初创公司尽调
尽调报告 Artificial Intelligence / AI Safety Pre-product research lab 2026-10-06

Safe Superintelligence Inc.

能获得巨额资本支持、但公开运营证据寥寥的私营 AI 安全实验室

SSI 是一项高确信度、尚无收入的 AI 安全押注,估值依靠创始人的选择权和算力获取能力,而不是可观察到的商业进展。

覆盖范围与披露说明

公开证据确认其采取纯研究路线、2024 年融资 $1B,以及后续于 2026 年与 NVIDIA 合作;但后续融资、估值、员工数、治理、收入和客户指标仍部分来自报道或尚未披露。

封面要素

成立时间 01
2024-06-19 [CO001]
已知最新估值 02
32000 USDm [CO006]
2024 年已确认融资 03
1000 USDm [CO005]
员工数估计 04
50 employees [CO011]
公开收入 05
— [CO012]

公司概况

Safe Superintelligence Inc. 是一家私营、尚未推出产品的 AI 安全研究实验室,由 Ilya Sutskever、Daniel Gross 和 Daniel Levy 于 2024 年创立。公司公开表示只专注于构建安全的超级智能,并确认在 Palo Alto 和 Tel Aviv 设有办公室;公开资料尚未显示其产品、收入或客户基础。公开证据还显示,公司确认 2024 年融资 $1 billion,后续据报以 $32 billion 估值融资,并于 2026 年与 NVIDIA 建立旨在扩大算力获取的合作。

官网
ssi.inc
成立时间
2024-06-19
创始人
Ilya Sutskever, Daniel Gross, Daniel Levy
创立地点
Palo Alto and Tel Aviv
总部
Palo Alto, CA, USA
产品
唯一规划中的产品是安全的超级智能;目前没有公开模型、API 或商业软件界面。
客户
没有公开客户;未来买家尚未披露。
商业模式
私人资本和战略算力支持为纯研究实验室提供资金;未披露公开收入模式。
阶段
Pre-product research lab
融资情况
已确认 2024 年融资 $1B;后续 2025 年融资和 2026 年 NVIDIA 合作均来自报道或尚未披露,而非公司确认。
[CO001, CO002, CO004, CO005, CO006, CO007, CO011, CO012]

执行摘要

主要优势

  • 单一聚焦的 AI 安全方向,有顶尖创始人履历和充足融资能力支撑
  • 已确认及有报道的主要算力供应商战略关系
  • 如果实验室取得真正的技术突破,上行空间可能极大

主要风险

  • 没有公开产品、论文、基准测试或收入流可供评估
  • 高度依赖创始人执行力,且私人治理信息披露有限
  • 估值和累计融资额部分来自报道,无法独立审计

未决问题

  • SSI 具体在开展哪些技术研究?用什么可量化的里程碑定义进展?
  • 股权结构表中的确切融资条款、所有权利和治理保护是什么?
  • 在 2026 年扩充算力后,SSI 当前的烧钱速度和剩余资金可支撑期限是多少?
  • SSI 是否设有正式的外部安全治理或使命锁定机制?

目录

Chapter 01

01公司概况

1.1 身份、使命与运营阶段

Safe Superintelligence Inc. 于 2024 年 6 月由 Ilya Sutskever、Daniel Gross 和 Daniel Levy 创立 [CO001]。公司自称是一家美国公司,并确认在 Palo Alto 和 Tel Aviv 设有办公室;第三方资料则将 Palo Alto 列为总部 [CO002]。其战略范围异常狭窄:只有一个目标和一个规划中的产品——安全的超级智能;安全与能力同步推进,而不是先做一系列过渡性的商业产品 [CO003]。 公开证据仍将 SSI 定位在纯研究、尚未推出产品的阶段。截至 2026 年 8 月和 9 月,独立评审未找到公开模型、API、产品界面、论文或技术博客文章 [CO004]。这并不证明内部没有研究,只说明外部无法评估产品、基准测试或已发布的方法。现有证据也没有公司发布的收入或客户数指标,因此二者都是尽调缺口,而不是零值 [CO012]。因此,SSI 更像一个由资本支持的研究项目,而不是已经拥有可观察商业牵引力的软件运营商。

快照 KPI 表
指标数值或状态证据日期置信度缺口或限定
身份美国私营 AI 研究公司;成立于 2024 年 6 月2026-09中现有证据未包含确切法律实体和注册记录
地点公司确认在 Palo Alto 和 Tel Aviv 设有办公室;据报 Palo Alto 是总部2026-09中公司描述了办公室,但未在主页上将任何地点标为总部
产品与阶段纯研究;唯一规划产品是安全的超级智能;未发现公开模型或 API2026-09中内部研究进展和发布标准未披露
融资与估值2024 年公司确认 $1B;2025 年据报以 $32B 估值融资约 $2B;2026 年 NVIDIA 金额由两家公司未披露2026-09中报道金额不等于已确认可用现金
收入与客户在现有证据中未公开披露2026-10低不要因没有公开产品就推断收入为零或客户为零
员工数估计相互冲突:13 人、约 20–30 人和约 50 人2026-09低未找到官方当前员工数或办公室分布

日期表示每行所采用的最新留存证据;据报道的数字仍会标注来源,未获支持的运营数量视为证据缺口。

[CO001, CO002, CO003, CO004, CO005, CO006, CO007, CO008, CO011, CO012]

1.2 领导层集中与治理可见度

SSI 的创始团队结合了 Sutskever 的前沿模型研究背景、Gross 的运营和投资经验,以及 Levy 的大规模优化经验 [CO001]。领导结构在 2025 年发生变化:SSI 表示 Gross 在 6 月 29 日之后已不再任职,Sutskever 出任 CEO,Levy 出任总裁,技术团队继续向 Sutskever 汇报 [CO009, CO010]。如今两位留下的创始人构成可见的领导核心,科学和高管责任也集中在 Sutskever 身上。 现有公开资料没有披露董事会构成、投资人保护权利、独立安全委员会或使命锁定机制 [CO013]。资料也没有发现可供外部评审对照阈值、升级规则或监督机制的公开前沿安全框架 [CO014]。这些缺失不应被解读为内部不存在控制措施,而是形成了验证问题:投资人需要在得出 SSI 的使命得到制度性保护之前,取得保密治理文件、董事会会议记录和安全评审流程。Gross 离任后,这一要求更重要,因为最初担任 CEO 的创始人已经离开。

领导层与创始人表
人员SSI 当前或最后职务指定证据中的背景职能覆盖关键人物依赖
Ilya Sutskever联合创始人兼 CEO前 OpenAI 联合创始人兼首席科学家科学方向、执行权力、技术汇报线关键:投资逻辑与技术方向都与其高度绑定
Daniel Levy联合创始人兼总裁前 OpenAI 优化研究员;斯坦福大学计算机科学博士大规模训练与运营连续性高:公开确认的在任创始人仅剩两人之一
Daniel Gross联合创始人兼前 CEO;于 2025 年 6 月 29 日离职前 Apple 机器学习负责人、Cue 创始人及投资人初期高管与融资覆盖已兑现的交接风险:其运营职能已离开公司

背景和职务描述仅限指定的官方及独立资料;技术团队之外的现任董事席位和汇报线尚未公开。

[CO001, CO009, CO010, CO013, CO014]

1.3 没有公开运营基线的资本形成

最清晰的融资锚点是 SSI 自己在 2024 年 9 月发布的更新:公司从 NFDG、Andreessen Horowitz、Sequoia、DST Global 和 SV Angel 募集了 $1 billion [CO005]。独立来源随后报道,2025 年 Greenoaks 领投了一轮约 $2 billion 的融资,估值为 $32 billion,但 SSI 没有在现有官方证据中确认该轮融资或估值 [CO006]。2026 年 7 月,SSI 确认与 NVIDIA 建立战略合作,目标是将算力提高十倍;现有证据没有披露两家公司的投资金额 [CO007]。 二手来源将约 $5 billion 的投资归因于 NVIDIA,但保留的最强综合结论明确将该金额标为未经确认 [CO008]。因此,本章不把报道中的融资换算成已确认的可用现金,也不把 $8 billion 总额当作既定事实。规模同样不确定:现有估计从 13 名员工、约 20–30 名到约 50 名不等,没有官方当前人数 [CO011]。收入、客户、烧钱速度、剩余现金和现金跑道均无证据支持。因此,资本化情况主要通过融资公告和供应商承诺体现,而不是财务报表或运营 KPI。

利益相关方或投资人地图
利益相关方角色控制权或经济重要性证据状态尽调要点
NFDG2024 年投资人公司确认的 $1B 轮次中具名参与者公司确认参与确认所有权、董事会权利,以及 Gross 离职后的任何变化
Andreessen Horowitz2024 年投资人公司确认的 $1B 轮次中具名参与者公司确认参与确认按比例认购权、信息权和当前所有权
Sequoia Capital2024 年投资人公司确认的 $1B 轮次中具名参与者公司确认参与确认治理权利和投资期限
Greenoaks Capital据报道领投 2025 年轮次与据报道约 $2B 的轮次及 $32B 估值相关第三方报道;SSI 未在此确认获取融资文件,并确认估值的投前 / 投后口径
NVIDIA2026 年战略合作伙伴兼投资人计算供应商及投资人,参与旨在将算力扩大十倍的合作合作已确认;公司未披露投资金额单独确认现金投资、算力承诺、定价、排他性和治理权利

这是与决策最相关的已披露或据报道的利益相关方样本,并非完整股权结构表;经济权利和持股比例仍属私人信息。

[CO005, CO006, CO007, CO008, CO013, CO015]

1.4 里程碑与尽调含义

SSI 的公开里程碑显示其迅速获得资本和算力,但可供外部检验的产出有限。时间线从 2024 年 6 月成立、2024 年 9 月融资开始,随后是据报于 2025 年 4 月建立的 Google Cloud TPU 关系、2025 年 6–7 月的创始人交接,以及 2026 年 7 月与 NVIDIA 的合作 [CO001, CO005, CO009, CO010, CO015]。这些事件证明了融资能力、供应商获取能力以及 Gross 离任后的连续性,但不能独立证明技术表现或商业化。 最明确的反向信号不是公司错过了自己的承诺。一篇归因于投资人的报道指向 2026 年 8 月发布模型,但 SSI 从未确认该日期,评审截至 9 月也未发现发布 [CO016]。这应被视为未兑现的外部预期,而非企业承诺被打破。因此,下一步关键里程碑应是证据事件:模型或研究论文、可审计的安全方法、经核实的当前团队和治理图谱,以及将已募集资金与剩余现金分开的财务披露。在这些证据出现前,关键人物集中和信息不对称将主导公司概况。

