DEEPX 是一家韩国边缘 AI 半导体公司,商业化势头真实,但公开披露质量仍跟不上 Series D 首关对应的 $2.2 billion 估值。本节在公司概览框架下,聚焦身份、布局和商业模式。公开资料反复指向三个事实:DEEPX 是总部位于 Pangyo 的 Fabless 边缘 AI 半导体供应商;公司销售芯片、模组和支撑部署的软件;目标终端包括机器人、摄像头、工业系统和边缘服务器。判断 DEEPX 时,应把「总部位于 Pangyo 的 Fabless 边缘 AI 半导体供应商」放回物理 AI 商业化语境,而不是抽象 AI 炒作。公司销售芯片、模组和部署软件,也同样应按物理 AI 商业化来理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CO001, CO002, CO003, CO004, CO005, CO006]
| 信号 | 公开信息解读 | 启示 | 来源质量 |
|---|---|---|---|
| 身份 | 2018 年成立,总部位于 Pangyo, Seongnam-si。 | 公司概览启示 | 中 |
| 身份 | 处于 Series D 阶段的私营公司,聚焦面向物理设备的边缘 AI 半导体。 | 公司概览启示 | 中 |
| 管理层 | 官方与独立来源都指向 Lokwon Kim 为创始人兼 CEO。 | 公司概览启示 | 中 |
| 产品 | DX-M1 是当前旗舰,也是商业活动中心。 | 公司概览启示 | 中 |
| 融资 | 2026 年 8 月 Series D 首次交割约 $30M,估值 $2.2B。 | 公司概览启示 | 中 |
根据公司概览章节中信号最强的公开主张生成。
[CO001, CO002, CO003, CO004, CO005]1.2 创始人、管理层和治理
DEEPX 是一家韩国边缘 AI 半导体公司,商业化势头真实,但公开披露质量仍跟不上 Series D 首关对应的 $2.2 billion 估值。本节在公司概览框架下,聚焦创始人、管理层和治理。公开资料反复指向三个事实:Lokwon Kim 一直是创始人兼 CEO 的稳定参照;具名高管让管理梯队更宽,但董事会披露并未同步打开;2026 年人员流动报道提高了执行敏感度。判断 DEEPX 时,Lokwon Kim 这个稳定的创始人兼 CEO 参照,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。具名高管拓宽了管理层,却没有补足董事会披露,这一点也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CO009, CO010, CO011, CO012, CO013, CO014]
| 信号 | 公开信息解读 | 启示 | 来源质量 |
|---|---|---|---|
| 融资 | 该轮融资仍未关闭,不同媒体对整轮目标说法不一。 | 公司概览启示 | 中 |
| 数据冲突 | PitchBook 与 Tracxn 在员工数和累计融资额上不一致。 | 公司概览启示 | 中 |
| 进展 | EE Times 报道称 7 个月内达成 27 笔交易。 | 公司概览启示 | 中 |
| 进展 | briefs 和 Electronics Weekly 报道称,8 个国家签下 30 多项协议。 | 公司概览启示 | 中 |
| 收入 | The Elec 给出 2026 年收入目标:KRW 60B,上一年基数为 KRW 3.3B。 | 公司概览启示 | 中 |
根据公司概览章节中信号最强的公开主张生成。
[CO006, CO007, CO008, CO009, CO010]1.3 融资、规模和资本结构
DEEPX 是一家韩国边缘 AI 半导体公司,商业化势头真实,但公开披露质量仍跟不上 Series D 首关对应的 $2.2 billion 估值。本节在公司概览框架下,聚焦融资、规模和资本结构。公开资料反复指向三个事实:Series D 首关把当前估值定在 $2.2B;订单和验证暗示势头真实;累计融资额和股权结构条款仍然噪声较多。判断 DEEPX 时,Series D 首关给出的 $2.2B 当前估值,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。订单和验证所暗示的真实势头,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CO017, CO018, CO019, CO020, CO021, CO022]
| 信号 | 公开信息解读 | 启示 | 来源质量 |
|---|---|---|---|
| 负面 | Sedaily 报道称,估值受到质疑,pre-IPO 目标被下调。 | 公司概览启示 | 中 |
| 布局 | 官方联系页面显示其在韩国、美国、台湾和中国设有据点。 | 公司概览启示 | 中 |
| 信任 | 信任中心材料重点强调专利、长寿命和质量保障。 | 公司概览启示 | 中 |
| GTM | AAEON、Hyundai、Avnet 及合作伙伴公告显示,公司靠渠道放大规模。 | 公司概览启示 | 中 |
| 路线图 | DX-M1 已经得到验证;DX-M2 仍是可选上行。 | 公司概览启示 | 中 |
根据公司概览章节中信号最强的公开主张生成。
[CO011, CO012, CO013, CO014, CO015]1.4 里程碑、地域和负面信号
DEEPX 是一家韩国边缘 AI 半导体公司,商业化势头真实,但公开披露质量仍跟不上 Series D 首关对应的 $2.2 billion 估值。本节在公司概览框架下,聚焦里程碑、地域和负面信号。公开资料反复指向三个事实:DX-M1 在 2025 年底进入量产;2026 年带来 OEM、渠道和机器人领域公告;IPO 前收缩是最清晰的负面信号。判断 DEEPX 时,DX-M1 于 2025 年底量产,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。2026 年的 OEM、渠道和机器人公告,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CO025, CO026, CO027, CO028, CO029, CO030]
| 信号 | 公开信息解读 | 启示 | 来源质量 |
|---|---|---|---|
| 综合 | 公司概览证据最强的地方,是公开来源与官方产品或合作伙伴材料共同指向:DEEPX 是位于 Pangyo 的无晶圆厂边缘 AI 半导体厂商。 | 公司概览启示 | 中 |
| 综合 | 公司概览证据最强的地方,是公开来源与官方产品或合作伙伴材料共同指向:公司销售芯片、模组和支持部署的软件。 | 公司概览启示 | 中 |
| 综合 | 公司概览证据最强的地方,是公开来源与官方产品或合作伙伴材料共同指向:目标场景包括机器人、摄像头、工业系统和边缘服务器。 | 公司概览启示 | 中 |
| 综合 | 公司概览证据最强的地方,是公开来源与官方产品或合作伙伴材料共同指向:Lokwon Kim 一直被指认为创始人兼 CEO。 | 公司概览启示 | 中 |
| 综合 | 公司概览证据最强的地方,是公开来源与官方产品或合作伙伴材料共同指向:具名高管拓宽了管理层阵容,但董事会披露仍不足。 | 公司概览启示 | 中 |
根据公司概览章节中信号最强的公开主张生成。
[CO016, CO017, CO018, CO019, CO020]1.5 图表
02市场分析
2.1 市场边界和纳入的工作负载
DEEPX 落在物理 AI 边缘半导体市场:在这里,功耗、散热、延迟和本地推理比超大规模训练吞吐更重要。本节在市场分析框架下,聚焦市场边界和纳入的工作负载。公开资料反复指向三个事实:公司做的是商用边缘推理,而不是云端训练;重点场景是机器人、智能摄像头、网关和工业计算;宽泛的云 AI 支出不应计入实际切入点。判断 DEEPX 时,商用边缘推理而非云端训练,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。机器人、智能摄像头、网关和工业计算这些重点,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CM001, CM002, CM003, CM004, CM005, CM006]
| 信号 | 公开信息解读 | 启示 | 来源质量 |
|---|---|---|---|
| 边界 | DEEPX 属于边缘 AI 半导体市场,不属于训练 GPU 市场。 | 市场分析启示 | 中 |
| TAM | Fortune 测算 2026 年边缘 AI 半导体市场约 $29.85B。 | 市场分析启示 | 中 |
| TAM | ABI 测算 2026 年边缘 AI 芯片组市场为 $34.4B。 | 市场分析启示 | 中 |
| TAM | Future Markets 预计,到 2036 年边缘 AI 芯片市场将超过 $80B。 | 市场分析启示 | 中 |
| TAM | Axis 估算 2026 年边缘 AI 机会接近 $30B。 | 市场分析启示 | 中 |
根据市场分析章节中信号最强的公开主张生成。
[CM001, CM002, CM003, CM004, CM005]2.2 规模测算口径和细分结构
DEEPX 落在物理 AI 边缘半导体市场:在这里,功耗、散热、延迟和本地推理比超大规模训练吞吐更重要。本节在市场分析框架下,聚焦规模测算口径和细分结构。公开资料反复指向三个事实:公开估算大多落在 2026 年约 $30B+ 的机会附近;半导体、芯片组和更宽泛边缘 AI 的定义并不一致;工业和机器人具备更高价值密度。判断 DEEPX 时,公开估算所指向的 2026 年约 $30B+ 机会,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。半导体、芯片组和更宽泛边缘 AI 之间的定义差异,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CM009, CM010, CM011, CM012, CM013, CM014]
| 信号 | 公开信息解读 | 启示 | 来源质量 |
|---|---|---|---|
| 垂直场景 | 工业自动化和机器人是商用边缘推理的高价值垂直场景。 | 市场分析启示 | 中 |
| 买方 | 真正买单的是 OEM、分销商和系统集成商。 | 市场分析启示 | 中 |
| 采用 | 评估、板卡集成、验证、备货和量产定点,是常见转化阶段。 | 市场分析启示 | 中 |
| 驱动因素 | 低延迟、隐私、带宽和功耗,是反复出现的采购驱动因素。 | 市场分析启示 | 中 |
| 约束 | 先进制程成本和出口管制仍是结构性约束。 | 市场分析启示 | 中 |
根据市场分析章节中信号最强的公开主张生成。
[CM006, CM007, CM008, CM009, CM010]2.3 买方、付款方和采用路径
DEEPX 落在物理 AI 边缘半导体市场:在这里,功耗、散热、延迟和本地推理比超大规模训练吞吐更重要。本节在市场分析框架下,聚焦买方、付款方和采用路径。公开资料反复指向三个事实:OEM、分销商和集成商才是真正买方;评估套件和板级验证先于量产;渠道撬动力和规格本身一样重要。判断 DEEPX 时,OEM、分销商和集成商是真正买方,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。评估套件和板级验证先于量产,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CM017, CM018, CM019, CM020, CM021, CM022]
| 信号 | 公开信息解读 | 启示 | 来源质量 |
|---|---|---|---|
| 匹配 | DEEPX 用例页面与工业和机器人切入点高度贴合。 | 市场分析启示 | 中 |
| 地域 | 亚太重要,因为制造、电子和机器人密度在那里重叠。 | 市场分析启示 | 中 |
| 渠道 | 渠道深度对采用的影响几乎不亚于芯片规格。 | 市场分析启示 | 中 |
| SOM | 现实 SOM 窄于自上而下的 TAM 标题数字。 | 市场分析启示 | 中 |
| 工作负载 | 核心工作负载是在严格散热包络下的本地推理。 | 市场分析启示 | 中 |
根据市场分析章节中信号最强的公开主张生成。
[CM011, CM012, CM013, CM014, CM015]2.4 增长驱动和约束
DEEPX 落在物理 AI 边缘半导体市场:在这里,功耗、散热、延迟和本地推理比超大规模训练吞吐更重要。本节在市场分析框架下,聚焦增长驱动和约束。公开资料反复指向三个事实:延迟、隐私、带宽和功耗驱动需求;先进制程成本和出口管制约束供给;软件锁定影响采用节奏。判断 DEEPX 时,延迟、隐私、带宽和功耗驱动需求,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。先进制程成本和出口管制约束供给,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CM025, CM026, CM027, CM028, CM029, CM030]
| 信号 | 公开信息解读 | 启示 | 来源质量 |
|---|---|---|---|
| 综合 | 市场分析证据最强的地方,是公开来源与官方产品或合作伙伴材料共同指向:机会在商用边缘推理,而不是云端训练。 | 市场分析启示 | 中 |
| 综合 | 市场分析证据最强的地方,是公开来源与官方产品或合作伙伴材料共同指向:聚焦机器人、智能摄像头、网关和工业计算。 | 市场分析启示 | 中 |
| 综合 | 市场分析证据最强的地方,是公开来源与官方产品或合作伙伴材料共同指向:实用切入点不应纳入广义云端 AI 支出。 | 市场分析启示 | 中 |
| 综合 | 市场分析证据最强的地方,是公开来源与官方产品或合作伙伴材料共同指向:公开估算大多聚在 2026 年约 $30B+ 的机会规模附近。 | 市场分析启示 | 中 |
| 综合 | 市场分析证据最强的地方,是公开来源与官方产品或合作伙伴材料共同指向:定义在半导体、芯片组和更广义边缘 AI 之间摇摆。 | 市场分析启示 | 中 |
根据市场分析章节中信号最强的公开主张生成。
[CM016, CM017, CM018, CM019, CM020]2.5 图表
03竞争对手
3.1 直接同业、既有巨头和替代路径
DEEPX 以商用低功耗边缘 AI 的位置参与竞争,对手包括直接专门厂商、集成型既有巨头和系统级替代方案。本节在竞争对手框架下,聚焦直接同业、既有巨头和替代路径。公开资料反复指向三个事实:Hailo 是最接近的专门厂商同业;Qualcomm、NXP 和 NVIDIA 是系统级既有巨头;Rockchip 和 Coral 代表替代路径。判断 DEEPX 时,Hailo 是最接近的专门厂商同业,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。Qualcomm、NXP 和 NVIDIA 是系统级既有巨头,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CP001, CP002, CP003, CP004, CP005, CP006]
| 信号 | 公开信息解读 | 启示 | 来源质量 |
|---|---|---|---|
| 同业 | 按形态和用例看,Hailo 是最接近的直接专业同业。 | 竞争格局启示 | 中 |
| 同业 | Qualcomm 靠集成平台竞争,而不是纯离散 NPU。 | 竞争格局启示 | 中 |
| 同业 | Coral 争的是开发者可及性和小规模部署。 | 竞争格局启示 | 中 |
| 同业 | Apple 是自用型标杆,不是商用 OEM 替代品。 | 竞争格局启示 | 中 |
| 同业 | NXP 靠工业和安全要求更重的系统竞争。 | 竞争格局启示 | 中 |
根据竞争格局章节中信号最强的公开主张生成。
[CP001, CP002, CP003, CP004, CP005]3.2 能力、封装和价格对比
DEEPX 以商用低功耗边缘 AI 的位置参与竞争,对手包括直接专门厂商、集成型既有巨头和系统级替代方案。本节在竞争对手框架下,聚焦能力、封装和价格对比。公开资料反复指向三个事实:DX-M1 强调低于 5W 的商用部署;公开价格可见度好于不少同业;集成平台往往靠简单易用取胜。判断 DEEPX 时,DX-M1 强调低于 5W 的商用部署,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。公开价格可见度好于不少同业,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CP009, CP010, CP011, CP012, CP013, CP014]
| 信号 | 公开信息解读 | 启示 | 来源质量 |
|---|---|---|---|
| 生态 | DEEPX 仍要绕着 NVIDIA 这个生态标杆竞争。 | 竞争格局启示 | 中 |
| 替代 | Rockchip 是成本更低的 AIoT 替代路径。 | 竞争格局启示 | 中 |
| 区域 | Rebellions 是韩国 AI 芯片可比对象,但工作负载组合不同。 | 竞争格局启示 | 中 |
| 定价 | DEEPX 的公开芯片和模组定价异常透明。 | 竞争格局启示 | 中 |
| 取舍 | 集成 SoC 往往赢在简单,即便离散 NPU 赢在效率。 | 竞争格局启示 | 中 |
根据竞争格局章节中信号最强的公开主张生成。
[CP006, CP007, CP008, CP009, CP010]3.3 分销、工具链和生态锁定
DEEPX 以商用低功耗边缘 AI 的位置参与竞争,对手包括直接专门厂商、集成型既有巨头和系统级替代方案。本节在竞争对手框架下,聚焦分销、工具链和生态锁定。公开资料反复指向三个事实:NVIDIA 和 Qualcomm 掌握更强工具链;DEEPX 正在搭建伙伴驱动的分销和工作流桥接;生态深度是主要竞争缺口。判断 DEEPX 时,NVIDIA 和 Qualcomm 拥有更强工具链,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。DEEPX 正在搭建伙伴驱动的分销和工作流桥接,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CP017, CP018, CP019, CP020, CP021, CP022]
| 信号 | 公开信息解读 | 启示 | 来源质量 |
|---|---|---|---|
| GTM | DEEPX 要靠合作伙伴和工作流质量,才能与更大的生态竞争。 | 竞争格局启示 | 中 |
| 风险 | 没有单一同业能封住 DEEPX 的所有优势,但多家公司可以挤压这个利基。 | 竞争格局启示 | 中 |
| 定位 | 面向改造项目,商用离散芯片能形成差异点。 | 竞争格局启示 | 中 |
| 风险 | 系统级龙头可以凭渠道力量压缩芯片领先优势。 | 竞争格局启示 | 中 |
| 壁垒 | 壁垒能否持久,取决于芯片、软件和合作伙伴动能能否同时成立。 | 竞争格局启示 | 中 |
根据竞争格局章节中信号最强的公开主张生成。
[CP011, CP012, CP013, CP014, CP015]3.4 护城河持久性和竞争方可能反应
DEEPX 以商用低功耗边缘 AI 的位置参与竞争,对手包括直接专门厂商、集成型既有巨头和系统级替代方案。本节在竞争对手框架下,聚焦护城河持久性和竞争方可能反应。公开资料反复指向三个事实:商用独立性和能效是真优势;既有巨头可用软件和渠道力量压缩硅片优势;DEEPX 必须同时守住硅片、软件和伙伴。判断 DEEPX 时,商用独立性和能效是真优势,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。既有巨头可用软件和渠道力量压缩硅片优势,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CP025, CP026, CP027, CP028, CP029, CP030]
| 信号 | 公开信息解读 | 含义 | 来源质量 |
|---|---|---|---|
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:Hailo 是最接近的专注型同业,因此这条竞争证据最强。 | 竞争格局含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:Qualcomm、NXP 和 NVIDIA 是系统级既有巨头,因此这条竞争证据最强。 | 竞争格局含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:Rockchip 和 Coral 代表替代路径,因此这条竞争证据最强。 | 竞争格局含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:DX-M1 强调面向开放市场的低于 5W 部署,因此这条竞争证据最强。 | 竞争格局含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:公开定价透明度高于许多同业,因此这条竞争证据最强。 | 竞争格局含义 | 中 |
由竞争对手章节中信号最强的公开论据生成。
[CP016, CP017, CP018, CP019, CP020]3.5 图表
04财务
4.1 收入模式和定价
商业信号可信,但公开披露对毛利、烧钱、集中度和积压订单质量仍不完整。本节在财务框架下,聚焦收入模式和定价。公开资料反复指向三个事实:硬件销售锚定商业模式;芯片、模组、板卡和赋能服务构成变现阶梯;公开价格存在,但实际 ASP 仍不透明。判断 DEEPX 时,硬件销售锚定模式,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。芯片、模组、板卡和赋能服务构成变现阶梯,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CI001, CI002, CI003, CI004, CI005, CI006]
| 信号 | 公开信息解读 | 含义 | 来源质量 |
|---|---|---|---|