FO001: 公司里程碑时间线

公开里程碑区分公司确认的事件、报道中的融资,以及未经确认的发布预期。

日期是宣布日期或来源审阅日期;2025 年融资与 2026 年 8 月发布窗口仍来自第三方报道,而非公司确认。

SSI 宣布成立 [CO001, CO003]

$1B 融资获确认 [CO005]

Google Cloud 算力关系见报道 [CO015]

$2B 轮融资与 $32B 估值见报道 [CO006]

Gross 离开 SSI [CO010]

领导层整合 [CO009]

NVIDIA 宣布合作 [CO007]

未找到公开产品 [CO004]

未确认的发布窗口已过 [CO016]

[CO001, CO003, CO004, CO005, CO006, CO007, CO009, CO010, CO015, CO016]

1.5 展示项

Chapter 02

02市场分析

2.1 市场边界、范围内能力与规模测算排除项

Safe Superintelligence Inc.(SSI)通过严格的结构性排除界定市场范围:公司只追求一个目标——构建安全的超级智能,并明确拒绝参与近期商业软件市场。与依靠消费者聊天机器人、开发者 API 和企业智能副驾交叉补贴研究的传统基础模型公司不同,SSI 将中间产品周期视为运营干扰。因此,客户支持自动化、企业搜索和编程助手等传统企业软件类别都完全不在公司的可服务范围内。其范围内领域仅限于基础通用人工智能和自主对齐架构。实验室已从 Google Cloud TPU 和下一代 Nvidia Vera Rubin 系统获得算力,但没有向第三方提供商业算力或托管服务。此外,SSI 在 arXiv 上没有公开研究论文,没有模型卡,也没有 Anthropic 或 OpenAI 那样的版本化前沿安全框架,形成了一条封闭的运营边界:研究进展完全在内部评估,而不是通过公开商业或学术基准衡量。[CM001, CM007, CM009, CM011]

Safe Superintelligence 的市场边界与范围排除
类别纳入范围排除范围目标买方 / 付款方战略理由
基础研究实验室长期安全超级智能研发与核心算法对齐企业工作流 SaaS、任务自动化插件、消费级聊天机器人战略股权和算力合作伙伴(Nvidia、Alphabet)让研究人员不受商业管理开销和产品交付周期牵制
商业 API 服务在超级智能里程碑完成前不提供任何服务REST API、按 token 计费的端点、模型微调平台企业开发者(当前排除)避免过早商业化和功能竞赛
前沿安全验证能力与安全并行协同开发公开的前沿安全框架、已发布预印本、外部红队测试内部研究团队、未来的主权评估机构保护专有架构洞见,避免追逐虚荣指标
算力基础设施接入多年期集群使用(TPU、Vera Rubin)商品化现货 GPU 租赁、商用算力转售通过股权和战略协议合作的基础设施供应商锁定前沿扩展所需的专用大规模集群

边界标准源自 SSI 官方创始使命声明和第三方实验室追踪;企业 SaaS 产品及开发者 API 明确不在当前运营范围内。

[CM001, CM007, CM009, CM011]

2.2 规模测算视角、算力支出基准与 TAM 约束

用标准可服务市场框架分析 Safe Superintelligence Inc.,需要把理论经济潜力与已经实现的市场需求分开。在理论上限情景下,全面自动化认知劳动代表着数十万亿美元的可寻址宏观经济影响,覆盖全球知识工作。但按当前商业现实衡量,SSI 的可服务、可获取市场支出目前为零,因为实验室没有商业合同、API 或计费基础设施。相比之下,商业前沿同行迅速扩大企业软件渗透:截至 2026 年末,Anthropic 的年化收入运行率超过 $65 billion,OpenAI 约为 $40 billion,说明市场对运营型 AI 工作流的需求极大。同时,私人资本市场给 SSI 的估值为 $32 billion——这一估值最初来自 Greenoaks Capital 领投的 2025 年 4 月融资,并得到后续 Nvidia 算力投资的支持——约为 Thinking Machines Lab($13.5 billion 二级市场标记)的两点四倍、World Labs($5 billion)的六倍。按其约 13 至 50 人的精简团队计算,这一估值相当于每名员工超过 $600 million,说明 SSI 的定价依据是 AGI 突破的纯看涨期权,而不是贴现现金流或近期企业软件支出。[CM002, CM003, CM004, CM005, CM006, CM007, CM010, CM015]

市场规模测算视角与可触达 AI 支出基准
测算视角参考实体 / 领域基准指标值市场时点方法 / 来源市场规模限制 / 风险
近期已实现收入Safe Superintelligence Inc.$0 已披露商业收入2026Vorp Labs 和 Value Add VC 追踪的公司直接状态纯研究阶段无法采用传统收入倍数
前沿商业同业Anthropic 与 OpenAI$65B 与 $40B 收入运行率2026据报道的年度化商业及 API 收入运行率商业实验室通过企业 SaaS 工作流变现,而 SSI 不覆盖这些场景
私营实验室资本化SSI、Thinking Machines、World Labs$32B、$13.5B、$5B 估值2025-2026风投轮次和老股交易估值(Greenoaks、Forge、Bloomberg)反映投资人对创始人人才的稀缺性定价,而非折现现金流
累计部署资本SSI 专用算力与股权$8B 累计融资 / 承诺2026融资轮次及 Nvidia 战略算力合作将资本承诺和硬件分配与运营预算混为一谈

基准数据综合自 Value Add VC、Vorp Labs 和 GetLatka 数据库;私营实验室的估值数字代表据报道的风投轮次和经纪商老股交易指示价。

[CM002, CM003, CM004, CM005, CM006]
FM001: 超级智能的约束式市场规模金字塔

六层规模视角,从推测性的全球认知劳动力替代,逐层下探至 SSI 当前已经实现的可寻址商业收入。

各层宽度仅表示层级,不代表数值比例。

宏观经济数字代表宽泛的行业规模视角;运行率收入与私营实验室估值估计基于截至 2026 年末的第三方市场报道。

[CM002, CM004, CM005, CM006, CM007, CM010, CM015]

2.3 买方类型、战略支持与采购路径

Safe Superintelligence Inc. 没有销售商业产品,因此当前资本化主要由战略企业支持者和风险投资人推动,而非传统软件采购部门。迄今主要提供资本和资源的是超大规模硬件及基础设施巨头,尤其是 Nvidia:据报道,Nvidia 于 2026 年 7 月在提供 Vera Rubin 算力的同时承诺了一笔未披露的数十亿美元投资,金额为 $5 billion;Alphabet 则在参与风险投资的同时提供了 Google Cloud TPU 算力。这些战略伙伴将 SSI 视为高确信度对冲工具和专业研究载体,可用于压力测试下一代硬件架构。若未来实现安全超级智能,买方群体将与标准 SaaS 买方明显不同:目标买家会包括寻求地缘政治技术对等的主权防务机构、统筹全自主运营单元的跨国公司,以及系统性研究联盟。年度 SaaS 许可或按 token 消费计价等标准企业软件采购路径不会主导采用;采购更可能接近战略合资、主权防务收购或高度监管的关键任务基础设施特许经营。[CM001, CM008, CM009, CM013, CM014]

买方画像与采购采用路径
细分市场潜在买方主要终端用户付款实体主要工作流预算归属采用触发因素
战略云与芯片供应商超大规模平台(Nvidia、Alphabet)AI 基础设施工程师企业风险投资与研发拨款下一代架构训练与工作负载协同设计CEO / CTO 战略储备研究突破已获验证,足以支撑 10 倍算力规模
自主企业采用者全球财富 500 强管理层自主系统运营者企业战略转型端到端认知工作流替代董事会层面的创新与资本预算可验证安全的自主推理能力超过人类专家基准
主权与国防机构国家安全与主权 AI 项目战略防务分析师国家技术与国防预算关键国家基础设施与战略对齐国防部 / 主权财富基金地缘政治 AI 对等能力,以及经认证的防止非对齐突破机制
前沿研究联盟高级 AI 对齐研究机构超级对齐研究人员慈善与安全捐赠基金协同多智能体对齐测试研究资助委员会正式发布可审计的安全与对齐架构

潜在客户画像反映安全超级智能的长期潜在部署渠道;截至 2026 年末不存在商业客户关系。

[CM001, CM008, CM009, CM013]

2.4 结构性驱动、安全验证与商业化瓶颈

支撑 Safe Superintelligence Inc. 的投资逻辑来自强劲的结构性驱动因素:预训练 Transformer 的传统扩展规律快速耗尽,行业对真正推理突破的需求上升,以及机构逐渐认识到必须同步解决对齐与能力问题。然而,公司面临严重的采用瓶颈和执行约束。其中首要问题是完全缺乏外部验证:一则暗示 2026 年 8 月首次发布模型的投资人传闻最终没有公开产物,市场观察者仍看不到公开权重、演示或模型卡。此外,联合创始人兼首任 CEO Daniel Gross 于 2025 年中离开并负责 Meta Superintelligence Labs,此前 Mark Zuckerberg 的数十亿美元收购尝试遭到拒绝,凸显资金充足的商业集团带来的激烈人才竞争。实验室在 Palo Alto 和 Tel Aviv 仅有 13 至 50 名员工,能隔离运营官僚主义,却也让它暴露于关键人物依赖和算力谈判风险。在没有可验证里程碑的情况下,实验室的市场信誉完全依赖 Ilya Sutskever 的个人技术声誉。[CM010, CM011, CM012, CM013, CM015]