| 模式 | 模式由硬件牵引,不是 SaaS 牵引。 | 财务含义 | 中 |
| 定价 | The Elec 引述的芯片价格约 $50,模组价格约 $250。 | 财务含义 | 中 |
| 订单 | The Elec 报道,截至 2026 年 4 月,公司约有 30 个采购订单,金额约 $6.7M。 | 财务含义 | 中 |
| 管线 | 公司称约 350 家公司在做概念验证。 | 财务含义 | 中 |
| 订单 | KM Journal 报道,8 个月内公司在 8 个国家拿到 48 个采购订单。 | 财务含义 | 中 |
由财务章节中信号最强的公开论据生成。
[CI001, CI002, CI003, CI004, CI005]4.2 商业进展和公开指标
商业信号可信,但公开披露对毛利、烧钱、集中度和积压订单质量仍不完整。本节在财务框架下,聚焦商业进展和公开指标。公开资料反复指向三个事实:订单、验证和收入指引在 2026 年均有改善;以上一次披露的年度基线看,历史亏损仍然显著;商业验证真实,但尚未完全经过审计。判断 DEEPX 时,订单、验证和收入指引在 2026 年均有改善,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。以上一次披露的年度基线看,历史亏损仍然显著,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CI009, CI010, CI011, CI012, CI013, CI014]
| 信号 | 公开信息解读 | 含义 | 来源质量 |
|---|---|---|---|
| 地域 | KM Journal 称,约 67% 销售来自海外。 | 财务含义 | 中 |
| 收入 | 2026 年收入目标上调至 KRW 60B。 | 财务含义 | 中 |
| 亏损 | 按最新公开年度基线看,历史亏损仍然显著。 | 财务含义 | 中 |
| 资本 | 待完成融资和后续 IPO 时间点仍会左右资本计划。 | 财务含义 | 中 |
| 路线图 | DX-M2 放大上行空间,也抬高资金需求。 | 财务含义 | 中 |
由财务章节中信号最强的公开论据生成。
[CI006, CI007, CI008, CI009, CI010]4.3 成本结构、烧钱和资金充足性
商业信号可信,但公开披露对毛利、烧钱、集中度和积压订单质量仍不完整。本节在财务框架下,聚焦成本结构、烧钱和资金充足性。公开资料反复指向三个事实:先进制程下 Fabless 扩张仍然昂贵;未完成融资和之后 IPO 仍然重要;DX-M2 同时抬高上行空间和资金需求。判断 DEEPX 时,先进制程下 Fabless 扩张仍然昂贵,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。未完成融资和之后 IPO 仍然重要,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CI017, CI018, CI019, CI020, CI021, CI022]
| 信号 | 公开信息解读 | 含义 | 来源质量 |
|---|---|---|---|
| 缺口 | 毛利率、现金跑道和在手订单质量仍未公开披露。 | 财务含义 | 中 |
| 质量 | 订单证明变现潜力,不等于证明单位经济模型可持续。 | 财务含义 | 中 |
| 质量 | 运营层面盈亏平衡的表述,不等于经审计盈利。 | 财务含义 | 中 |
| 毛利 | 渠道经济性很可能大幅影响实际毛利率。 | 财务含义 | 中 |
| IPO | IPO 应视为一个选项,不等于风险已经出清。 | 财务含义 | 中 |
由财务章节中信号最强的公开论据生成。
[CI011, CI012, CI013, CI014, CI015]4.4 财务判断和缺失披露
商业信号可信,但公开披露对毛利、烧钱、集中度和积压订单质量仍不完整。本节在财务框架下,聚焦财务判断和缺失披露。公开资料反复指向三个事实:单位经济性仍是披露最薄弱的一层;积压订单质量和客户集中度尚未解决;要做估值级投资判断,仍需要经审计细节。判断 DEEPX 时,单位经济性仍是披露最薄弱的一层,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。积压订单质量和客户集中度尚未解决,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CI025, CI026, CI027, CI028, CI029, CI030]
| 信号 | 公开信息解读 | 含义 | 来源质量 |
|---|---|---|---|
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:硬件销售锚定商业模式,因此这条财务证据最强。 | 财务含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:芯片、模组、板卡和开发赋能拼出变现阶梯,因此这条财务证据最强。 | 财务含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:已有公开定价,但实际 ASP 仍不透明,因此这条财务证据最强。 | 财务含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:2026 年订单、验证和收入指引都在改善,因此这条财务证据最强。 | 财务含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:按最近披露的年度基线看,历史亏损仍然显著,因此这条财务证据最强。 | 财务含义 | 中 |
由财务章节中信号最强的公开论据生成。
[CI016, CI017, CI018, CI019, CI020]| 信号 | 公开信息解读 | 含义 | 来源质量 |
|---|---|---|---|
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:商业化证明真实存在,但尚未完全经过审计验证,因此这条财务证据最强。 | 财务含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:先进制程的 fabless 扩张仍然烧钱,因此这条财务证据最强。 | 财务含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:待完成融资和后续 IPO 仍然关键,因此这条财务证据最强。 | 财务含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:DX-M2 同时抬高上行空间和资金需求,因此这条财务证据最强。 | 财务含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:单位经济模型仍是披露最薄弱的一层,因此这条财务证据最强。 | 财务含义 | 中 |
由财务章节中信号最强的公开论据生成。
[CI021, CI022, CI023, CI024, CI025]4.5 图表
05产品与技术
5.1 产品组合和形态
DEEPX 的产品故事已覆盖芯片、模组、板卡、软件工具,并延伸到更有野心的 DX-M2 生成式边缘 AI 路线图。本节在产品与技术框架下,聚焦产品组合和形态。公开资料反复指向三个事实:DX-M1、DX-M1M、DX-H1 和 DXNN 拼出更完整的栈;公司现在覆盖芯片、模组、板卡和开发工作流;DX-M2 是路线图上的可选上行层。判断 DEEPX 时,DX-M1、DX-M1M、DX-H1 和 DXNN 拼出更完整的栈,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。公司现在覆盖芯片、模组、板卡和开发工作流,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CE001, CE002, CE003, CE004, CE005, CE006]
| 信号 | 公开信息解读 | 含义 | 来源质量 |
|---|---|---|---|
| 产品 | DX-M1 是一颗 25 TOPS INT8 边缘 AI 芯片。 | 产品与技术含义 | 中 |
| 功耗 | DX-M1 工作在 1W-5W 功耗包络内,媒体还提到更低的平均功耗。 | 产品与技术含义 | 中 |
| 质量 | DX-M1 支持车规级工作范围。 | 产品与技术含义 | 中 |
| 模组 | DX-M1M 将相近算力封装进集成内存的模组。 | 产品与技术含义 | 中 |
| 产品 | DX-H1 V-NPU 把产品线延伸到 50 TOPS、面向视频的板卡。 | 产品与技术含义 | 中 |
由产品与技术章节中信号最强的公开论据生成。
[CE001, CE002, CE003, CE004, CE005]5.2 软件栈和部署流程
DEEPX 的产品故事已覆盖芯片、模组、板卡、软件工具,并延伸到更有野心的 DX-M2 生成式边缘 AI 路线图。本节在产品与技术框架下,聚焦软件栈和部署流程。公开资料反复指向三个事实:DXNN 处理编译、优化、运行步骤;Ultralytics 和 Wind River 集成改善可用性;开发者路径真实存在,但仍小于既有巨头生态。判断 DEEPX 时,DXNN 处理编译、优化、运行步骤,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。Ultralytics 和 Wind River 集成改善可用性,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CE009, CE010, CE011, CE012, CE013, CE014]
| 信号 | 公开信息解读 | 含义 | 来源质量 |
|---|---|---|---|
| 产品 | DX-H1 Quattro 把产品线延伸到 100 TOPS 低矮型板卡。 | 产品与技术含义 | 中 |
| 软件 | DXNN 承接编译、优化、仿真和运行时工作流。 | 产品与技术含义 | 中 |
| 软件 | Ultralytics 为开发者新增 DEEPX 导出路径。 | 产品与技术含义 | 中 |
| 信任 | 与 Wind River 集成,有助于强化关键任务定位。 | 产品与技术含义 | 中 |
| 路线图 | DX-M1 已能在设备端运行 SLM,拓宽了产品叙事。 | 产品与技术含义 | 中 |
由产品与技术章节中信号最强的公开论据生成。
[CE006, CE007, CE008, CE009, CE010]5.3 制造、信任和质量信号
DEEPX 的产品故事已覆盖芯片、模组、板卡、软件工具,并延伸到更有野心的 DX-M2 生成式边缘 AI 路线图。本节在产品与技术框架下,聚焦制造、信任和质量信号。公开资料反复指向三个事实:Samsung 5nm 和公开产品规格增加可信度;AEC-Q100 范围运行能力有助于工业定位;广泛公开认证证据仍比产品主张更薄。判断 DEEPX 时,Samsung 5nm 和公开产品规格增加可信度,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。AEC-Q100 范围运行能力有助于工业定位,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CE017, CE018, CE019, CE020, CE021, CE022]
| 信号 | 公开信息解读 | 含义 | 来源质量 |
|---|---|---|---|
| 路线图 | DX-M2 正按 2nm、低于 5W 的生成式边缘芯片开发。 | 产品与技术含义 | 中 |
| 规格 | 以私营初创公司标准看,产品页面披露的部署规格异常具体。 | 产品与技术含义 | 中 |
| 信任 | 公开信任证据更多落在专利和产品规格上,第三方认证较弱。 | 产品与技术含义 | 中 |
| 集成 | 板卡和嵌入式系统伙伴证明了系统级集成路径。 | 产品与技术含义 | 中 |
| 风险 | 关键产品风险在于路线图野心跑得比验证深度更快。 | 产品与技术含义 | 中 |
由产品与技术章节中信号最强的公开论据生成。
[CE011, CE012, CE013, CE014, CE015]5.4 路线图雄心和技术风险
DEEPX 的产品故事已覆盖芯片、模组、板卡、软件工具,并延伸到更有野心的 DX-M2 生成式边缘 AI 路线图。本节在产品与技术框架下,聚焦路线图雄心和技术风险。公开资料反复指向三个事实:端侧 SLM 叙事把故事从视觉扩展出去;2nm DX-M2 雄心很大,也吃资本;基准测试和验证深度仍需要独立支撑。判断 DEEPX 时,端侧 SLM 叙事把故事从视觉扩展出去,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。2nm DX-M2 雄心很大、资本需求高,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CE025, CE026, CE027, CE028, CE029, CE030]
| 信号 | 公开信息解读 | 含义 | 来源质量 |
|---|---|---|---|
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:DX-M1、DX-M1M、DX-H1 和 DXNN 拼出更完整的技术栈,因此这条产品与技术证据最强。 | 产品与技术含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:公司如今覆盖芯片、模组、板卡和开发工作流,因此这条产品与技术证据最强。 | 产品与技术含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:DX-M2 是路线图里的可选性层,因此这条产品与技术证据最强。 | 产品与技术含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:DXNN 承接编译、优化、运行步骤,因此这条产品与技术证据最强。 | 产品与技术含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:Ultralytics 与 Wind River 集成提升可用性,因此这条产品与技术证据最强。 | 产品与技术含义 | 中 |
由产品与技术章节中信号最强的公开论据生成。
[CE016, CE017, CE018, CE019, CE020]| 信号 | 公开信息解读 | 含义 | 来源质量 |
|---|---|---|---|
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:开发者路径已经跑通,但规模仍小于既有巨头生态,因此这条产品与技术证据最强。 | 产品与技术含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:Samsung 5nm 和公开产品规格增加可信度,因此这条产品与技术证据最强。 | 产品与技术含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:AEC-Q100 范围的运行能力强化工业定位,因此这条产品与技术证据最强。 | 产品与技术含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:广泛公开认证证据仍比产品主张更薄,因此这条产品与技术证据最强。 | 产品与技术含义 | 中 |
| 综合 | 公开来源与官方产品 / 合作伙伴材料共同指向:设备端 SLM 叙事把故事拓宽到视觉之外,因此这条产品与技术证据最强。 | 产品与技术含义 | 中 |
由产品与技术章节中信号最强的公开论据生成。
[CE021, CE022, CE023, CE024, CE025]5.5 图表
06客户
6.1 客户细分和采用证明
与一般非上市 AI 芯片初创公司相比,DEEPX 的客户证明更强,但重复订单质量和集中度披露仍不足。本节在客户框架下,聚焦客户细分和采用证明。公开资料反复指向三个事实:证明覆盖 OEM、分销商和集成商;垂直场景包括机器人、安防监控、智能工厂和嵌入式计算;进入市场高度依赖伙伴协助。判断 DEEPX 时,证明覆盖 OEM、分销商和集成商,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。机器人、安防监控、智能工厂和嵌入式计算这些垂直场景,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CU001, CU002, CU003, CU004, CU005, CU006]
| 信号 | 公开信息解读 | 含义 | 来源质量 |
|---|---|---|---|
| 客群 | 客户证明覆盖 OEM、分销商、集成商和终端应用伙伴。 | 客户含义 | 中 |
| 管线 | 公司称,在大规模量产前已有超过 300 项验证。 | 客户含义 | 中 |
| 进展 | EE Times 报道,公司 7 个月内达成 27 笔交易。 | 客户含义 | 中 |
| 牵引力 | briefs 与 Electronics Weekly 报道,已在 8 个国家达成 30 多项生产协议。 | 客户启示 | 中 |
| 牵引力 | KM Journal 后来报道,8 个国家已有 48 份采购订单。 | 客户启示 | 中 |
基于客户章节采用的信号最强公开论点生成。
[CU001, CU002, CU003, CU004, CU005]6.2 具名部署和结果质量
与一般非上市 AI 芯片初创公司相比,DEEPX 的客户证明更强,但重复订单质量和集中度披露仍不足。本节在客户框架下,聚焦具名部署和结果质量。公开资料反复指向三个事实:Hyundai、AAEON、Baidu、POSCO DX 和 DFI 是最强具名证明;部分证据已达到量产级,部分仍是伙伴级;经济结果可见度仍有限。判断 DEEPX 时,Hyundai、AAEON、Baidu、POSCO DX 和 DFI 是最强具名证明,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。部分证据已达到量产级、部分仍是伙伴级,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CU009, CU010, CU011, CU012, CU013, CU014]
| 信号 | 公开解读 | 启示 | 来源质量 |
|---|---|---|---|
| 验证 | Hyundai 是最强的具名机器人验证。 | 客户启示 | 中 |
| 验证 | AAEON 是最清晰的工业 OEM 规模化伙伴。 | 客户启示 | 中 |
| 验证 | Baidu 关联渠道已伸入无人机和机器人。 | 客户启示 | 中 |
| 验证 | POSCO DX 和工业用例强化了工厂场景的适配度。 | 客户启示 | 中 |
| 渠道 | 可见渠道栈扩大了全球触达。 | 客户启示 | 中 |
基于客户章节采用的信号最强公开论点生成。
[CU006, CU007, CU008, CU009, CU010]6.3 渠道扩张和地域
与一般非上市 AI 芯片初创公司相比,DEEPX 的客户证明更强,但重复订单质量和集中度披露仍不足。本节在客户框架下,聚焦渠道扩张和地域。公开资料反复指向三个事实:Avnet、DigiKey、WPG、日本伙伴和全球活动扩大触达;按公开报道,销售已经具备相当国际化程度;渠道宽度仍可能掩盖经济集中度。判断 DEEPX 时,Avnet、DigiKey、WPG、日本伙伴和全球活动扩大触达,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。按公开报道,销售已经具备相当国际化程度,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CU017, CU018, CU019, CU020, CU021, CU022]
| 信号 | 公开解读 | 启示 | 来源质量 |
|---|---|---|---|
| 缺口 | 公开数据仍缺少留存和复购指标。 | 客户启示 | 中 |
| 验证质量 | 具名验证的成熟度与经济细节差异很大。 | 客户启示 | 中 |
| 风险 | 渠道看起来很宽,仍可能掩盖经济集中度。 | 客户启示 | 中 |
| 地域 | 公开销售已明显国际化。 | 客户启示 | 中 |
| 持久性 | 客户故事里,持久性仍是公开资料最难验证的一环。 | 客户启示 | 中 |
基于客户章节采用的信号最强公开论点生成。
[CU011, CU012, CU013, CU014, CU015]6.4 留存、满意度和缺失分母
与一般非上市 AI 芯片初创公司相比,DEEPX 的客户证明更强,但重复订单质量和集中度披露仍不足。本节在客户框架下,聚焦留存、满意度和缺失分母。公开资料反复指向三个事实:NRR、GRR 或重复订单率均未公开;头部客户组合未披露;持久性是客户尽调的主要缺口。判断 DEEPX 时,NRR、GRR 或重复订单率均未公开,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。头部客户组合未披露,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CU025, CU026, CU027, CU028, CU029, CU030]
| 信号 | 公开解读 | 启示 | 来源质量 |
|---|---|---|---|
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,客户证据最强;验证覆盖 OEM、分销商和集成商。 | 客户启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,客户证据最强;垂直场景包括机器人、安防、智能工厂和嵌入式计算。 | 客户启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,客户证据最强;上市路径高度依赖伙伴助力。 | 客户启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,客户证据最强;Hyundai、AAEON、Baidu、POSCO DX 和 DFI 是最强的具名验证。 | 客户启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,客户证据最强;部分证据已达生产级,部分仍停留在伙伴级。 | 客户启示 | 中 |
基于客户章节采用的信号最强公开论点生成。
[CU016, CU017, CU018, CU019, CU020]| 信号 | 公开解读 | 启示 | 来源质量 |
|---|---|---|---|
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,客户证据最强;经济结果能见度仍有限。 | 客户启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,客户证据最强;Avnet、DigiKey、WPG、日本伙伴和全球活动扩大了触达。 | 客户启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,客户证据最强;按公开报道,销售已明显国际化。 | 客户启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,客户证据最强;渠道广度仍可能掩盖经济集中度。 | 客户启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,客户证据最强;NRR、GRR 和复购率均未公开。 | 客户启示 | 中 |
基于客户章节采用的信号最强公开论点生成。
[CU021, CU022, CU023, CU024, CU025]6.5 图表
07风险
7.1 最高优先级风险簇
关键风险集中在晶圆代工集中度、路线图雄心、融资依赖、生态竞争和公开披露有限。本节在风险框架下,聚焦最高优先级风险簇。公开资料反复指向三个事实:融资、路线图、供应、生态和披露风险相互连着;一处失败会很快传导到其他几层;监控需要落到运营指标,而不是停留在叙事。判断 DEEPX 时,融资、路线图、供应、生态和披露风险相互连接,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。一处失败会很快传导到其他几层,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CR001, CR002, CR003, CR004, CR005, CR006]
| 信号 | 公开解读 | 启示 | 来源质量 |
|---|---|---|---|
| 供应 | 与 Samsung 挂钩的制造环节,是最清晰的结构性运营依赖。 | 风险启示 | 中 |
| 融资 | 融资轮仍未关闭,且据报道 IPO 前目标缩水,融资风险仍在。 | 风险启示 | 中 |
| 路线图 | 相比 DX-M1,DX-M2 把路线图推入更依赖执行的阶段。 | 风险启示 | 中 |
| 伙伴 | 伙伴主导增长提升速度,也抬高依赖。 | 风险启示 | 中 |
| 地缘政治 | 出口管制和地缘政治是可信的行业风险。 | 风险启示 | 中 |
基于风险章节采用的信号最强公开论点生成。
[CR001, CR002, CR003, CR004, CR005]7.2 监管、出口和供应链风险
关键风险集中在晶圆代工集中度、路线图雄心、融资依赖、生态竞争和公开披露有限。本节在风险框架下,聚焦监管、出口和供应链风险。公开资料反复指向三个事实:Samsung 集中度是最清晰的结构性敞口;出口管制和地缘政治会同时影响供给与需求;先进制程切换抬高执行风险。判断 DEEPX 时,Samsung 集中度是最清晰的结构性敞口,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。出口管制和地缘政治会同时影响供给与需求,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CR009, CR010, CR011, CR012, CR013, CR014]
| 信号 | 公开解读 | 启示 | 来源质量 |
|---|---|---|---|
| 人员 | 创始人居中,抬高关键人风险。 | 风险启示 | 中 |
| 人员 | 2026 年高管离职抬高了执行敏感度。 | 风险启示 | 中 |
| 质量 | 公开质量和安全证据,比其任务关键型雄心更薄。 | 风险启示 | 中 |
| 商业 | 一旦转化停滞,溢价估值可能迅速压缩。 | 风险启示 | 中 |
| 缓释 | 更充分披露是主要缓释因素之一。 | 风险启示 | 中 |
基于风险章节采用的信号最强公开论点生成。
[CR006, CR007, CR008, CR009, CR010]7.3 伙伴、客户和人员风险
关键风险集中在晶圆代工集中度、路线图雄心、融资依赖、生态竞争和公开披露有限。本节在风险框架下,聚焦伙伴、客户和人员风险。公开资料反复指向三个事实:伙伴驱动增长会带来集中度暴露面;创始人中心性和高管流动抬高人员风险;关键任务场景需要更强质量证明。判断 DEEPX 时,伙伴驱动增长会带来集中度暴露面,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。创始人中心性和高管流动抬高人员风险,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CR017, CR018, CR019, CR020, CR021, CR022]
| 信号 | 公开解读 | 启示 | 来源质量 |
|---|---|---|---|
| 否决 | DX-M2 延迟会同时打击护城河和融资叙事。 | 风险启示 | 中 |
| 集中度 | 从经济口径看,渠道多样性可能比标识列表显示的更窄。 | 风险启示 | 中 |
| 缓释 | 当前良率证据有帮助,但不能消除供应风险。 | 风险启示 | 中 |
| 质量 | 任务关键型垂直场景会要求更强的验证包。 | 风险启示 | 中 |
| 宏观 | 宏观需求不降低融资纪律要求。 | 风险启示 | 中 |
基于风险章节采用的信号最强公开论点生成。
[CR011, CR012, CR013, CR014, CR015]7.4 缓释措施和终止标准
关键风险集中在晶圆代工集中度、路线图雄心、融资依赖、生态竞争和公开披露有限。本节在风险框架下,聚焦缓释措施和终止标准。公开资料反复指向三个事实:融资进展、DX-M2 时间表和重复量产客户是关键监控项;多元化和更好披露是主要缓释手段;终止触发器应可衡量,而不是靠猜测。判断 DEEPX 时,融资进展、DX-M2 时间表和重复量产客户是关键监控项,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。多元化和更好披露是主要缓释手段,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CR025, CR026, CR027, CR028, CR029, CR030]
| 信号 | 公开解读 | 启示 | 来源质量 |
|---|---|---|---|
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,风险证据最强;融资、路线图、供应、生态和披露风险彼此相连。 | 风险启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,风险证据最强;单点失效可能很快传导到其他多个层面。 | 风险启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,风险证据最强;监控必须落到运营层,而不是停在口号。 | 风险启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,风险证据最强;Samsung 集中度是最清晰的结构性敞口。 | 风险启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,风险证据最强;出口管制和地缘政治可能同时影响供给与需求。 | 风险启示 | 中 |
基于风险章节采用的信号最强公开论点生成。
[CR016, CR017, CR018, CR019, CR020]| 信号 | 公开解读 | 启示 | 来源质量 |
|---|---|---|---|
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,风险证据最强;先进制程切换会抬高执行风险。 | 风险启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,风险证据最强;伙伴主导增长会制造集中度暴露面。 | 风险启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,风险证据最强;创始人居中和高管更替抬高人员风险。 | 风险启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,风险证据最强;任务关键型用例需要更强质量验证。 | 风险启示 | 中 |
| 综合 | 公开来源与官方产品或伙伴材料相互印证时,风险证据最强;融资进度、DX-M2 时间点和复购生产客户是关键监控项。 | 风险启示 | 中 |
基于风险章节采用的信号最强公开论点生成。
[CR021, CR022, CR023, CR024, CR025]7.5 图表
08估值
8.1 投资论点和建议
当前估值足够可信,值得认真研究;但相对于公开财务和客户证明披露,价格仍偏高,今天还不到买入评级。本节在估值框架下,聚焦投资论点和建议。公开资料反复指向三个事实:真实进展让公司值得继续研究;公开披露仍落后于当前价格;今天更合适的结论是继续研究。判断 DEEPX 时,真实进展让公司值得继续研究,应放在物理 AI 商业化语境里看,而不是抽象 AI 炒作。公开披露仍落后于当前价格,也应按同一语境理解。主要注意点是,公开信号的证据强度并不相同;本章会区分官方产品或伙伴表述、独立报道和真正未解的缺口。[CV001, CV002, CV003, CV004, CV005, CV006]
| 信号 | 公开解读 | 启示 | 来源质量 |
|---|---|---|---|
| 判断 | 公司值得认真尽调,但以当前公开估值看,还不到坚定买入。 | 估值启示 | 中 |
| 估值 | 最清晰的估值锚是 Series D 首关的 $2.2B。 | 估值启示 | 中 |
| 验证 | 商业验证真实存在,但相对价格仍偏早。 | 估值启示 | 中 |
| 可比 | Rebellions 是最接近的区域私营估值参照。 | 估值启示 | 中 |
| 可比 | 上市龙头有助于框定估值天花板,但不是直接的创业公司可比对象。 | 估值启示 | 中 |
基于估值章节采用的信号最强公开论点生成。
[CV001, CV002, CV003, CV004, CV005]8.2 融资背景和价格支撑
当前估值值得认真研究,但相对公开披露的财务与客户验证仍偏高,当前还不足以给出买入结论。本节把估值问题拆到融资背景与价格支撑。多处公开资料反复指向三点:Series D 首次交割确立估值锚;更大一轮融资和未来 IPO 仍在变化;价格支撑取决于收入质量证据能否更扎实。落到 DEEPX,Series D 首次交割这个估值锚,以及更大一轮融资和未来 IPO 仍未定,都应放在物理 AI 商业化里看,而不是抽象 AI 热度。主要提醒是,公开信号的证据权重不同;本章会区分官方产品或合作伙伴声明、独立报道和真正未解决的缺口。[CV009, CV010, CV011, CV012, CV013, CV014]
| 信号 | 公开解读 | 启示 | 来源质量 |
|---|---|---|---|
| 多头 | 多头情景取决于转化与 DX-M2 执行。 | 估值启示 | 中 |
| 空头 | 空头情景是:市场先给融资可选性定价,耐久经济性还没跟上。 | 估值启示 | 中 |
| 上调 | 审计后的收入质量和复购客户,是上调判断的主要触发器。 | 估值启示 | 中 |
| 下行 | 融资停滞或 DX-M2 延误,可能很快压缩估值。 | 估值启示 | 中 |
| 缺口 | 现有披露不足以支持不看价格的买入。 | 估值启示 | 中 |
基于估值章节采用的信号最强公开论点生成。
[CV006, CV007, CV008, CV009, CV010]8.3 乐观、基准、悲观情景与可比对象
当前估值值得认真研究,但相对公开披露的财务与客户验证仍偏高,当前还不足以给出买入结论。本节把估值问题拆到乐观、基准、悲观情景与可比对象。多处公开资料反复指向三点:Rebellions 是最接近的区域性私募可比对象;上市龙头只能作为上限可比,不是直接参照;结果取决于转化和路线图执行。落到 DEEPX,Rebellions 这个最接近的区域性私募可比,以及上市龙头只能作为上限可比而非直接匹配,都应放在物理 AI 商业化里看,而不是抽象 AI 热度。主要提醒是,公开信号的证据权重不同;本章会区分官方产品或合作伙伴声明、独立报道和真正未解决的缺口。[CV017, CV018, CV019, CV020, CV021, CV022]
| 信号 | 公开解读 | 启示 | 来源质量 |
|---|---|---|---|
| 立场 | 正确立场应跟着证据走,不是反对公司。 | 估值含义 | 中 |
| 风险 | 即使技术执行没掉链子,客户集中度也可能压住上行空间。 | 估值含义 | 中 |
| 成熟度 | DEEPX 比典型 AI 芯片初创公司更成熟,但离当前价格隐含阶段还有距离。 | 估值含义 | 中 |
| 路径 | 证据更扎实,或价格更便宜,判断才可能上调。 | 估值含义 | 中 |
| 建议 | 当前建议是继续研究。 | 估值含义 | 中 |
基于估值章节中信号最强的公开说法生成。
[CV011, CV012, CV013, CV014, CV015]8.4 推翻投资判断的因素与最终尽调问题
当前估值值得认真研究,但相对公开披露的财务与客户验证仍偏高,当前还不足以给出买入结论。本节把估值问题拆到推翻投资判断的因素与最终尽调问题。多处公开资料反复指向三点:审计财务、留存、集中度和路线图证明仍是闸门问题;更强证据或更好价格可能改善判断;当前立场对证据敏感,并非反对公司。落到 DEEPX,审计财务、留存、集中度和路线图证明这些闸门问题,以及更强证据或更好价格可能改善判断,都应放在物理 AI 商业化里看,而不是抽象 AI 热度。主要提醒是,公开信号的证据权重不同;本章会区分官方产品或合作伙伴声明、独立报道和真正未解决的缺口。[CV025, CV026, CV027, CV028, CV029, CV030]
| 信号 | 公开信息解读 | 影响 | 来源质量 |
|---|---|---|---|
| 综合 | 公开来源与官方产品或合作伙伴材料都指向同一方向:真实牵引力让这家公司值得进入投资研究。 | 估值含义 | 中 |
| 综合 | 公开来源与官方产品或合作伙伴材料都指向同一方向:公开披露仍跟不上当前价格。 | 估值含义 | 中 |
| 综合 | 公开来源与官方产品或合作伙伴材料都指向同一方向:今天合适的判断是继续研究。 | 估值含义 | 中 |
| 综合 | 公开来源与官方产品或合作伙伴材料都指向同一方向:Series D 首次交割给估值定下锚点。 | 估值含义 | 中 |
| 综合 | 公开来源与官方产品或合作伙伴材料都指向同一方向:更大一轮融资和未来 IPO 仍有变数。 | 估值含义 | 中 |
基于估值章节中信号最强的公开说法生成。
[CV016, CV017, CV018, CV019, CV020]| 信号 | 公开信息解读 | 影响 | 来源质量 |
|---|---|---|---|
| 综合 | 公开来源与官方产品或合作伙伴材料都指向同一方向:当前价格能否站住,要看收入质量证据能否补强。 | 估值含义 | 中 |
| 综合 | 公开来源与官方产品或合作伙伴材料都指向同一方向:Rebellions 是区域内最接近的未上市可比公司。 | 估值含义 | 中 |
| 综合 | 公开来源与官方产品或合作伙伴材料都指向同一方向:公开市场龙头只能当估值上限参照,不是直接可比。 | 估值含义 | 中 |
| 综合 | 公开来源与官方产品或合作伙伴材料都指向同一方向:结果取决于转化能否落地、路线图能否兑现。 | 估值含义 | 中 |
| 综合 | 公开来源与官方产品或合作伙伴材料都指向同一方向:经审计财务、留存、客户集中度和路线图证明,仍是放行前必须追问的问题。 | 估值含义 | 中 |
基于估值章节中信号最强的公开说法生成。
[CV021, CV022, CV023, CV024, CV025]| 信号 | 公开信息解读 | 影响 | 来源质量 |
|---|---|---|---|
| 综合 | 公开来源与官方产品或合作伙伴材料都指向同一方向:证据更扎实,或价格更便宜,判断才可能上调。 | 估值含义 | 中 |
| 综合 | 公开来源与官方产品或合作伙伴材料都指向同一方向:当前立场跟证据走,不是看空公司。 | 估值含义 | 中 |
| 综合 | DEEPX 当前估值判断仍偏建设性但保持谨慎,缺失披露仍可能改写结论。 | 估值含义 | 中 |
| 综合 | 把估值、市场、产品和客户叙事放在物理 AI 商业化框架下看,比当作纯 AI 热点故事更可信。 | 估值含义 | 中 |
| 综合 | 产品、合作伙伴的公开证据明显强于财务韧性和治理。 | 估值含义 | 中 |
基于估值章节中信号最强的公开说法生成。
[CV026, CV027, CV028, CV029, CV030]8.5 图表
免责声明
本报告基于截至 2026-08-08 的公开证据,不构成投资建议。公开来源相互冲突时,本报告保留不确定性, 不强行给出虚假的精确性。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | Founded in 2018 and headquartered in Pangyo, Seongnam-si. | 中 | SO001, SO009 |
| CO002 | Private Series D company focused on edge AI semiconductors for physical devices. | 中 | SO001, SO023 |
| CO003 | Lokwon Kim is the founder and CEO across official and independent sources. | 中 | SO008, SO001, SO005 |
| CO004 | DX-M1 is the current flagship and center of commercial activity. | 中 | SO011 |
| CO005 | Series D first close was about $30M at $2.2B valuation in August 2026. | 中 | SO002, SO003 |
| CO006 | Round remains open and full-round targets vary across outlets. | 中 | SO002, SO003 |
| CO007 | PitchBook and Tracxn disagree on headcount and lifetime capital raised. | 中 | SO001 |
| CO008 | EE Times reported 27 deals in seven months. | 中 | SO004 |
| CO009 | briefs and Electronics Weekly reported 30-plus agreements across eight countries. | 中 | SO002, SO003 |
| CO010 | The Elec set a 2026 revenue target of KRW 60B from a KRW 3.3B prior-year base. | 中 | SO006 |
| CO011 | Sedaily reported a scaled-back pre-IPO target amid valuation skepticism. | 中 | SO007 |
| CO012 | Official contact pages show Korea, U.S., Taiwan, and China locations. | 中 | SO009 |
| CO013 | Trust-center materials emphasize patents, longevity, and quality posture. | 中 | SO024 |
| CO014 | AAEON, Hyundai, Avnet, and partner announcements show channel-led scale-up. | 中 | SO013, SO015 |
| CO015 | DX-M1 is proven; DX-M2 is still optionality. | 中 | SO011, SO014 |
| CO016 | Company Overview evidence is strongest where public sources and official product or partner materials point in the same direction on Pangyo-based fabless edge-AI semiconductor vendor. | 中 | SO001, SO002, SO003 |
| CO017 | Company Overview evidence is strongest where public sources and official product or partner materials point in the same direction on sells chips, modules, and deployment-enabling software. | 中 | SO001, SO002, SO003 |
| CO018 | Company Overview evidence is strongest where public sources and official product or partner materials point in the same direction on targets robots, cameras, industrial systems, and edge servers. | 中 | SO001, SO002, SO003 |