2.5 展示项

Chapter 03

03竞争格局

3.1 竞争集合与买方替代方案

两个独立目录将 OpenAI、Anthropic、Google DeepMind、xAI 和 Mistral AI 列入 SSI 的直接前沿实验室对比集合,另一个目录还识别出更长的新实验室尾部 [CP001]。这一集合在战略上并不均衡:SSI 被描述为专注安全的自研实验室,OpenAI 被描述为覆盖更广泛的对齐、红队测试、评估和内容安全活动,而 Thinking Machines 和 World Labs 等较新的研究实验室在现有证据中已经公开命名交付产物 [CP002, CP003, CP012]。内部自建和等待更成熟供应商都是有效的买方替代方案,但现有资料没有量化任一路径,因此将其列为明确未知项,而不是虚构的竞争者 [CP015]。

竞争者画像表
替代方案类别规模 / 融资证据目标细分市场有证据支持的差异化指定证据的局限
Safe Superintelligence Inc.专有前沿安全实验室一个来源确认或报道约 $3B 股权融资;更高总额存在争议未披露商业买方细分市场唯一明确目标是安全超级智能未发现公开模型、API、定价页、基准或安全框架
OpenAI商业前沿 AI 公司对比页面列示融资额 $13B+指定证据未说明公开重点更广,覆盖对齐、红队测试、评估和内容安全指定证据无法确立当前定价、产品范围或融资可比性
Anthropic前沿 AI 安全公司指定证据未确立指定证据未确立被列为直接同业,且已发布前沿安全框架没有指定来源提供当前产品、定价或规模画像
Google DeepMind超大规模云厂商的 AI 研究实验室指定证据未确立指定证据未确立被列为直接同业,且已发布前沿安全框架没有指定来源提供当前产品、定价或规模画像
xAI前沿 AI 公司指定证据未确立指定证据未确立被列为直接同业,且已发布前沿安全框架没有指定来源提供当前产品、定价或规模画像
Mistral AI开放权重模型开发商指定证据未确立指定证据未确立Distill Intelligence 将其列为 SSI 竞争者指定来源不支持更深入的能力或商业比较
Thinking Machines Lab新兴前沿研究实验室据一份分析师来源,融资 $2B,老股交易估值约 $13.5B根据具名 Tinker API 推断为开发者微调用户据报道 Tinker 微调 API 已于 2025 年 10 月上线老股估值不是定价的新股轮次;定价尚未确立
World Labs空间 AI 研究实验室据一份分析师来源,融资 $1.23B,估值约 $5B根据具名 Marble 产品推断为空间计算用户据报道 Marble 空间产品已上线对比仅由一个指定来源提供;定价尚未确立
内部自建前沿 AI买方替代方案未知具备研究和算力能力的组织可能掌控数据和部署指定证据未量化或刻画
维持现状 / 延后采用不购买替代方案未知无法验证安全性或 ROI 的买方避免近期切换和部署风险没有指定来源量化延后采用或其成本

竞争格局仅基于指定资料池;未知单元格为有意保留,融资口径存在冲突,老股交易估值不视为定价轮次估值。

[CP001, CP002, CP003, CP004, CP005, CP008, CP009, CP012, CP015]

3.2 能力证据与信任立场

最影响决策的对比不是基准排名,而是外部可检查证据的数量。一份检查 SSI 官方渠道和公开模型目录的资料截至 2026 年 8 月 15 日未发现已发布模型、API、产品、开发者文档或定价页面 [CP004]。同一评审没有发现 SSI 发布的前沿安全框架,却找到了 Anthropic、OpenAI、Google DeepMind、Meta 和 xAI 的已发布框架 [CP005]。SSI 对广义对齐的关注是可见的,但落实情况不可见:MindStudio 认为,基于有限公开表述的详细架构说法仍属推测 [CP006]。因此,矩阵和图表衡量的是现有证据中的披露情况,而不是底层模型质量或安全性。

功能与能力证据矩阵
尽调标准SSIOpenAI其他具名前沿同业解读
公开模型或 API截至 2026-08-15 未发现指定证据未评估Thinking Machines:据报道 Tinker 已上线;World Labs:据报道 Marble 已上线资料池支持 SSI 存在披露缺口,但不足以形成完整技术排名
开发者文档或定价页截至 2026-08-15 未发现指定证据未评估指定证据未评估无法进行可靠的同业价格比较
已发布研究或技术博客截至 2026-08-15 未发现指定证据未评估指定证据未评估SSI 所称的研究优势无法由外部复现
已发布前沿安全框架未发现已发现已发布框架已发现 Anthropic、Google DeepMind、Meta 和 xAI 的框架框架存在不等于实施质量得到证明
公开安全重点对齐与负责任 AI对齐、红队测试、负责任 AI、评估和内容安全指定证据未持续刻画OpenAI 的公开范围描述更广
技术方法细节公开细节被描述为推测指定证据未评估指定证据未评估无法支持架构层面的比较

本矩阵记录指定来源能够确立的内容,并非断言未评估的同业不具备相应能力。

[CP002, CP003, CP004, CP005, CP006, CP011, CP012]
FP001: 指定证据可见性地图

指定来源池中可见的商业产物与安全流程证据的序数地图,不代表底层能力排名。

条目商业产物证据深度(1-5) (1–5)安全流程证据深度(1-5) (1–5)说明
1. SSI 1 1 截至所审阅日期,来源中未发现公开产品或前沿安全框架。
2. OpenAI 2 5 指定来源池识别出一个框架和宽泛的安全范围,但没有提供产品或定价概况。
3. Anthropic 1 5 指定来源池识别出一个框架;未评估商业证据。
4. Google DeepMind 1 5 指定来源池识别出一个框架;未评估商业证据。
5. Thinking Machines Lab 5 1 指定来源池提到已发布的 Tinker API;未评估安全流程证据。
6. World Labs 5 1 指定来源池提到已发布的 Marble 产品;未评估安全流程证据。

分析师指定的 1-5 证据深度分数仅总结该来源池。分数为 1 表示指定证据未建立相关事实,不代表能力或流程不存在。

[CP004, CP005, CP012, CP016]

3.3 定价、分发与供应获取

SSI 尚无法按已实现价格、合同期限、折扣或客户切换成本进行比较。现有证据未见公开定价界面或商业套餐,竞争对比页面也没有提供可靠的同行价格卡片 [CP011]。算力获取更为可见:资料描述了先使用 Google Cloud TPU、后与 NVIDIA 合作获得 Vera Rubin 访问权,但不能据此声称任一供应关系具有排他性,也不能声称由此形成专有分发能力 [CP010]。公开员工数估计同样存在重大差异,从约 13 人到约 50 人不等,来源时间和方法各异,因此本章不把任何一个数字换算成精确的当前运营规模 [CP007]。

定价与产品包装比较
替代方案公开价格 / 单位合同或产品包折扣 / 实际价格尽调含义
SSI未发现公开定价页未发现模型、API 或产品包未知没有买方切换成本或支付意愿证据
OpenAI指定证据未确立指定证据未确立未知比较前获取最新的一手价目表
Anthropic指定证据未确立指定证据未确立未知比较前获取最新的一手价目表
Thinking Machines Lab指定证据未确立据报道 Tinker 微调 API 已上线未知具名产品提供了验证入口,但经济性仍未知
内部自建 / 维持现状私有成本结构内部项目或延后购买未知访谈目标买方,量化算力、人才、集成和延误成本

指定来源不包含当前同业一手定价或合同条款;所有未获支持的字段均标记为未知,而非估计值。

[CP002, CP004, CP011, CP012, CP015]

3.4 护城河持久性与替代风险

SSI 当前的护城河更应理解为资本、算力和稀缺研究人才的获取能力,而不是产品锁定效应。即便资本基线也存在争议:Vorp Labs 将约 $3 billion 的已确认及已报道股权融资与未披露的 NVIDIA 投资分开,而 Value Add VC 则在报道中的 NVIDIA $5 billion 支票基础上给出约 $8 billion 的总额 [CP008, CP009]。后者还报道 Thinking Machines 已交付 Tinker、World Labs 已交付 Marble,而 SSI 仍没有公开产物 [CP012]。截至 9 月 25 日,传闻中的 SSI 2026 年 8 月模型发布仍未出现,但该传闻从未是 SSI 的承诺 [CP013]。没有模型、基准或方法披露,研究优势和切换成本都无法验证,商品化和执行时点因此成为核心竞争风险 [CP014, CP015]。

3.5 展示项

Chapter 04

04财务情况

4.1 收入模式与商业证据

现有证据只支持 SSI 处于商业化前的研究阶段,而不是已经运营的收入模式。多个独立来源称 SSI 没有已发布产品,也未披露收入;评审的来源集合没有找到付费客户、客户合同、定价页面或活跃商业渠道 [CI001, CI002]。某个市场数据页面将其收入模式标为“订阅”,但同一页面又称 SSI 尚未商业化,并表示公司没有披露收入或财务指标;因此,该分类不能证明实际存在订阅业务 [CI003, CI004]。 现有证据中没有经验证的公开记录定义收入流、可计费单位、标价或实际价格、折扣政策、合同期限或收入确认方法 [CI005]。融资和算力获取是研究投入,不是收入。更稳妥的结论不是“收入为零”,而是收入未公开披露;截至 2026 年 9 月,现有证据也没有证明其存在商业产品。管理层尽调必须确定未来变现会采用 API 消费、许可、企业合同或其他机制,不能预设为 SaaS 模式 [CI001, CI002, CI005]。

收入来源与变现证据
潜在收入来源机制单位当前公开状态证据质量尽调要点
当前商业收入没有已验证机制未披露收入;未发现已发布产品多个独立二手来源提供收入台账、客户清单和已签合同
模型或 API 访问可能按使用量收费未发现公开 API、模型或定价页留存证据未确立提供上线计划、计费单位和单位成本模型
企业授权可能授权或私有部署未发现公开报价或合同留存证据未确立提供销售管线、合同模板、期限和支持义务
NVIDIA 关系投资加算力合作非收入单位融资和供应商准入;未披露客户收入据报道已建立合作,经济条款未披露需区分股权现金、算力额度、采购以及相互义务

截至 2026-10-06 的部分公开证据图谱;null 表示没有找到有依据的计费单位,不代表价格为零,可能机制只是尽调分类,并非公司计划。

[CI001, CI002, CI003, CI004, CI005, CI007, CI008]
FI001: 收入模型桥接:从报道中的输入到未披露的输出

公开记录能够追溯到研究资本与算力访问,但下游每一步商业转化仍未核实。

节点 / 连接节点 / 起点终点说明
节点 1 报道中的融资 [capital]各来源金额不同,且不等于当前现金。
节点 2 算力访问 [compute]据报道,NVIDIA 合作目标是实现数量级扩张。
节点 3 研究活动 [research]没有公开的成本分配或产出经济性。
节点 4 商业产品 [product]保留证据未证明有已发布产品。
节点 5 客户合同 [contract]未发现公开合同、定价或可计费单位。
节点 6 已确认收入与毛利 [revenue]没有披露收入或利润率桥接。

仅为概念性有序序列:前端节点是报道中的输入,后端节点未披露或未核实,不暗示任何转化率或货币价值。

[CI002, CI007, CI013, CI016]

4.2 GTM 证据与单位经济学缺口

SSI 的公开财务信息不足以支持销售效率分析。现有来源没有披露获客成本、销售周期、回本周期、渠道经济性、留存、利用率,或可用于计算这些指标的任何分母 [CI006]。没有公开定价页面或客户合同,也无法检验标价到实际价格的差异、折扣、最低承诺或收入集中度 [CI002, CI005]。这些是不可用的输入,而不是零值;不应拿假定已经商业化的同行基准替代它们。 毛利率和交付经济性同样无法分析。据报道,NVIDIA 关系增加了 Vera Rubin 访问权,计划将 SSI 算力扩大一个数量级,使算力成为最清晰的可观察成本驱动因素 [CI007, CI013]。但公开证据没有披露算力究竟是购买、抵扣、融资、预留还是实物出资,也没有毛利率、营运资本、资本开支或服务交付成本数据 [CI008, CI014]。在产品和计费单位形成之前,传统 SaaS 单位经济学不是表现较弱,而是在结构上尚未成熟。

定价与合同证据
输入项公开值置信度重要性尽调要求
标价与计费单位不可获得确定收入分母和产品包装提供经批准的价目表及计费单位定义
实际价格与折扣不可获得检验价格完整性和贡献利润率提供发票、折扣表和让利安排
合同期限与承诺不可获得决定订单积压、集中度和营运资金提供已签署合同、续约条款和付款计划
收入确认政策不可获得区分订单额、账单额、ARR 和已确认收入提供会计政策及递延收入滚动表

留存来源未提供这些商业输入;null 表示没有可获得的公开证据,不代表金额为零。

[CI002, CI005, CI006]

4.3 资本充足性与融资依赖

历史融资时间线属于公司概况;从财务投资判断角度,关键是这些融资对当前资源意味着什么。公开来源对此并不一致。Vorp Labs 将官方确认的 2024 年 $1 billion 融资与多家媒体报道的约 $2 billion 2025 年融资分开,并将 2026 年 NVIDIA 金额视为两家公司未披露 [CI009]。其他数据库和分析称三轮融资合计约 $8 billion,而 GetLatka 仍称一轮融资为 $2 billion [CI010, CI011]。差异反映了来源处理方式和数据陈旧程度,因此不能把单一累计总额当作经审计现金。 即使融资额正确,也不能证明可用现金,因为保留的来源没有披露当前现金、烧钱速度、受限资金、债务、项目融资、已承诺算力支出,或现金与实物对价的构成 [CI012, CI014]。据报算力扩大十倍可能提升研究能力,但也让基础设施义务成为现金跑道问题的核心 [CI007, CI013]。因此,公开证据无法计算现金跑道、确定下一轮融资触发点,或检验融资依赖是否下降 [CI015]。

单位经济与资本充足性
指标公开值置信度投资判断意义尽调要求
收入 / ARR不可获得无法检验收入质量和增长提供按月确认的收入、ARR 桥接表和客户集中度
CAC / 回本周期 / 销售周期不可获得无法计算 GTM 效率提供获客支出、队列订单额和销售周期数据
毛利率 / 服务成本不可获得无法建模利润率路径按产品单元提供算力、支持和交付成本
现金 / 月度烧钱速度 / 现金跑道不可获得融资额不等于可用现金提供银行余额、现金流量表、受限现金和月度烧钱速度
算力承诺 / 资本开支 / 债务不可获得算力扩大十倍可能带来重大义务提供 NVIDIA 和云厂商合同、采购承诺、租赁、债务及担保权益

公开来源对累计融资额的说法不一致,也没有披露这些经营分母;null 表示证据缺失,不代表数值为零。

[CI006, CI007, CI008, CI009, CI010, CI011, CI012, CI013, CI014, CI015, CI016]

4.4 财务结论

目前无法根据公开证据对 SSI 的收入质量、毛利路径和资本充足性评级。证据支持其商业化前、研究优先的定位,以及明显扩大的规划算力规模,但不支持收入、ARR、毛利率、烧钱速度、现金跑道或销售效率估计 [CI001, CI006, CI013, CI016]。NVIDIA 合作可能降低前沿算力获取风险,但投资金额和合同经济性未披露,无法判断它提供的是现金、容量、采购义务、战略依赖,还是几者组合 [CI007, CI008, CI012]。 核心尽调障碍不只是缺少传统 SaaS 指标,而是缺少一座连接已报道融资与研究产出的会计及合同桥梁。投资判断需要当前现金和月度烧钱速度、算力承诺和单位成本、债务或担保权益、员工成本、资金计划用途、产品里程碑,以及明确的商业化政策 [CI012, CI014, CI015]。在这些记录可获得前,财务逻辑只是押注技术成功的期权,而不是可预测的运营业务 [CI016]。

Chapter 05

05产品与技术

5.1 唯一规划产品,没有公开产品界面

SSI 表示只有一个目标和一个规划产品:安全的超级智能。这是研究目标,而不是今天可用的 SKU。截至 2026 年 9 月下旬,现有证据没有发现公开模型、API、聊天机器人、定价页面、模型卡或开发者文档。因此,没有可支持的买家、用户工作流、部署模式、服务水平记录或可量化客户收益可供评估。直线式结构可能减少中间商业化带来的干扰,但也消除了暴露可靠性、集成和支持问题的常规产品反馈闭环。因此,产品成熟度应归类为尚未发布的研究,而不是 alpha 软件或可部署平台。[CE001, CE002, CE011, CE013]

产品模块与公开成熟度矩阵
模块 / 产品线买方或用户公开状态 / 成熟度声称的差异化尽调缺口
安全超级智能研究计划未公开确认未发布的研究目标单一目标、直线式聚焦没有架构、基准测试、发布标准或可演示产品
公开模型或推理服务没有公开用户未发布没有中间产品周期没有权重、API、定价、文档或生产遥测
开发者集成界面没有公开开发者用户不可获得没有声称的集成优势没有 SDK、代码库、软件包、兼容性矩阵或支持流程
安全评估与治理层内部团队;未披露外部利益相关方方法未公开记录公司称安全与能力同步推进没有框架、阈值、审计证据、模型卡或事故协议

根据截至 2026-09-28 的指定证据整理的公开状态矩阵;null 或不可获得的界面属于证据缺口,不证明内部系统不存在。

[CE001, CE002, CE004, CE011, CE013]

5.2 架构未披露;算力是可观察层

公开证据支持的是基础设施轮廓,而不是模型架构。Google Cloud 于 2025 年为 SSI 研究提供 TPU,2026 年 7 月与 NVIDIA 的合作又增加了 Vera Rubin 平台访问权,并据公司称将算力扩大一个数量级。工作负载分配、集群规模、软件栈、训练方法、模型参数、数据策略和迁移计划均未披露。SSI 表示安全与能力同步推进,但这没有说明控制机制,也没有证明性能。关于类脑学习、样本效率、持续学习或内部价值函数的评论仍属推测;在没有一手技术披露前,不应将其描述为 SSI 已实施的技术。[CE005, CE006, CE007, CE008, CE009, CE014]

受证据约束的技术与运营架构
层级 / 组件公开支持的作用依赖项风险
产品目标开发一个安全的超级智能系统SSI 研究计划没有中间交付物或外部成熟度测试
模型与学习方法未公开披露内部研究架构评论具有推测性且无法复现
安全与能力设计公司称两者同步推进未披露的内部控制没有公开安全框架、阈值或外部审计
训练算力2025 年使用 Google TPU;2026 年宣布获得 NVIDIA Vera Rubin 访问权限Google Cloud 和 NVIDIA供应商集中度、迁移、容量、成本和排期未披露

本表将已披露基础设施与未知的内部架构分开;不会根据算力合作关系推断软件层、模型设计或认证。

[CE001, CE005, CE006, CE007, CE008, CE009, CE014]
FE001: 可公开观察的产品与技术栈

该技术栈将 SSI 陈述的目标与已披露的算力关系,同私有方法以及缺失的外部产品和保障界面区分开来。

计划中的产品
一个安全超级智能系统;没有公开可用的中间商业产品。
模型架构
未披露;参数、数据、训练、基准或可复现性细节均未公开。
安全与能力路径
SSI 表示安全与能力同步推进,但没有公布实施控制措施。
2025 年算力
Google Cloud TPU 容量支持研究工作负载。
2026 年扩容
已宣布获得 NVIDIA Vera Rubin 访问权限,并声称算力将实现数量级扩张。
外部界面
没有公开模型、API、开发者文档、论文、模型卡或安全框架。

仅为概念性证据地图;层级顺序不代表已披露的内部软件架构或实测技术表现。

[CE002, CE005, CE006, CE007, CE008, CE015]

5.3 部署、集成、可靠性与路线图

SSI 尚未公开部署或集成界面,因此无法评估正常运行时间、延迟、支持、安全运营、版本管理、兼容性和事故历史。一名投资人转述了 2026 年 8 月发布模型的预期,但 SSI 没有确认该日期;截至 9 月下旬,现有证据也没有发现发布。错过第三方传闻不是错过公司承诺,而是说明外部路线图尚未验证。NVIDIA 关系和 Sutskever 关于研究值得扩大规模的表述显示其愿意投入算力,但两者都不能证明基准质量、安全运行、生产准备度或发布日期。[CE010, CE012, CE013, CE015, CE016]