| CO019 | Company Overview evidence is strongest where public sources and official product or partner materials point in the same direction on Lokwon Kim is the consistent founder-CEO reference. | 中 | SO001, SO002, SO003 |
| CO020 | Company Overview evidence is strongest where public sources and official product or partner materials point in the same direction on named executives broaden the bench but not board disclosure. | 中 | SO001, SO002, SO003 |
| CO021 | Company Overview evidence is strongest where public sources and official product or partner materials point in the same direction on 2026 turnover reports raise execution sensitivity. | 中 | SO001, SO002, SO003 |
| CO022 | Company Overview evidence is strongest where public sources and official product or partner materials point in the same direction on Series D first close set the current mark at $2.2B. | 中 | SO001, SO002, SO003 |
| CO023 | Company Overview evidence is strongest where public sources and official product or partner materials point in the same direction on orders and validations imply real momentum. | 中 | SO001, SO002, SO003 |
| CO024 | Company Overview evidence is strongest where public sources and official product or partner materials point in the same direction on lifetime capital and cap-table terms remain noisy. | 中 | SO001, SO002, SO003 |
| CO025 | Company Overview evidence is strongest where public sources and official product or partner materials point in the same direction on DX-M1 moved into mass production late 2025. | 中 | SO001, SO002, SO003 |
| CO026 | Company Overview evidence is strongest where public sources and official product or partner materials point in the same direction on 2026 brought OEM, channel, and robotics announcements. | 中 | SO001, SO002, SO003 |
| CO027 | Company Overview evidence is strongest where public sources and official product or partner materials point in the same direction on pre-IPO contraction is the clearest adverse signal. | 中 | SO001, SO002, SO003 |
| CO028 | The current company-overview read for DEEPX remains constructive but cautious because missing disclosure could still change the conclusion. | 中 | SO001, SO002, SO003 |
| CO029 | The valuation, market, product, and customer narrative is more credible when read as a physical-AI commercialization story than as a pure AI hype story. | 中 | SO001, SO002, SO003 |
| CO030 | Public evidence quality is materially better on product and partnerships than on financial durability and governance. | 中 | SO001, SO002, SO003 |
| CO031 | Additional diligence could still change the current company-overview conclusion because key disclosures remain private. | 低 | SO001 |
| CO032 | Additional diligence could still change the current company-overview conclusion because key disclosures remain private. | 低 | SO001 |
| CO033 | Additional diligence could still change the current company-overview conclusion because key disclosures remain private. | 低 | SO001 |
| CO034 | Additional diligence could still change the current company-overview conclusion because key disclosures remain private. | 低 | SO001 |
| CO035 | Additional diligence could still change the current company-overview conclusion because key disclosures remain private. | 低 | SO001 |
| CM001 | DEEPX belongs to the edge AI semiconductor market, not the training-GPU market. | 中 | SM003, SM006, SM011 |
| CM002 | Fortune sizes the 2026 edge AI semiconductor market at about $29.85B. | 中 | SM003 |
| CM003 | ABI sizes the 2026 edge AI chipset market at $34.4B. | 中 | SM004 |
| CM004 | Future Markets projects the edge AI chip market above $80B by 2036. | 中 | SM002 |
| CM005 | Axis groups 2026 edge-AI opportunity near $30B. | 中 | SM001 |
| CM006 | Industrial automation and robotics are high-value verticals for merchant edge inference. | 中 | SM004, SM005, SM008 |
| CM007 | OEMs, distributors, and system integrators are the real buyers. | 中 | SM016, SM018 |
| CM008 | Evaluation, board integration, validation, stocking, and production win are the normal conversion stages. | 中 | SM023, SM016, SM018 |
| CM009 | Latency, privacy, bandwidth, and power are recurring purchase drivers. | 中 | SM006, SM003 |
| CM010 | Advanced-node cost and export controls remain structural constraints. | 中 | SM003, SM019 |
| CM011 | DEEPX use-case pages line up closely with the industrial and robotics wedge. | 中 | SM011, SM008, SM014 |
| CM012 | APAC matters because manufacturing, electronics, and robotics density overlap there. | 中 | SM003, SM004 |
| CM013 | Channel depth influences adoption nearly as much as silicon specs. | 中 | SM005, SM017 |
| CM014 | The practical SOM is narrower than the top-down TAM headlines. | 中 | SM003, SM002, SM001 |
| CM015 | The relevant workload is local inference under tight thermal envelopes. | 中 | SM007, SM013, SM009 |
| CM016 | Market Analysis evidence is strongest where public sources and official product or partner materials point in the same direction on merchant edge inference rather than cloud training. | 中 | SM001, SM002, SM003 |
| CM017 | Market Analysis evidence is strongest where public sources and official product or partner materials point in the same direction on focus on robots, smart cameras, gateways, and industrial compute. | 中 | SM001, SM002, SM003 |
| CM018 | Market Analysis evidence is strongest where public sources and official product or partner materials point in the same direction on exclude broad cloud-AI spend from the practical wedge. | 中 | SM001, SM002, SM003 |
| CM019 | Market Analysis evidence is strongest where public sources and official product or partner materials point in the same direction on public estimates cluster around a roughly $30B+ 2026 opportunity. | 中 | SM001, SM002, SM003 |
| CM020 | Market Analysis evidence is strongest where public sources and official product or partner materials point in the same direction on definitions vary between semiconductors, chipsets, and broader edge AI. | 中 | SM001, SM002, SM003 |
| CM021 | Market Analysis evidence is strongest where public sources and official product or partner materials point in the same direction on industrial and robotics carry higher value density. | 中 | SM001, SM002, SM003 |
| CM022 | Market Analysis evidence is strongest where public sources and official product or partner materials point in the same direction on OEMs, distributors, and integrators are the real buyers. | 中 | SM001, SM002, SM003 |
| CM023 | Market Analysis evidence is strongest where public sources and official product or partner materials point in the same direction on evaluation kits and board validation precede production. | 中 | SM001, SM002, SM003 |
| CM024 | Market Analysis evidence is strongest where public sources and official product or partner materials point in the same direction on channel leverage matters as much as specs. | 中 | SM001, SM002, SM003 |
| CM025 | Market Analysis evidence is strongest where public sources and official product or partner materials point in the same direction on latency, privacy, bandwidth, and power drive demand. | 中 | SM001, SM002, SM003 |
| CM026 | Market Analysis evidence is strongest where public sources and official product or partner materials point in the same direction on advanced-node cost and export controls constrain supply. | 中 | SM001, SM002, SM003 |
| CM027 | Market Analysis evidence is strongest where public sources and official product or partner materials point in the same direction on software lock-in shapes adoption timing. | 中 | SM001, SM002, SM003 |
| CM028 | The current market-analysis read for DEEPX remains constructive but cautious because missing disclosure could still change the conclusion. | 中 | SM001, SM002, SM003 |
| CM029 | The valuation, market, product, and customer narrative is more credible when read as a physical-AI commercialization story than as a pure AI hype story. | 中 | SM001, SM002, SM003 |
| CM030 | Public evidence quality is materially better on product and partnerships than on financial durability and governance. | 中 | SM001, SM002, SM003 |
| CM031 | Additional diligence could still change the current market-analysis conclusion because key disclosures remain private. | 低 | SM001 |
| CM032 | Additional diligence could still change the current market-analysis conclusion because key disclosures remain private. | 低 | SM001 |
| CM033 | Additional diligence could still change the current market-analysis conclusion because key disclosures remain private. | 低 | SM001 |
| CM034 | Additional diligence could still change the current market-analysis conclusion because key disclosures remain private. | 低 | SM001 |
| CM035 | Additional diligence could still change the current market-analysis conclusion because key disclosures remain private. | 低 | SM001 |
| CP001 | Hailo is the closest direct specialist peer by form factor and use case. | 中 | SP001, SP011 |
| CP002 | Qualcomm competes through integrated platforms rather than pure discrete NPUs. | 中 | SP002 |
| CP003 | Coral competes for developer accessibility and small-scale deployment. | 中 | SP003 |
| CP004 | Apple is a captive benchmark, not a merchant OEM substitute. | 中 | SP004 |
| CP005 | NXP competes through industrial and safety-heavy systems. | 中 | SP006 |
| CP006 | NVIDIA remains the ecosystem benchmark DEEPX must work around. | 中 | SP007, SP018 |
| CP007 | Rockchip is a lower-cost AIoT substitute path. | 中 | SP005 |
| CP008 | Rebellions is a relevant Korean AI-silicon comparator with a different workload mix. | 中 | SP009 |
| CP009 | DEEPX has unusually visible public chip and module pricing. | 中 | SP015, SP016 |
| CP010 | Integrated SoCs often win on simplicity even if discrete NPUs win on efficiency. | 中 | SP002, SP006, SP007 |
| CP011 | DEEPX depends on partner and workflow quality to compete with bigger ecosystems. | 中 | SP017, SP012, SP024 |
| CP012 | No single peer closes every DEEPX advantage, but several can crowd the niche. | 中 | SP001, SP002, SP007 |
| CP013 | Merchant discreteness is a meaningful differentiator for retrofit programs. | 中 | SP012, SP023 |
| CP014 | System incumbents can compress a silicon lead through channel power. | 中 | SP002, SP007 |
| CP015 | Moat durability depends on silicon, software, and partner momentum together. | 中 | SP024 |
| CP016 | Competitors evidence is strongest where public sources and official product or partner materials point in the same direction on Hailo is the closest specialist peer. | 中 | SP001, SP002, SP003 |
| CP017 | Competitors evidence is strongest where public sources and official product or partner materials point in the same direction on Qualcomm, NXP, and NVIDIA are system-level incumbents. | 中 | SP001, SP002, SP003 |
| CP018 | Competitors evidence is strongest where public sources and official product or partner materials point in the same direction on Rockchip and Coral represent substitute paths. | 中 | SP001, SP002, SP003 |
| CP019 | Competitors evidence is strongest where public sources and official product or partner materials point in the same direction on DX-M1 emphasizes sub-5W merchant deployment. | 中 | SP001, SP002, SP003 |
| CP020 | Competitors evidence is strongest where public sources and official product or partner materials point in the same direction on public pricing visibility is better than many peers. | 中 | SP001, SP002, SP003 |
| CP021 | Competitors evidence is strongest where public sources and official product or partner materials point in the same direction on integrated platforms often win on simplicity. | 中 | SP001, SP002, SP003 |
| CP022 | Competitors evidence is strongest where public sources and official product or partner materials point in the same direction on NVIDIA and Qualcomm own stronger tool chains. | 中 | SP001, SP002, SP003 |
| CP023 | Competitors evidence is strongest where public sources and official product or partner materials point in the same direction on DEEPX is building partner-led distribution and workflow bridges. | 中 | SP001, SP002, SP003 |
| CP024 | Competitors evidence is strongest where public sources and official product or partner materials point in the same direction on ecosystem depth is the main competitive gap. | 中 | SP001, SP002, SP003 |
| CP025 | Competitors evidence is strongest where public sources and official product or partner materials point in the same direction on merchant discreteness and power efficiency are real advantages. | 中 | SP001, SP002, SP003 |
| CP026 | Competitors evidence is strongest where public sources and official product or partner materials point in the same direction on incumbents can compress silicon advantage with software and channel power. | 中 | SP001, SP002, SP003 |
| CP027 | Competitors evidence is strongest where public sources and official product or partner materials point in the same direction on DEEPX must defend silicon, software, and partners together. | 中 | SP001, SP002, SP003 |
| CP028 | The current competitors read for DEEPX remains constructive but cautious because missing disclosure could still change the conclusion. | 中 | SP001, SP002, SP003 |
| CP029 | The valuation, market, product, and customer narrative is more credible when read as a physical-AI commercialization story than as a pure AI hype story. | 中 | SP001, SP002, SP003 |
| CP030 | Public evidence quality is materially better on product and partnerships than on financial durability and governance. | 中 | SP001, SP002, SP003 |
| CP031 | Additional diligence could still change the current competitors conclusion because key disclosures remain private. | 低 | SP001 |
| CP032 | Additional diligence could still change the current competitors conclusion because key disclosures remain private. | 低 | SP001 |
| CP033 | Additional diligence could still change the current competitors conclusion because key disclosures remain private. | 低 | SP001 |
| CP034 | Additional diligence could still change the current competitors conclusion because key disclosures remain private. | 低 | SP001 |
| CP035 | Additional diligence could still change the current competitors conclusion because key disclosures remain private. | 低 | SP001 |
| CI001 | The model is hardware-led rather than SaaS-led. | 中 | SI014, SI001 |
| CI002 | The Elec cited about $50 chip pricing and about $250 module pricing. | 中 | SI001 |
| CI003 | The Elec reported roughly 30 purchase orders worth about $6.7M by April 2026. | 中 | SI001 |
| CI004 | The company reported about 350 proof-of-concept companies. | 中 | SI001, SI017 |
| CI005 | KM Journal reported 48 purchase orders across eight countries within eight months. | 中 | SI004 |
| CI006 | KM Journal said about 67% of sales were overseas. | 中 | SI004 |
| CI007 | 2026 revenue target was raised to KRW 60B. | 中 | SI001 |
| CI008 | Historic losses remained material on the latest public annual baseline. | 中 | SI001, SI002 |
| CI009 | Open funding and later IPO timing still matter to the capital plan. | 中 | SI003, SI002, SI011 |
| CI010 | DX-M2 increases funding needs along with upside. | 中 | SI005 |
| CI011 | Gross margin, runway, and backlog quality are still undisclosed publicly. | 中 | SI008, SI002 |
| CI012 | Orders prove monetization potential, not durable unit economics. | 中 | SI001, SI012 |
| CI013 | Operational breakeven commentary is not the same as audited profitability. | 中 | SI004 |
| CI014 | Channel economics likely matter heavily to realized margins. | 中 | SI016 |
| CI015 | IPO should be treated as an option, not a completed de-risking event. | 中 | SI003, SI002 |
| CI016 | Financials evidence is strongest where public sources and official product or partner materials point in the same direction on hardware sales anchor the model. | 中 | SI001, SI002, SI003 |
| CI017 | Financials evidence is strongest where public sources and official product or partner materials point in the same direction on chips, modules, cards, and enablement create the monetization ladder. | 中 | SI001, SI002, SI003 |
| CI018 | Financials evidence is strongest where public sources and official product or partner materials point in the same direction on public pricing exists but realized ASPs are still opaque. | 中 | SI001, SI002, SI003 |
| CI019 | Financials evidence is strongest where public sources and official product or partner materials point in the same direction on orders, validations, and revenue guidance all improved in 2026. | 中 | SI001, SI002, SI003 |
| CI020 | Financials evidence is strongest where public sources and official product or partner materials point in the same direction on historic losses remain material on the last disclosed annual baseline. | 中 | SI001, SI002, SI003 |
| CI021 | Financials evidence is strongest where public sources and official product or partner materials point in the same direction on commercial proof is real but not yet fully audited. | 中 | SI001, SI002, SI003 |
| CI022 | Financials evidence is strongest where public sources and official product or partner materials point in the same direction on advanced-node fabless scaling remains expensive. | 中 | SI001, SI002, SI003 |
| CI023 | Financials evidence is strongest where public sources and official product or partner materials point in the same direction on open funding and a later IPO still matter. | 中 | SI001, SI002, SI003 |
| CI024 | Financials evidence is strongest where public sources and official product or partner materials point in the same direction on DX-M2 raises both upside and funding needs. | 中 | SI001, SI002, SI003 |
| CI025 | Financials evidence is strongest where public sources and official product or partner materials point in the same direction on unit economics remain the weakest disclosed layer. | 中 | SI001, SI002, SI003 |
| CI026 | Financials evidence is strongest where public sources and official product or partner materials point in the same direction on backlog quality and customer concentration are unresolved. | 中 | SI001, SI002, SI003 |
| CI027 | Financials evidence is strongest where public sources and official product or partner materials point in the same direction on valuation-grade underwriting still needs audited detail. | 中 | SI001, SI002, SI003 |
| CI028 | The current financials read for DEEPX remains constructive but cautious because missing disclosure could still change the conclusion. | 中 | SI001, SI002, SI003 |
| CI029 | The valuation, market, product, and customer narrative is more credible when read as a physical-AI commercialization story than as a pure AI hype story. | 中 | SI001, SI002, SI003 |
| CI030 | Public evidence quality is materially better on product and partnerships than on financial durability and governance. | 中 | SI001, SI002, SI003 |
| CI031 | Additional diligence could still change the current financials conclusion because key disclosures remain private. | 低 | SI001 |
| CI032 | Additional diligence could still change the current financials conclusion because key disclosures remain private. | 低 | SI001 |
| CI033 | Additional diligence could still change the current financials conclusion because key disclosures remain private. | 低 | SI001 |
| CI034 | Additional diligence could still change the current financials conclusion because key disclosures remain private. | 低 | SI001 |
| CI035 | Additional diligence could still change the current financials conclusion because key disclosures remain private. | 低 | SI001 |
| CE001 | DX-M1 is a 25 TOPS INT8 edge AI chip. | 中 | SE001 |
| CE002 | DX-M1 runs in a 1W-5W envelope with lower average-power references in press. | 中 | SE001 |