工作流、部署与用例证据
用户任务当前公开工作流SSI 解决方案可衡量收益局限
训练和评估内部研究系统封闭式研究工作流模型和训练流程未披露没有公开基准测试数据、方法、评估设计和可靠性结果均不可获得
扩大研究算力先有 Google TPU 关系,随后获得 NVIDIA Vera Rubin 访问权限由合作伙伴提供加速器容量公司报告称算力扩大一个数量级实际容量、利用率、成本、工作负载拆分和交付时间未披露
将模型集成到软件没有公开工作流没有模型、API、SDK 或文档无法衡量没有集成、安全、隐私、支持或服务水平证据
部署前审查安全性没有公开的保障工作流据称安全与能力同步推进无法独立衡量没有公开框架、阈值、审计、模型卡或发布门槛

算力扩大十倍的数字是公司 / 合作伙伴的说法,不是测得的吞吐量或模型性能结果;不可获得的字段需要私下技术尽调。

[CE006, CE007, CE008, CE011, CE013]

5.4 差异化是聚焦主张,尚不是可审计护城河

在公开记录中,SSI 的差异化是组织层面和愿景层面的:一家小而专注的实验室,打算在不维持中间商业产品的情况下同步发展安全与能力。技术差异化证据并未公开。现有来源没有发现论文、基准、模型卡、代码库、前沿安全框架、外部审计或记录在案的发布阈值。这种不透明可能保护研究,但阻止投资人和实践者比较安全控制、衡量进展或测试可复现性。因此核心尽调问题不是使命是否独特,而是私有证据能否在继续扩大规模前证明可重复的技术优势和可执行的安全控制。[CE003, CE004, CE006, CE011, CE015]

5.5 展示项

Chapter 06

06客户情况

6.1 商业界面与客户分层

SSI 尚未公开能让企业、开发者或研究人员成为可观察用户的产品界面,因此现有证据无法支持传统客户分层分析。独立评审未发现已发布模型、API、定价页面、开发者文档或基准,而 SSI 的直线式战略将安全超级智能置于中间产品之前。因此,下面的分层图是对潜在买方和用户群体的尽调分类,不是 SSI 当前服务这些群体的主张。评审来源没有找到付费账户、试点客户、生产部署或客户案例。缺失说明公开商业验证较弱,但不能证明不存在私下评估或未披露关系。[CU001, CU002, CU010, CU014]

客户细分表
细分买方 / 用户 / 付款方潜在用例有公开证据的规模收入 / 战略价值尽调缺口
企业买方买方、用户和付款方未披露未来在企业工作流中使用 SSI 模型审阅样本中没有具名账户或部署未知;没有定价或合同证据获取具名账户、部署状态、合同金额和结果
开发者和技术用户未披露开发者画像评估模型、API 或研究产物未找到已发布模型、API、SDK 或开发者文档未知;没有使用或变现界面获取访问日志、活跃用户分母和 API 条款
试点与设计合作伙伴经济买方和最终用户未披露安全测试、红队测试或受控评估审阅样本中没有具名试点或案例研究可能有验证价值,但目前没有证据获取试点协议、测试范围、结果和转化状态
基础设施交易对手NVIDIA 及据报道存在的 Google Cloud 关系算力供应和技术合作具名合作伙伴关系,不是客户部署具有战略运营价值;没有客户收入证据分开核查供应商、投资人、合作伙伴和客户的经济关系

这些细分是尽调分类,不是 SSI 报告的客户队列;null 或缺失规模表示指定的八个来源未提供该指标。

[CU001, CU002, CU005, CU007, CU014]

6.2 采用、留存与扩张仍无法衡量

公开采用证据在首个客户里程碑之前就中断了。Value Add VC 报道称截至 2026 年 9 月 25 日没有披露收入,多家来源也称预期的 8 月模型发布尚未出现;但已披露收入不等于 ARR,发布时点来自投资人而不是 SSI 的承诺。没有来源提供账户数、活跃用户、地点、利用率、积压订单、NRR、GRR、流失、续约、合同长度、满意度或队列数据。因此恰当结论不是留存很差,而是目前无法评估留存和扩张。在 SSI 提供带日期、定义分母的运营证据前,任何投资判断模型都应将这些字段保留为 null。[CU003, CU004, CU009, CU013]

客户增长 / 采用轨迹表
指标数值截至来源依据置信度含义缺失分母 / 限定
已披露收入报告为 $02026-09-25Value Add VC 和 AgentMarketCap中没有可用于判断客户采用的公开收入锚点第三方报道;不是 ARR,也不是公司经审计披露
已发布模型、API 或产品未找到2026-09-25Vorp Labs、MindStudio、Value Add VC 和 AgentMarketCap中没有可观察的用户引导界面公开渠道检查不排除私下内部产物
具名付费客户2026-10-06指定的八个来源中没有具名客户低无法公开验证商业牵引力实际客户群体和搜索完整性未知
账户、用户、地点或利用率2026-10-06指定来源池中没有定义分母的指标低无法计算采用轨迹需要有日期的队列、活动定义和全部潜在群体
NRR、GRR、流失、续约或合同期限2026-10-06指定来源池中没有留存指标低无法评估持久性和扩张需要队列周期、期初基数、排除项和合同条款
算力容量宣布扩大一个数量级2026-07-27NVIDIA-SSI 联合合作公告中提升研究能力,但不能证明采用情况没有客户、使用量或收入分母

各数值保留来源中的指标和日期;null 表示指定来源池中不可获得,报告的 $0 是已披露收入,不是 ARR 或现金。

[CU001, CU002, CU003, CU004, CU005, CU009, CU013]

6.3 已命名的交易对手不等于已命名客户

NVIDIA 是现有资料中最重要的已命名外部关系,但联合公告描述的是长期战略合作、投资、Vera Rubin 访问权和技术协作,并没有描述 NVIDIA 购买、试用或部署 SSI 产品。二手报道同样将 Google Cloud 描述为过去的算力提供商。把任一供应商或投资人当作客户证据,会夸大牵引力,也会掩盖 SSI 对资本和稀缺算力的依赖。因此,客户证明表明确记录关系类型,并为未披露的终端客户群保留开放行。这个区分很重要:即使客户集中度未知,供应商集中度也可能提高执行风险。[CU005, CU006, CU007, CU012]

具名客户证明表
具名方 / 群体关系部署 / 用例生产环境还是试点结果局限
NVIDIA投资人、算力供应商和技术合作伙伴Vera Rubin 访问权限及算力平台合作都不是;未描述 SSI 客户部署SSI 称算力将增加一个数量级联合公告未将 NVIDIA 认定为 SSI 产品客户
Google Cloud据报道曾是算力供应商用 TPU 开展 SSI 研究都不是;供应商关系据报道获得研究算力访问权限未报道 SSI 产品采购、试点或结果
终端客户未披露指定来源池中没有具名用例未知没有可独立验证的客户结果在此来源池中缺失不证明不存在私下关系

这份不完整的枚举将供应商和投资人与客户区分开;结果仅限于引用来源所述内容,没有留存证据。

[CU002, CU005, CU006, CU007, CU012]

6.4 判断客户质量前必须发生什么

下一项可信的采用信号不是又一条融资新闻,而是外部可检查的产品或研究产物,随后有明确的评估、试点和生产证据。Enterprise DNA 明确认为 SSI 的工作距离商业采用还有数年;IBM 一场讨论也提出了更广泛的问题:一个由资本支持、只有单一目标的研究组织,最终必须证明回报。旅程图将今天已观察到的内容与尚未解决的转化关卡分开。投资人在评估收入持久性或 land-and-expand 价值前,应要求客户身份、付款方和用户角色、部署状态、带日期的使用分母、结果、合同条款及续约证据。[CU008, CU011, CU013, CU014]

FU001: 客户证据旅程

SSI 已进入研究与算力扩展阶段,但每个客户验证和扩张阶段仍缺乏公开证据。

这是证据成熟度序列,并非预测,也不是声称 SSI 将完成每个阶段;尚未解决的阶段属于尽调要求。

[CU001, CU004, CU005, CU013, CU014]

6.5 展示项

Chapter 07

07风险

7.1 风险态势:证据不透明是核心暴露点

SSI 表示其唯一产品是安全的超级智能,安全应与能力同步推进,但这项使命声明不能证明任一结果已经实现。最强的公开风险信号是缺少分阶段技术证明:截至来源注明的 2026 年评审日期,现有来源没有 SSI 归属的论文、基准主张、公开模型、API 或前沿安全框架。一份技术综述还提醒,当前对齐方法只覆盖问题的一部分,不能构成端到端保证。因此结果是较高的剩余投资判断不确定性,而不是技术失败的证明。适当的缓释措施是将资本和算力获取与可独立复核的评估、治理阈值和回滚标准绑定;这些产物持续缺失就是投资逻辑破坏信号。[CR001, CR002, CR003, CR004, CR005]

FR001: 风险热力图

对最重大的公开证据风险进行序数综合,将未知状态与已观察事件分开。

序数标签综合指定来源与明确缺口;它们不是概率、事件数量或预测,也不证明未记录事件曾经或未曾发生。

[CR002, CR003, CR004, CR007, CR008, CR010, CR013, CR015]

7.2 治理、法律与监管态势

公开记录支持存在披露风险,但不能据此认定 SSI 违法或遭遇监管事件。Vorp Labs 报告称没有已发布的前沿安全框架;评审的公司分析也没有展示决策权、独立发布批准、事故报告、隐私控制、知识产权争议或特定司法辖区的部署义务。由于引用记录中的 SSI 尚未发布公开产品,未来产品界面和适用要求也无法可靠界定。这些都是重大尽调缺口。在投资或追加资本前,法律顾问应梳理美国和以色列实体,检查诉讼和监管往来,审阅 IP 转让,并测试独立机构能否暂停扩展或部署。[CR003, CR004, CR015, CR016]