| CE003 | DX-M1 supports automotive-grade operating ranges. | 中 | SE001 |
| CE004 | DX-M1M packages similar compute into a module with integrated memory. | 中 | SE002 |
| CE005 | DX-H1 V-NPU extends the line into a 50 TOPS video-focused card. | 中 | SE003 |
| CE006 | DX-H1 Quattro extends the line into a 100 TOPS low-profile card. | 中 | SE004 |
| CE007 | DXNN handles compile, optimize, simulate, and runtime workflow. | 中 | SE005, SE007 |
| CE008 | Ultralytics added a DEEPX export path for developers. | 中 | SE012, SE014, SE015 |
| CE009 | Wind River integration helps mission-critical positioning. | 中 | SE013, SE016 |
| CE010 | DX-M1 now running SLMs on device broadens the product narrative. | 中 | SE010 |
| CE011 | DX-M2 is being developed as a 2nm sub-5W generative edge chip. | 中 | SE011, SE023 |
| CE012 | Product pages expose unusually concrete deployment specs for a private startup. | 中 | SE001, SE003, SE004 |
| CE013 | Public trust evidence is stronger on patents and product specs than on third-party certifications. | 中 | SE009 |
| CE014 | Board and embedded-system partners prove system-level integration paths. | 中 | SE017, SE018, SE022 |
| CE015 | The key product risk is roadmap ambition outrunning validation depth. | 中 | SE023, SE011 |
| CE016 | Product and Technology evidence is strongest where public sources and official product or partner materials point in the same direction on DX-M1, DX-M1M, DX-H1, and DXNN create a fuller stack. | 中 | SE001, SE002, SE003 |
| CE017 | Product and Technology evidence is strongest where public sources and official product or partner materials point in the same direction on the company now addresses chips, modules, cards, and dev workflows. | 中 | SE001, SE002, SE003 |
| CE018 | Product and Technology evidence is strongest where public sources and official product or partner materials point in the same direction on DX-M2 is the roadmap optionality layer. | 中 | SE001, SE002, SE003 |
| CE019 | Product and Technology evidence is strongest where public sources and official product or partner materials point in the same direction on DXNN handles compile-optimize-run steps. | 中 | SE001, SE002, SE003 |
| CE020 | Product and Technology evidence is strongest where public sources and official product or partner materials point in the same direction on Ultralytics and Wind River integrations improve usability. | 中 | SE001, SE002, SE003 |
| CE021 | Product and Technology evidence is strongest where public sources and official product or partner materials point in the same direction on developer path is real but still smaller than incumbent ecosystems. | 中 | SE001, SE002, SE003 |
| CE022 | Product and Technology evidence is strongest where public sources and official product or partner materials point in the same direction on Samsung 5nm and public product specs add credibility. | 中 | SE001, SE002, SE003 |
| CE023 | Product and Technology evidence is strongest where public sources and official product or partner materials point in the same direction on AEC-Q100-range operation helps industrial positioning. | 中 | SE001, SE002, SE003 |
| CE024 | Product and Technology evidence is strongest where public sources and official product or partner materials point in the same direction on broad public certification evidence remains thinner than product claims. | 中 | SE001, SE002, SE003 |
| CE025 | Product and Technology evidence is strongest where public sources and official product or partner materials point in the same direction on on-device SLM messaging broadens the story beyond vision. | 中 | SE001, SE002, SE003 |
| CE026 | Product and Technology evidence is strongest where public sources and official product or partner materials point in the same direction on DX-M2 on 2nm is ambitious and capital-intensive. | 中 | SE001, SE002, SE003 |
| CE027 | Product and Technology evidence is strongest where public sources and official product or partner materials point in the same direction on benchmark and validation depth still need independent support. | 中 | SE001, SE002, SE003 |
| CE028 | The current product-tech read for DEEPX remains constructive but cautious because missing disclosure could still change the conclusion. | 中 | SE001, SE002, SE003 |
| CE029 | The valuation, market, product, and customer narrative is more credible when read as a physical-AI commercialization story than as a pure AI hype story. | 中 | SE001, SE002, SE003 |
| CE030 | Public evidence quality is materially better on product and partnerships than on financial durability and governance. | 中 | SE001, SE002, SE003 |
| CE031 | Additional diligence could still change the current product-tech conclusion because key disclosures remain private. | 低 | SE001 |
| CE032 | Additional diligence could still change the current product-tech conclusion because key disclosures remain private. | 低 | SE001 |
| CE033 | Additional diligence could still change the current product-tech conclusion because key disclosures remain private. | 低 | SE001 |
| CE034 | Additional diligence could still change the current product-tech conclusion because key disclosures remain private. | 低 | SE001 |
| CE035 | Additional diligence could still change the current product-tech conclusion because key disclosures remain private. | 低 | SE001 |
| CU001 | Customer proof spans OEMs, distributors, integrators, and end-application partners. | 中 | SU001, SU006 |
| CU002 | The company reported more than 300 validations before broad production scale. | 中 | SU010 |
| CU003 | EE Times reported 27 deals in seven months. | 中 | SU012 |
| CU004 | briefs and Electronics Weekly reported 30-plus production agreements in eight countries. | 中 | SU014, SU015 |
| CU005 | KM Journal later reported 48 purchase orders across eight countries. | 中 | SU011 |
| CU006 | Hyundai provides the strongest named robotics proof. | 中 | SU004, SU005 |
| CU007 | AAEON is the clearest industrial OEM scale partner. | 中 | SU001, SU002 |
| CU008 | Baidu-linked channels extend into drones and robotics. | 中 | SU013 |
| CU009 | POSCO DX and industrial use cases reinforce factory relevance. | 中 | SU007 |
| CU010 | The visible channel stack expands global reach. | 中 | SU006, SU003 |
| CU011 | Public data still lacks retention and repeat-order metrics. | 中 | SU016, SU011 |
| CU012 | Named proof is heterogeneous in maturity and economic detail. | 中 | SU009, SU008, SU001 |
| CU013 | Channel breadth can still hide economic concentration. | 中 | SU006, SU002 |
| CU014 | Sales already appear materially international. | 中 | SU011 |
| CU015 | Durability remains the hardest part of the customer story to verify publicly. | 中 | SU016, SU014 |
| CU016 | Customers evidence is strongest where public sources and official product or partner materials point in the same direction on proof spans OEMs, distributors, and integrators. | 中 | SU001, SU002, SU003 |
| CU017 | Customers evidence is strongest where public sources and official product or partner materials point in the same direction on verticals include robotics, surveillance, smart factory, and embedded compute. | 中 | SU001, SU002, SU003 |
| CU018 | Customers evidence is strongest where public sources and official product or partner materials point in the same direction on the route to market is heavily partner-assisted. | 中 | SU001, SU002, SU003 |
| CU019 | Customers evidence is strongest where public sources and official product or partner materials point in the same direction on Hyundai, AAEON, Baidu, POSCO DX, and DFI are the strongest named proofs. | 中 | SU001, SU002, SU003 |
| CU020 | Customers evidence is strongest where public sources and official product or partner materials point in the same direction on some evidence is production-grade, some is still partner-grade. | 中 | SU001, SU002, SU003 |
| CU021 | Customers evidence is strongest where public sources and official product or partner materials point in the same direction on economic outcome visibility remains limited. | 中 | SU001, SU002, SU003 |
| CU022 | Customers evidence is strongest where public sources and official product or partner materials point in the same direction on Avnet, DigiKey, WPG, Japan partners, and global events expand reach. | 中 | SU001, SU002, SU003 |
| CU023 | Customers evidence is strongest where public sources and official product or partner materials point in the same direction on sales are already materially international by public reports. | 中 | SU001, SU002, SU003 |
| CU024 | Customers evidence is strongest where public sources and official product or partner materials point in the same direction on channel breadth may still mask economic concentration. | 中 | SU001, SU002, SU003 |
| CU025 | Customers evidence is strongest where public sources and official product or partner materials point in the same direction on no NRR, GRR, or repeat-order rates are public. | 中 | SU001, SU002, SU003 |
| CU026 | Customers evidence is strongest where public sources and official product or partner materials point in the same direction on top-customer mix is undisclosed. | 中 | SU001, SU002, SU003 |
| CU027 | Customers evidence is strongest where public sources and official product or partner materials point in the same direction on durability is the main customer diligence gap. | 中 | SU001, SU002, SU003 |
| CU028 | The current customers read for DEEPX remains constructive but cautious because missing disclosure could still change the conclusion. | 中 | SU001, SU002, SU003 |
| CU029 | The valuation, market, product, and customer narrative is more credible when read as a physical-AI commercialization story than as a pure AI hype story. | 中 | SU001, SU002, SU003 |
| CU030 | Public evidence quality is materially better on product and partnerships than on financial durability and governance. | 中 | SU001, SU002, SU003 |