监管 / 法律风险登记表
排名规则 / 许可 / 案件司法辖区公开状态可能性严重性缓解成熟度剩余暴露尽调路径
1前沿安全治理与部署控制美国 / 以色列;其他地区取决于部署经审阅来源未发现可比的公开 SSI 前沿安全框架高证据缺口严重没有公开证据高检查内部阈值、独立发布权限、评估结果、暂停标准和董事会会议记录。
2面向产品的 AI、隐私、安全和报告义务在产品、用户、数据流和部署市场确定前无法判断所引用的 2026 年 8 月至 9 月记录中没有公开模型或 API适用性存在中等不确定性高无法公开评估高在部署前获取架构、数据流、预期用途、模型访问、事故响应和市场进入计划。
3诉讼、执法、知识产权所有权及监管事故处置美国 / 以色列指定证据未能确定;这不等于不存在未知若出现不利情况则为高未知未知要求提供诉讼和监管机构证明、知识产权转让、前雇主豁免、保险清单和事故日志。

严重性和可能性是锚定所引公开披露缺口的定性分析判断,不是事故频率或法律结论;“未知”表示指定公开证据无法确立该事项。

[CR003, CR004, CR006, CR015, CR016]

7.3 运营、技术与领导层执行

SSI 的直线式路线限制了外部验证。Vorp Labs 截至 2026 年 8 月 15 日没有找到公开模型或 API;Value Add VC 报道称,投资人曾提出 8 月发布窗口,但截至 9 月 25 日仍没有模型。该传闻不是 SSI 的承诺,因此证据支持的是里程碑不确定性,而不是公司承诺被打破。第三方分析还认为,避免迭代部署可能限制真实世界压力测试。领导层集中增加了执行敏感性:现有来源报告 Daniel Gross 于 2025 年离任,之后 Ilya Sutskever 和 Daniel Levy 分别担任 CEO 和总裁。缓释措施需要董事会批准的继任安排、记录在案的里程碑评审、独立红队证据,以及在扩展速度超过保证能力时预先约定停止。[CR006, CR007, CR010, CR014]

运营 / 质量 / 安全风险登记表
排名失效模式证据可能性严重性缓解成熟度剩余暴露终止标准 / 未解决缺口
1在进一步扩大规模前无法独立测试安全保障所引用的 2026 年审阅中没有公开 SSI 论文、基准测试声明或前沿安全框架高证据缺口严重没有公开证据高如果独立评估者无法检查方法、失效类别、阈值和回滚权限,就停止扩大规模。
2技术里程碑延期在外部仍无法衡量9 月 25 日仍没有公开模型,此前投资人称 8 月可能发布,但该时间窗未经确认高可观测性风险高没有公开里程碑节奏高要求提供有日期的内部里程碑和证据包;单独错过传闻时间不构成公司违约。
3关键人物与领导层连续性Gross 于 2025 年离职;Sutskever 出任 CEO,Levy 出任总裁中高没有公开的继任梯队证据高要求提供继任安排、留任方案、关键人物保险、董事会监督和分布式技术签字机制。
4现实验证和反馈有限没有公开模型或 API;第三方分析指出,由于未部署,压力测试有所减少中高内部控制未披露高在任何不可逆或高自主性使用前,要求独立对抗性评估和沙盒证据。

行按剩余严重性排序;可能性标签描述的是证据与可观测性风险,而非实测失败概率;缓释措施的成熟度仅限于指定来源允许外部审阅的范围。

[CR002, CR003, CR004, CR006, CR007, CR010, CR014]

7.4 伙伴、财务与模型风险

算力获取和融资既是优势,也是集中风险。来源描述 SSI 先使用 Google Cloud TPU,随后于 2026 年与 Nvidia 合作,预计将算力扩大一个数量级;但在所评审记录中,两家公司没有披露投资金额。因此,融资总额不能互换:GetLatka 在一页最后更新于 2025 年的资料上记录一轮 $2 billion;Vorp Labs 将约 $3 billion 的已确认及已报道股权与未披露的 Nvidia 投资分开;Value Add VC 则将报道中的 Nvidia $5 billion 金额计入后估计约 $8 billion。没有任何来源证明可用现金、烧钱速度、现金跑道或融资条款。缓释措施应是核实股权结构表、现金、合同和供应商退出安排,而不是从估值或融资新闻外推。[CR008, CR009, CR011, CR012, CR013, CR016]

合作伙伴 / 依赖风险登记表
排名依赖项对手方角色集中度失败情景严重性缓释措施剩余暴露
1前沿算力与战略资本NvidiaVera Rubin 访问权限、合作关系与未披露投资下一阶段的重大依赖交付、分配、商业条款或战略优先级变化限制计划中的扩展高审计合同、核验容量里程碑、托管过渡方案,并与替代供应商做基准比较在条款与可迁移性得到核实前为高
2训练平台迁移Google Cloud 与 Nvidia据报道先以 TPU 为研究基础,随后扩展至 Nvidia两个战略基础设施生态移植或编排复杂度拖慢实验,或削弱可复现性高双平台可复现性测试、数据可移植性,以及明确的迁移验收标准中高
3融资条款与资本充足性私人投资者联合体与 Nvidia股权与战略融资条款、现金余额与投资者权利未披露表面融资额高估可用现金,或造成使命、稀释或流动性约束高核实现金、股权结构表、优先权、契约、董事会权利、承诺算力、烧钱速度与现金跑道高
4外部验证与市场反馈独立评估者和未来用户安全挑战、部署反馈与需求证据引用的评审中没有公开产品界面研究在缺乏外部可观察证据证明安全与能力里程碑一致的情况下推进高在扩大授权前,分阶段开展独立评估,并进行规模有限且可逆的试点高

融资额不等同于可用现金;Nvidia 的金额仍是来源报道,而非公司披露;依赖严重性是分析师综合判断,并非量化的宕机或合同违约概率。

[CR006, CR008, CR009, CR011, CR012, CR013, CR014, CR016]

7.5 展示项

Chapter 08

08估值

8.1 估值背景与资本历史

Safe Superintelligence Inc. 在前沿人工智能研究中处于前所未有的估值位置。公司在 Greenoaks Capital 领投的 2025 年 4 月融资中估值 $32 billion,并将这一基准维持到 2026 年末。据报 2026 年 7 月宣布与 Nvidia 的无定价战略合作后,累计融资约达 $8 billion,Bloomberg 报道该合作金额为 $5 billion。尽管资本积累巨大,SSI 仍没有商业收入、企业 API,且尚未见公开模型部署;按估计的 50 人团队计算,隐含纸面估值约为每名员工 $640 million。Hiive 等平台的二级市场报价曾显示股价约 $93.64,但实际交易流动性仍严重受限。在没有商业现金流或经审计财务报表的情况下,$32 billion 估值体现的是对创始人技术能力的纯看涨期权,而不是传统运营倍数。[CV001, CV002, CV003, CV004, CV005, CV010, CV013]

建议摘要表
维度评级 / 数值驱动因素 / 证据基础决策含义
建议Research-More尚无收入的研究实验室,收入为 $0,且没有公开产品在技术里程碑得到核实前,暂不配置资本
信心Medium已记录的 $8B 融资与 Nvidia 合作,对上完全的研究保密跟踪二级市场与算力部署,不承担一级市场风险
风险评级High资本密集度高、依赖算力,且缺乏外部治理框架入场前要求高级别清算优先权或技术尽调权限
估值立场Expensive$32B 估值标记未变,在没有商业收入的情况下意味着人均 $640M等待估值倍数压缩或有定价的里程碑轮次后再参与

摘要综合了融资历史报道、公开算力合作关系,以及截至 2026 年末仍无商业收入这一事实。

[CV001, CV004, CV005, CV013]

8.2 投资逻辑与反向逻辑

SSI 的核心投资逻辑建立在创始人稀缺性和研究纯粹性上。实验室坚持只专注于安全超级智能的直线式使命,消除了 OpenAI 和 Anthropic 等竞争者常见的组织管理成本、产品管理摩擦和商业妥协。这一聚焦使命吸引顶尖研究人员来到 Palo Alto 和 Tel Aviv,也获得 Sequoia Capital、Andreessen Horowitz 和 Greenoaks Capital 的支持。此外,2026 年 7 月的 Nvidia 合作让 SSI 能专门获取下一代 Vera Rubin 基础设施,预计算力扩大十倍。反向逻辑则强调严重的运营脆弱性:SSI 没有发布技术论文或模型卡,也错过了市场投资人所引用的 2026 年 8 月等传闻外部交付窗口。此外,联合创始人 Daniel Gross 于 2025 年中离开并加入 Meta,此前收购要约遭拒,凸显高管更替风险,也使技术执行高度集中于 Ilya Sutskever 和 Daniel Levy。[CV006, CV007, CV008, CV009]

正向逻辑 / 反向逻辑表
正向逻辑角度核心论点证据与反证哪些变化会改变判断
创始人与人才护城河Ilya Sutskever 的履历无需营销投入,就能吸引顶尖研究人才与资本。Sequoia、a16z、Greenoaks 支持了 $3B 股权融资;已组建约 50 名顶尖研究人员的团队。除 Daniel Gross 外,出现实质性的研究人员离职或联合创始人流失。
直线式对齐只专注于安全超级智能,避开困扰 OpenAI 和 Anthropic 的商业干扰。SSI 不发布消费应用、API 或模型卡,从而保持专注。竞争实验室通过商业化迭代部署,率先解决可扩展对齐问题。
算力规模优势Nvidia 合作借助下一代 Vera Rubin 基础设施带来 10 倍算力扩张。Nvidia 在长期战略芯片供应协议之外,还取得了股权。没有收入无法消化算力成本,或芯片交付延迟。
估值脱节$32B 估值对应 $0 收入,带来巨大的下轮融资与清算优先权悬置。对一个没有公开概念验证(PoC)的实验室而言,人均 $640M 前所未有。公开发布具有变革性且安全对齐的推理模型。

该分析将创始人主导的技术选择权,与极端的商业不透明度和资本密集度进行对照。

[CV001, CV004, CV006, CV008, CV009]

8.3 情景建模与下行保护

评估潜在投资回报,需要对 2027 年和 2028 年之前分化的开发路径进行明确的概率加权。乐观情景是:SSI 在样本效率或可扩展对齐上实现根本性算法突破,累计资本需求接近 $22 billion,估值跃升至 $60 billion–$200 billion。基准情景是:SSI 继续进行深度隐身研究,吸收 Nvidia 的算力承诺,并通过无定价的战略企业扩展维持 $32 billion 的估值底线。然而在悲观和困境情景下,技术里程碑停滞,大量算力消耗耗尽流动性。由于数十亿美元融资批次经常附带高级清算优先权和棘轮条款,一旦技术停滞迫使公司进行下轮融资或困境资产整合,普通股股东将面临严重稀释和减记风险。[CV002, CV003, CV011, CV012]

乐观 / 基准 / 悲观情景表
情景假设估值 / 回报逻辑主要下行风险概率信号
乐观情景(超级智能突破)SSI 在样本效率与对齐上取得突破;以 $80B-$200B 估值完成 $10B+ 超大轮融资。按当前估值标记回报 2.5x - 6.0x;主权资本与超大规模云厂商资本入场。全球监管介入,或出于国家安全而将算力国有化。15% - 低
基准情景(隐身研发延续)SSI 维持 $32B 估值底线,消耗 Nvidia 算力,并在 2027 年前私下扩展。回报 1.0x;在未定价的战略追加融资中,估值标记维持 $32B 不变。追加数十亿美元算力融资造成过度稀释。55% - 高
悲观情景(执行失败 / 下轮融资)传闻中的模型未展现能力跃升;核心研究人员离职;算力成本耗尽可用现金。估值减记 50% - 80%;高级清算优先权抹去普通股权益。下轮融资重组,或被战略合作伙伴强制收购。25% - 中
极端悲观情景(过时 / 解散)前沿竞争者解决推理对齐问题;SSI 的架构假设被证明不可行。资本全部损失,或折价出售算力配额等资产。完全清算,剩余资本仅返还给高级优先股持有人。5% - 低

概率分布与估值倍数反映二元化的研究结果,以及中途缺乏现金流。

[CV001, CV002, CV003, CV011, CV012]

8.4 建议逻辑与尽调触发点

综合非凡创始人才华与严重估值不透明,结论是明确的继续研究建议,并配以昂贵的估值立场。投资人看不清清算优先权级别和股权结构表的治理安排前,应避免追逐接近每股 $93.64 的二级市场报价。持续监测必须聚焦明确的投资逻辑破坏触发点,包括算力交付长期延迟、Palo Alto 或 Tel Aviv 研究群体的关键人员离职,或竞争对手在可证明安全架构上取得突破、使 SSI 的专有直线式方法变得多余。新增本金投入应继续以经审计的里程碑验证、外部安全治理披露,或反映更广泛市场发现的定价轮为前提。[CV004, CV006, CV010, CV013, CV014]

FV001: 建议逻辑

将 SSI 的规模、证明点、风险与估值串联到继续研究建议的决策流程。

节点 / 连接节点 / 起点终点说明
节点 1 资本规模 [N1]$8B 累计融资,以及与 Nvidia 达成的 10x 算力交易
节点 2 商业证明 [N2]商业收入为零、公开模型为零、已发表论文为零
节点 3 估值基准 [N3]$32B 静态估值在尚无收入的情况下意味着人均 $640M
节点 4 执行风险 [N4]未定价的融资保密、缺乏安全框架,以及关键创始人离任
节点 5 尽调触发条件 [N5]接近 $93.64/股的二级市场指示价,以及即将到来的算力交付里程碑
节点 6 投资结论 [N6]继续研究建议:跟踪二级市场估值标记,等待经过核实的技术里程碑

将资本密集度、缺乏产品证明与估值悬置综合为尽调门槛的顺序逻辑结构。

[CV004, CV006, CV010, CV013, CV014]