| CU031 | Additional diligence could still change the current customers conclusion because key disclosures remain private. | 低 | SU001 |
| CU032 | Additional diligence could still change the current customers conclusion because key disclosures remain private. | 低 | SU001 |
| CU033 | Additional diligence could still change the current customers conclusion because key disclosures remain private. | 低 | SU001 |
| CU034 | Additional diligence could still change the current customers conclusion because key disclosures remain private. | 低 | SU001 |
| CU035 | Additional diligence could still change the current customers conclusion because key disclosures remain private. | 低 | SU001 |
| CR001 | Samsung-linked manufacturing is the clearest structural operating dependency. | 中 | SR004, SR003 |
| CR002 | Funding risk remains active because the round is open and the pre-IPO target reportedly shrank. | 中 | SR001 |
| CR003 | DX-M2 pushes the roadmap into a more execution-sensitive regime than DX-M1. | 中 | SR003 |
| CR004 | Partner-led growth increases both speed and dependency. | 中 | SR013, SR009 |
| CR005 | Export controls and geopolitics are credible sector risks. | 中 | SR002, SR021 |
| CR006 | Founder centrality raises key-person risk. | 中 | SR018 |
| CR007 | 2026 executive departures increase execution sensitivity. | 中 | SR004 |
| CR008 | Public quality and security evidence is thinner than the mission-critical ambition. | 中 | SR005, SR010 |
| CR009 | If conversion stalls, the premium valuation can compress quickly. | 中 | SR001 |
| CR010 | Better disclosure is one of the main mitigants. | 中 | SR001 |
| CR011 | A DX-M2 delay would hit both moat and financing narratives. | 中 | SR003 |
| CR012 | Channel diversity may be narrower economically than the logo list suggests. | 中 | SR013 |
| CR013 | Current yield evidence helps but does not remove supply risk. | 中 | SR004 |
| CR014 | Mission-critical verticals will ask for stronger validation packages. | 中 | SR009, SR005 |
| CR015 | Macro demand does not reduce financing discipline. | 中 | SR021, SR001 |
| CR016 | Risks evidence is strongest where public sources and official product or partner materials point in the same direction on funding, roadmap, supply, ecosystem, and disclosure risks are interconnected. | 中 | SR001, SR002, SR003 |
| CR017 | Risks evidence is strongest where public sources and official product or partner materials point in the same direction on one failure can transmit quickly into several other layers. | 中 | SR001, SR002, SR003 |
| CR018 | Risks evidence is strongest where public sources and official product or partner materials point in the same direction on monitoring needs to be operational not rhetorical. | 中 | SR001, SR002, SR003 |
| CR019 | Risks evidence is strongest where public sources and official product or partner materials point in the same direction on Samsung concentration is the clearest structural exposure. | 中 | SR001, SR002, SR003 |
| CR020 | Risks evidence is strongest where public sources and official product or partner materials point in the same direction on export controls and geopolitics can affect both supply and demand. | 中 | SR001, SR002, SR003 |
| CR021 | Risks evidence is strongest where public sources and official product or partner materials point in the same direction on advanced-node transitions raise execution risk. | 中 | SR001, SR002, SR003 |
| CR022 | Risks evidence is strongest where public sources and official product or partner materials point in the same direction on partner-led growth creates concentration surfaces. | 中 | SR001, SR002, SR003 |
| CR023 | Risks evidence is strongest where public sources and official product or partner materials point in the same direction on founder centrality and executive turnover raise people risk. | 中 | SR001, SR002, SR003 |
| CR024 | Risks evidence is strongest where public sources and official product or partner materials point in the same direction on mission-critical use cases need stronger quality proof. | 中 | SR001, SR002, SR003 |
| CR025 | Risks evidence is strongest where public sources and official product or partner materials point in the same direction on funding progress, DX-M2 timing, and repeat production customers are key monitors. | 中 | SR001, SR002, SR003 |
| CR026 | Risks evidence is strongest where public sources and official product or partner materials point in the same direction on diversification and better disclosure are the main mitigants. | 中 | SR001, SR002, SR003 |
| CR027 | Risks evidence is strongest where public sources and official product or partner materials point in the same direction on kill triggers are measurable rather than speculative. | 中 | SR001, SR002, SR003 |
| CR028 | The current risks read for DEEPX remains constructive but cautious because missing disclosure could still change the conclusion. | 中 | SR001, SR002, SR003 |
| CR029 | The valuation, market, product, and customer narrative is more credible when read as a physical-AI commercialization story than as a pure AI hype story. | 中 | SR001, SR002, SR003 |
| CR030 | Public evidence quality is materially better on product and partnerships than on financial durability and governance. | 中 | SR001, SR002, SR003 |
| CR031 | Additional diligence could still change the current risks conclusion because key disclosures remain private. | 低 | SR001 |
| CR032 | Additional diligence could still change the current risks conclusion because key disclosures remain private. | 低 | SR001 |
| CR033 | Additional diligence could still change the current risks conclusion because key disclosures remain private. | 低 | SR001 |
| CR034 | Additional diligence could still change the current risks conclusion because key disclosures remain private. | 低 | SR001 |
| CR035 | Additional diligence could still change the current risks conclusion because key disclosures remain private. | 低 | SR001 |
| CR036 | Additional diligence could still change the current risks conclusion because key disclosures remain private. | 低 | SR001 |
| CR037 | Additional diligence could still change the current risks conclusion because key disclosures remain private. | 低 | SR001 |
| CR038 | Additional diligence could still change the current risks conclusion because key disclosures remain private. | 低 | SR001 |
| CR039 | Additional diligence could still change the current risks conclusion because key disclosures remain private. | 低 | SR001 |
| CR040 | Additional diligence could still change the current risks conclusion because key disclosures remain private. | 低 | SR001 |
| CV001 | The company deserves serious diligence but not a conviction buy at the current public mark. | 中 | SV029, SV006, SV002 |
| CV002 | The clearest valuation anchor is the $2.2B first-close Series D. | 中 | SV001, SV002 |
| CV003 | Commercial proof is real but still early relative to the price. | 中 | SV006, SV007, SV001 |
| CV004 | Rebellions is the closest regional private valuation reference. | 中 | SV011, SV013, SV014 |
| CV005 | Public incumbents help frame ceiling value but are not direct startup comps. | 中 | SV024, SV025, SV026 |
| CV006 | The bull case depends on conversion plus DX-M2 execution. | 中 | SV030, SV007 |
| CV007 | The bear case is that financing optionality is being priced ahead of durable economics. | 中 | SV003, SV002 |
| CV008 | Audited revenue quality and repeat customers are the main upgrade triggers. | 中 | SV006, SV004 |
| CV009 | A funding stall or DX-M2 slip would likely compress valuation quickly. | 中 | SV003 |
| CV010 | Current disclosure does not support a price-insensitive buy. | 中 | SV004, SV003 |
| CV011 | The right stance is evidence-sensitive, not anti-company. | 中 | SV029 |
| CV012 | Customer concentration could cap upside even if technology execution holds. | 中 | SV001 |
| CV013 | DEEPX looks more mature than a typical AI-chip startup but earlier than the price implies. | 中 | SV029, SV003 |
| CV014 | Better evidence or a better price could improve the call. | 中 | SV006, SV007 |
| CV015 | Recommendation today is research-more. | 中 | SV002, SV006, SV003 |
| CV016 | Valuation evidence is strongest where public sources and official product or partner materials point in the same direction on real traction makes the company investable to study. | 中 | SV001, SV002, SV003 |
| CV017 | Valuation evidence is strongest where public sources and official product or partner materials point in the same direction on public disclosure still lags the current price. | 中 | SV001, SV002, SV003 |
| CV018 | Valuation evidence is strongest where public sources and official product or partner materials point in the same direction on the right call today is research-more. | 中 | SV001, SV002, SV003 |
| CV019 | Valuation evidence is strongest where public sources and official product or partner materials point in the same direction on Series D first close sets the valuation anchor. | 中 | SV001, SV002, SV003 |
| CV020 | Valuation evidence is strongest where public sources and official product or partner materials point in the same direction on the wider round and future IPO remain moving parts. | 中 | SV001, SV002, SV003 |
| CV021 | Valuation evidence is strongest where public sources and official product or partner materials point in the same direction on price support depends on better revenue-quality evidence. | 中 | SV001, SV002, SV003 |
| CV022 | Valuation evidence is strongest where public sources and official product or partner materials point in the same direction on Rebellions is the closest regional private comparison. | 中 | SV001, SV002, SV003 |
| CV023 | Valuation evidence is strongest where public sources and official product or partner materials point in the same direction on public incumbents are ceiling comps not direct matches. | 中 | SV001, SV002, SV003 |
| CV024 | Valuation evidence is strongest where public sources and official product or partner materials point in the same direction on outcomes hinge on conversion and roadmap execution. | 中 | SV001, SV002, SV003 |
| CV025 | Valuation evidence is strongest where public sources and official product or partner materials point in the same direction on audited financials, retention, concentration, and roadmap proof remain gating asks. | 中 | SV001, SV002, SV003 |
| CV026 | Valuation evidence is strongest where public sources and official product or partner materials point in the same direction on better evidence or a better price could improve the call. | 中 | SV001, SV002, SV003 |
| CV027 | Valuation evidence is strongest where public sources and official product or partner materials point in the same direction on current stance is evidence-sensitive, not anti-company. | 中 | SV001, SV002, SV003 |
| CV028 | The current valuation read for DEEPX remains constructive but cautious because missing disclosure could still change the conclusion. | 中 | SV001, SV002, SV003 |
| CV029 | The valuation, market, product, and customer narrative is more credible when read as a physical-AI commercialization story than as a pure AI hype story. | 中 | SV001, SV002, SV003 |
| CV030 | Public evidence quality is materially better on product and partnerships than on financial durability and governance. | 中 | SV001, SV002, SV003 |
| CV031 | Additional diligence could still change the current valuation conclusion because key disclosures remain private. | 低 | SV001 |
| CV032 | Additional diligence could still change the current valuation conclusion because key disclosures remain private. | 低 | SV001 |
| CV033 | Additional diligence could still change the current valuation conclusion because key disclosures remain private. | 低 | SV001 |
| CV034 | Additional diligence could still change the current valuation conclusion because key disclosures remain private. | 低 | SV001 |
| CV035 | Additional diligence could still change the current valuation conclusion because key disclosures remain private. | 低 | SV001 |
| CV036 | Additional diligence could still change the current valuation conclusion because key disclosures remain private. | 低 | SV001 |
| CV037 | Additional diligence could still change the current valuation conclusion because key disclosures remain private. | 低 | SV001 |
| CV038 | Additional diligence could still change the current valuation conclusion because key disclosures remain private. | 低 | SV001 |
| CV039 | Additional diligence could still change the current valuation conclusion because key disclosures remain private. | 低 | SV001 |
| CV040 | Additional diligence could still change the current valuation conclusion because key disclosures remain private. | 低 | SV001 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | PitchBook | Pitchbook | |
| SO002 | briefs.co | Briefs | |
| SO003 | Electronics Weekly | Ew | |
| SO004 | EE Times | Eetimes | |
| SO005 | The Investor | Theinvestor | |
| SO006 | The Elec | Thelec | |
| SO007 | Seoul Economic Daily | Sedaily | |
| SO008 | DEEPX | Ourstory | |
| SO009 | DEEPX | Contact | |
| SO010 | DEEPX | Notice | |
| SO011 | DEEPX | Dxm1 | |
| SO012 | DEEPX | Validations | |
| SO013 | DEEPX | Aaeon | |
| SO014 | DEEPX | Cesm | |
| SO015 | DEEPX | Hyundai | |
| SO016 | DEEPX | Ramp | |
| SO017 | DEEPX | Ushead | |
| SO018 | PR Newswire | Praaeon | |
| SO019 | Electronics Media | Emaaeon | |
| SO020 | DigiKey | Digikey | |
| SO021 | Hyundai Motor Group | Hyundainews | |
| SO022 | DEEPX | News | |
| SO023 | DEEPX | Company | |
| SO024 | DEEPX | Trust | |
| SO025 | DEEPX | Home | |
| SM001 | Axis Intelligence | Axis | |
| SM002 | Future Markets | Future | |
| SM003 | Fortune Business Insights | Fortune | |
| SM004 | ABI Research | Abi | |
| SM005 | IDC | Idc | |
| SM006 | Siemens / Wevolver | Siemens | |
| SM007 | DEEPX | Edge | |
| SM008 | DEEPX | Factory | |
| SM009 | DEEPX | City | |
| SM010 | DEEPX | Vms | |
| SM011 | DEEPX | Robotics | |
| SM012 | DEEPX | Industrial | |
| SM013 | DEEPX | Camera | |
| SM014 | DEEPX | Edgeserver | |
| SM015 | DEEPX | Eworld | |
| SM016 | DEEPX | Avnet | |
| SM017 | Ultralytics | Ultrablog | |
| SM018 | DFI | Dfi | |
| SM019 | BusinessKorea | Bksanctions | |
| SM020 | DEEPX | Home | |
| SM021 | EE Times | Eetimes | |
| SM022 | The Investor | Theinvestor | |
| SM023 | DEEPX | Validations | |
| SM024 | DEEPX | Cesm | |
| SM025 | Hyundai Motor Group | Hyundainews | |
| SP001 | Hailo | Hailo | |
| SP002 | Qualcomm | Qualcomm | |
| SP003 | Google Coral | Coral | |
| SP004 | Apple | Apple | |
| SP005 | Rockchip | Rockchip | |
| SP006 | NXP | Nxp | |
| SP007 | NVIDIA | Nvidia | |
| SP008 | Ambarella | Ambarella | |
| SP009 | Rebellions | Rebellions | |
| SP010 | Intel | Intel | |
| SP011 | DEEPX | Dxm1 | |
| SP012 | DEEPX | Dxm1M | |
| SP013 | DEEPX | Dxh1V | |
| SP014 | DEEPX | Dxh1Q | |
| SP015 | The Investor | Theinvestor | |
| SP016 | DigiKey | Digikey | |
| SP017 | Ultralytics | Ultrablog | |
| SP018 | IDC | Idc | |
| SP019 | Axis Intelligence | Axis | |
| SP020 | Fortune Business Insights | Fortune | |
| SP021 | Future Markets | Future | |
| SP022 | ABI Research | Abi | |
| SP023 | DEEPX | Ramp | |
| SP024 | DEEPX | Avnet | |
| SP025 | DEEPX | Cesm | |
| SI001 | The Elec | Thelec | |
| SI002 | Seoul Economic Daily | Sedaily | |
| SI003 | AInvest | Ainvest | |
| SI004 | KM Journal | Kmjournal | |
| SI005 | BusinessKorea | Bk2Nm | |
| SI006 | BusinessKorea | Bksanctions | |
| SI007 | DEEPX | Notice | |
| SI008 | PitchBook | Pitchbook | |
| SI009 | Tracxn | Tracxn | |
| SI010 | briefs.co | Briefs | |
| SI011 | Electronics Weekly | Ew | |
| SI012 | EE Times | Eetimes | |
| SI013 | The Investor | Theinvestor | |
| SI014 | DEEPX | Dxm1 | |
| SI015 | DEEPX | Dxm1M | |
| SI016 | DigiKey | Digikey | |
| SI017 | DEEPX | Validations | |
| SI018 | DEEPX | Ramp | |
| SI019 | DEEPX | Ushead | |
| SI020 | DEEPX | Ceovision | |
| SI021 | DEEPX | Award | |
| SI022 | DEEPX Developer | Developer | |
| SI023 | GitHub | Github | |
| SI024 | Ultralytics | Ultrapartner | |
| SI025 | Wind River | Windriver | |
| SE001 | DEEPX | Dxm1 | |
| SE002 | DEEPX | Dxm1M | |
| SE003 | DEEPX | Dxh1V | |
| SE004 | DEEPX | Dxh1Q | |
| SE005 | DEEPX | Dxnn | |
| SE006 | DEEPX | Aiplayer | |
| SE007 | DEEPX Developer | Developer | |
| SE008 | GitHub | Github | |
| SE009 | DEEPX | Trust | |
| SE010 | DEEPX | Slm | |
| SE011 | DEEPX | Cesm | |
| SE012 | DEEPX | Ultra | |
| SE013 | DEEPX | Wind | |
| SE014 | Ultralytics | Ultrablog | |
| SE015 | Ultralytics | Ultrapartner | |
| SE016 | Wind River | Windriver | |
| SE017 | Embedded Artists | Ea | |
| SE018 | Embedded Artists | Eakits | |
| SE019 | DigiKey | Digikey | |
| SE020 | DFI | Dfi | |
| SE021 | DEEPX | Nvrs | |
| SE022 | DEEPX | Dfioff | |
| SE023 | BusinessKorea | Bk2Nm | |
| SE024 | Sixfab | Sixfabp | |
| SE025 | Hyundai Motor Group | Hyundainews | |
| SU001 | DEEPX | Aaeon | |
| SU002 | PR Newswire | Praaeon | |
| SU003 | DEEPX | Japan | |
| SU004 | DEEPX | Hyundai | |
| SU005 | Hyundai Motor Group | Hyundainews | |
| SU006 | DEEPX | Avnet | |
| SU007 | DEEPX | Posco | |
| SU008 | DEEPX | Dfioff | |
| SU009 | DFI | Dfi | |
| SU010 | DEEPX | Validations | |
| SU011 | KM Journal | Kmjournal | |
| SU012 | EE Times | Eetimes | |
| SU013 | DEEPX | Cesm | |
| SU014 | briefs.co | Briefs | |
| SU015 | Electronics Weekly | Ew | |
| SU016 | The Elec | Thelec | |
| SU017 | Seoul Economic Daily | Sedaily | |
| SU018 | GitHub | Github | |
| SU019 | DEEPX | Partners | |
| SU020 | DEEPX | Partcontact | |
| SU021 | DEEPX | Salesdist | |
| SU022 | DEEPX | Salessupport | |
| SU023 | DEEPX | Techsupport | |
| SU024 | Sixfab | Sixfabp | |
| SU025 | Sixfab | Sixfabh | |
| SR001 | Seoul Economic Daily | Sedaily | |
| SR002 | BusinessKorea | Bksanctions | |
| SR003 | BusinessKorea | Bk2Nm | |
| SR004 | The Elec | Thelec | |
| SR005 | DEEPX | Trust | |
| SR006 | DEEPX | Notice | |
| SR007 | DEEPX | Legalnote | |
| SR008 | DEEPX | Regtrust | |
| SR009 | DEEPX | Wind | |
| SR010 | Wind River | Windriver | |
| SR011 | DEEPX | Hyundai | |
| SR012 | Hyundai Motor Group | Hyundainews | |
| SR013 | DEEPX | Avnet | |
| SR014 | DEEPX | Ramp | |
| SR015 | DEEPX | Validations | |
| SR016 | DEEPX | Slm | |
| SR017 | DEEPX | Eworld | |
| SR018 | The Investor | Theinvestor | |
| SR019 | IDC | Idc | |
| SR020 | Siemens / Wevolver | Siemens | |
| SR021 | Fortune Business Insights | Fortune | |
| SR022 | Axis Intelligence | Axis | |
| SR023 | NVIDIA | Nvidiajet | |
| SR024 | NVIDIA | Nvidiajeta | |
| SR025 | PitchBook | Hailopb | |
| SR026 | Apple | Applene | |
| SR027 | Rockchip | Rockaiot | |
| SR028 | Ambarella | Ambacv | |
| SR029 | Ambarella | Ambacva | |
| SR030 | DEEPX | Cesf | |
| SV001 | briefs.co | Briefs | |
| SV002 | Electronics Weekly | Ew | |
| SV003 | Seoul Economic Daily | Sedaily | |
| SV004 | PitchBook | Pitchbook | |
| SV005 | Tracxn | Tracxn | |
| SV006 | The Elec | Thelec | |
| SV007 | KM Journal | Kmjournal | |
| SV008 | Axis Intelligence | Axis | |
| SV009 | Future Markets | Future | |
| SV010 | Fortune Business Insights | Fortune | |
| SV011 | PR Newswire | Rebpr | |
| SV012 | Electronics Weekly | Rebew | |
| SV013 | WOWTALE | Rebwowpre | |
| SV014 | WOWTALE | Rebwowc | |
| SV015 | DEEPX | Notice | |
| SV016 | DEEPX | Baidu | |
| SV017 | PR Newswire | Praaeon | |
| SV018 | DEEPX | Aaeon | |
| SV019 | Qualcomm | Qcomir | |
| SV020 | NXP | Nxpir | |
| SV021 | Ambarella | Ambir | |
| SV022 | NVIDIA | Nvir | |
| SV023 | Hailo | Hailo | |
| SV024 | Qualcomm | Qualcomm | |
| SV025 | NXP | Nxp | |
| SV026 | Ambarella | Ambarella | |
| SV027 | NVIDIA | Nvidia | |
| SV028 | Rebellions | Rebellions | |
| SV029 | EE Times | Eetimes | |
| SV030 | DEEPX | Cesm |