8.5 展示项

免责声明

本报告是基于公开证据的尽调快照,不构成投资建议。重要的财务、法律、技术和合同事实仍未公开,任何投资决策前都应直接向管理层及一手文件核实。

证据索引

结论
编号陈述可信度来源
CO001 Safe Superintelligence Inc. was founded in June 2024 by Ilya Sutskever, Daniel Gross, and Daniel Levy. 中 SO003, SO004, SO006
CO002 SSI describes itself as an American company with offices in Palo Alto and Tel Aviv, while third-party profiles identify Palo Alto as its headquarters. 中 SO001, SO004, SO008
CO003 SSI says its sole goal and intended product is a safe superintelligence, with safety and capabilities pursued together. 中 SO001, SO006
CO004 Independent source reviews through September 2026 found no public SSI model, API, product surface, research paper, or technical blog post. 中 SO003, SO004, SO007
CO005 SSI confirmed a $1 billion raise on September 4, 2024, from NFDG, Andreessen Horowitz, Sequoia, DST Global, and SV Angel. 中 SO002, SO003
CO006 Independent sources report that SSI raised approximately $2 billion in 2025 at a $32 billion valuation, led by Greenoaks, without confirmation from SSI in the assigned official evidence. 中 SO003, SO004, SO006, SO008
CO007 SSI confirmed a July 2026 strategic partnership and investment from NVIDIA that it says will increase its compute by tenfold, but the companies did not disclose the investment amount in assigned evidence. 中 SO002, SO003, SO006
CO008 Secondary sources attribute roughly $5 billion to NVIDIA's 2026 investment, but the amount remains unconfirmed by the companies. 中 SO003, SO004, SO007
CO009 SSI's July 2025 update names Ilya Sutskever as CEO and Daniel Levy as President, with the technical team reporting to Sutskever. 中 SO002, SO003, SO004
CO010 SSI states that co-founder Daniel Gross was officially no longer part of the company after June 29, 2025. 中 SO002, SO003, SO006
CO011 Assigned independent sources provide conflicting current headcount estimates of 13, roughly 20-30, and roughly 50, while SSI provides no official count. 低 SO003, SO004, SO006, SO007, SO008
CO012 The assigned public evidence contains no company-issued revenue figure or customer count for SSI. 中 SO001, SO002, SO003, SO004
CO013 SSI's board composition, investor control rights, mission-lock provisions, and independent safety-oversight structure are not disclosed in the assigned evidence. 低
CO014 Independent reviews report that SSI has not published a frontier-safety framework, preventing external comparison of its thresholds and oversight processes. 中 SO003, SO004, SO006
CO015 Independent sources report that SSI used Google Cloud TPUs in 2025 before announcing expanded access to NVIDIA systems in 2026. 中 SO003, SO004, SO006
CO016 An investor-attributed report pointed to an August 2026 SSI model release, but SSI did not confirm the date and no public release was found by September 25. 中 SO003, SO005, SO006, SO007
CM001 Safe Superintelligence Inc. is structured with a singular focus on developing safe superintelligence, deliberately avoiding intermediate commercial products, software tooling, or enterprise APIs. 中 SM001, SM005, SM010
CM002 Safe Superintelligence Inc. operates with zero disclosed commercial revenue and zero launched products while carrying a $32 billion private market valuation. 中 SM004, SM005, SM006
CM003 The company's $32 billion valuation benchmark was established in an April 2025 financing round led by Greenoaks Capital and has remained flat through late 2026. 中 SM004, SM005, SM009
CM004 Total capital raised by Safe Superintelligence Inc. reached approximately $8 billion following an unconfirmed $5 billion strategic investment and Vera Rubin compute partnership from Nvidia in July 2026. 中 SM004, SM005, SM009
CM005 Peer research lab Thinking Machines Lab was valued at approximately $13.5 billion on secondary indications after raising $2 billion, while World Labs raised $1.23 billion at roughly a $5 billion valuation. 中 SM004
CM006 Frontier foundation model operators Anthropic and OpenAI generated annualized run-rate revenues of $65 billion and $40 billion respectively by late 2026, serving enterprise API and consumer subscription markets. 中 SM004, SM007
CM007 Because Safe Superintelligence Inc. offers no commercial product or API, its immediate addressable enterprise software market spend is currently zero dollars. 中 SM001, SM004, SM010
CM008 Potential future buyers of safe superintelligence systems comprise hyperscale infrastructure operators, sovereign entities, and frontier research consortia requiring autonomous cognitive automation. 中 SM001, SM008, SM010
CM009 Safe Superintelligence Inc. previously utilized Google Cloud TPUs for research workloads starting in April 2025 before expanding its compute an order of magnitude via Nvidia's Vera Rubin systems in July 2026. 中 SM004, SM005, SM009
CM010 An investor report suggesting Safe Superintelligence Inc. would release its first model in August 2026 did not materialize, leaving the company with zero public model weights or released demos. 中 SM004, SM005, SM008
CM011 Safe Superintelligence Inc. maintains no public research papers on arXiv, no public model cards, and no published frontier safety framework comparable to frameworks published by Anthropic or OpenAI. 中 SM005, SM010
CM012 Former chief executive Daniel Gross co-founded Safe Superintelligence Inc. in 2024 with Ilya Sutskever and Daniel Levy before departing in mid-2025 to join Meta's superintelligence unit. 高 SM003, SM005, SM009
CM013 Meta Platforms unsuccessfully attempted to acquire Safe Superintelligence Inc. in mid-2025 before hiring co-founder Daniel Gross and offering to purchase up to 49 percent of venture fund NFDG. 高 SM003, SM009
CM014 Safe Superintelligence Inc. raised $1 billion in its initial September 2024 financing round from venture backers including Andreessen Horowitz, Sequoia Capital, DST Global, and SV Angel. 高 SM002, SM003, SM005
CM015 Safe Superintelligence Inc. maintains an exceptionally lean headcount estimated at between 13 and 50 employees across Palo Alto and Tel Aviv, implying over $600 million in valuation per employee. 中 SM004, SM006, SM007
CP001 AI Landscape and Distill Intelligence place OpenAI, Anthropic, Google DeepMind, xAI, and Mistral AI within SSI's direct frontier-lab comparison set. 中 SP001, SP007
CP002 Value Add VC reports that Thinking Machines Lab shipped its Tinker fine-tuning API and World Labs shipped its Marble spatial product, while SSI had no shipped product. 中 SP004
CP003 AI Security and Safety describes OpenAI's public focus as spanning alignment, red teaming, responsible AI, evaluation, and content safety, while describing SSI's focus as alignment and responsible AI. 中 SP008
CP004 Vorp Labs found no released SSI model, API, product, developer documentation, or pricing page on official channels or the checked public model catalog as of August 15, 2026. 中 SP005
CP005 Vorp Labs found no published SSI frontier safety framework while identifying published frameworks for Anthropic, OpenAI, Google DeepMind, Meta, and xAI. 中 SP005
CP006 MindStudio states that most specific claims about SSI's architecture and methods are speculation based on limited public statements. 中 SP003
CP007 Public headcount estimates are not comparable: GetLatka reports approximately 13 employees, while Global VC Megadeal Briefing estimates roughly 50 without a common measurement date or method. 低 SP002, SP006
CP008 Vorp Labs reports about $3 billion of confirmed-and-reported SSI equity before the 2026 NVIDIA investment, whose amount both companies left undisclosed. 中 SP005
CP009 Value Add VC claims SSI had raised roughly $8 billion after a reported $5 billion NVIDIA investment, conflicting with the more conservative confirmed-and-reported funding basis. 低 SP004
CP010 Sources describe SSI using Google Cloud TPUs and later securing NVIDIA Vera Rubin access, but the assigned evidence does not establish exclusive supply or proprietary distribution rights. 中 SP002, SP005
CP011 No public SSI pricing page, contract package, or realized price was identified by the dated source review. 中 SP005
CP012 The assigned evidence provides named shipped products for Thinking Machines Lab and World Labs but no equivalent public artifact for SSI. 中 SP004, SP005
CP013 Value Add VC reported that a rumored August 2026 SSI model release had not appeared by September 25, while noting that SSI never confirmed the timeline. 中 SP004
CP014 Without a public model, benchmark, method disclosure, or safety framework, SSI's claimed research advantage cannot yet be independently compared with frontier peers. 中 SP003, SP005
CP015 The assigned pool does not establish buyer switching costs, multi-homing behavior, internal-build economics, or the cost of deferring adoption. 低
CP016 The assigned evidence is asymmetric: it documents SSI's missing public artifacts and peer safety-framework visibility more strongly than it documents comparative technical capability. 中 SP004, SP005, SP008
CI001 Multiple independent sources published in 2026 describe SSI as having no released commercial product and no disclosed revenue. 中 SI001, SI003, SI004, SI005, SI007, SI008
CI002 The retained source set identifies no paying customer, customer contract, pricing page, or active commercial channel for SSI. 中 SI001, SI003, SI005, SI007, SI008
CI003 Multiples classifies SSI's revenue model as subscription and its customer focus as B2B. 低 SI008
CI004 Multiples also describes SSI as pre-commercial and says it has disclosed no revenue or financial metrics. 中 SI008
CI005 No retained evidence discloses list pricing, realized pricing, discounts, contract units, or a revenue-recognition method for SSI. 中 SI001, SI002, SI003, SI005, SI007, SI008
CI006 No retained evidence discloses SSI's customer acquisition cost, sales cycle, payback, channel economics, retention, utilization, or gross margin. 中 SI001, SI002, SI003, SI004, SI005, SI008
CI007 Independent coverage reports that SSI's NVIDIA partnership provides Vera Rubin access and is intended to expand compute by an order of magnitude. 中 SI003, SI005
CI008 Neither SSI nor NVIDIA disclosed the investment amount, while secondary reports described approximately $5 billion or multiple billions without company confirmation. 中 SI003, SI005
CI009 Vorp Labs distinguishes SSI's officially confirmed $1 billion 2024 raise from a reported approximately $2 billion 2025 round and an undisclosed 2026 NVIDIA investment. 中 SI005
CI010 Silicon Valley Investclub, Value Add VC, and Multiples report approximately $8 billion of cumulative financing across three rounds. 中 SI002, SI004, SI008
CI011 GetLatka reports $2 billion raised across one round, materially differing from sources that report three rounds and approximately $8 billion. 低 SI006
CI012 The retained evidence cannot translate funding raised into cash available because it does not disclose current cash, burn, restrictions, or financing obligations. 中 SI002, SI003, SI004, SI005, SI008
CI013 The reported order-of-magnitude compute expansion is the clearest observable cost and capital-intensity driver, but no associated cost is disclosed. 中 SI003, SI005
CI014 No retained source provides SSI's working capital, capex classification, service-delivery costs, debt, project-finance obligations, or compute purchase commitments. 中 SI002, SI003, SI004, SI005, SI008
CI015 Public evidence provides no supportable basis to calculate SSI's runway or next-round trigger. 中 SI002, SI003, SI004, SI005, SI008
CI016 Public evidence supports a pre-commercial, compute-intensive research posture, but it does not support underwriting revenue quality, margin path, or capital adequacy. 中 SI001, SI002, SI003, SI004, SI005, SI008
CE001 SSI publicly defines safe superintelligence as its sole planned product and research objective. 中 SE001, SE003, SE005, SE008
CE002 The assigned evidence found no publicly released SSI model, API, chatbot, pricing page, or other usable product through late September 2026. 中 SE004, SE005, SE008
CE003 The assigned evidence found no SSI research paper or technical blog post describing its model methods. 中 SE003, SE004, SE005, SE008
CE004 SSI had not published a frontier safety framework in the assigned evidence through September 2026. 中 SE005, SE008
CE005 SSI has not publicly disclosed its model architecture, training method, model scale, data strategy, or benchmark results. 中 SE001, SE003, SE005, SE007, SE008
CE006 SSI says it advances safety and capability together, but the public statement does not disclose an implementation or control framework. 中 SE001, SE003
CE007 Google Cloud supplied TPU capacity for SSI research in 2025. 中 SE002, SE003, SE005, SE008
CE008 SSI and NVIDIA announced access to the Vera Rubin platform in July 2026 with a claimed order-of-magnitude expansion in SSI compute. 中 SE002, SE003, SE005, SE008
CE009 The public record shows successive Google TPU and NVIDIA accelerator relationships, but it does not disclose workload allocation, migration status, capacity, utilization, or cost. 中 SE003, SE004, SE005, SE008
CE010 An investor relayed an August 2026 SSI model-release expectation that SSI did not publicly confirm and that had not materialized by late September 2026. 中 SE003, SE004, SE005, SE007
CE011 No public SSI developer documentation, model catalog entry, or integration surface was identified in the assigned evidence. 中 SE005, SE008
CE012 SSI operates its research program across Palo Alto and Tel Aviv. 中 SE002, SE003, SE005, SE008
CE013 SSI has not published deployment, integration, support, reliability, security, privacy, or service-level evidence for an external product. 中 SE001, SE005, SE008
CE014 Public discussion of sample efficiency, continual learning, and internal value functions is third-party speculation rather than verified SSI architecture. 低 SE007
CE015 The NVIDIA partnership and SSI's statement that its research was worthy of scaling are commitment signals, not independent proof of capability, safety, or production readiness. 中 SE004, SE005, SE007, SE008
CE016 SSI has not published an externally verifiable release timetable or intermediate technical milestones. 中 SE003, SE004, SE005, SE008
CU001 Independent public-record reviews found no released SSI model, API, product, pricing page, developer documentation, or benchmark by September 25, 2026. 中 SU003, SU004, SU005, SU006
CU002 None of the eight sources allocated to this chapter identifies a named paying SSI customer, customer pilot, production deployment, or customer outcome. 中 SU003, SU004, SU005, SU006
CU003 Value Add VC reported zero disclosed SSI revenue as of September 25, 2026, while AgentMarketCap separately described SSI as having zero revenue in April 2026. 中 SU004, SU005
CU004 The rumored August 2026 model release came from investor commentary rather than an SSI announcement and had not appeared publicly by September 25, 2026. 中 SU003, SU005, SU006
CU005 The July 2026 NVIDIA-SSI announcement describes an investment, Vera Rubin access, and technical collaboration expected to increase SSI's compute by an order of magnitude. 中 SU001, SU002, SU006
CU006 The joint NVIDIA-SSI release does not characterize NVIDIA as a buyer, pilot user, or production customer of an SSI product. 中 SU002
CU007 Secondary sources describe Google Cloud as an SSI compute provider, not as a customer deploying SSI technology. 中 SU005, SU006
CU008 Enterprise DNA assessed that SSI's work would not reach business users for years, highlighting the gap between frontier research and enterprise adoption. 中 SU001
CU009 The assigned evidence provides no NRR, GRR, churn, renewal, contract-length, satisfaction, or cohort metric with which to assess SSI customer durability. 中 SU004, SU005, SU006
CU010 Latka reports company funding and headcount metadata but provides no customer count, usage, deployment, or retention evidence for SSI. 低 SU007
CU011 An IBM expert panel argued that a capital-backed, single-goal AI research organization would eventually face pressure to demonstrate returns or commercial relevance. 中 SU008
CU012 With customers undisclosed, the evidenced concentration is in capital and compute counterparties rather than in customer revenue. 中 SU001, SU002, SU005, SU006
CU013 Customer retention and expansion cannot be evaluated until SSI discloses a customer population, dated usage cohorts, contract terms, and renewal outcomes. 中 SU001, SU005, SU006
CU014 SSI's public customer-evidence journey currently reaches research and compute scaling but not an inspectable release, named pilot, production deployment, or renewal. 中 SU001, SU002, SU003, SU005, SU006
CR001 SSI states that its sole product is safe superintelligence and that safety and capability should advance together. 中 SR002, SR003, SR007
CR002 A technical survey concludes that current alignment techniques address parts of the safety problem but do not compose into an end-to-end proof. 中 SR001
CR003 Assigned 2026 reviews report that SSI had published no papers, benchmark claims, public evaluations, or detailed technical milestones. 中 SR003, SR005, SR009
CR004 Vorp Labs reports that SSI had not published a frontier safety framework comparable on thresholds, reviews, or update cadence with those of named frontier peers. 中 SR005, SR009
CR005 A Constellation Research analyst said SSI's direction and milestones were unclear at launch. 中 SR002
CR006 Vorp Labs found no released SSI model, API, or product as of its August 15, 2026 review. 中 SR005
CR007 Value Add VC reported no public model by September 25, 2026 after an outside investor described an August window, while emphasizing that SSI itself had not promised that date. 中 SR004
CR008 The 2026 Nvidia partnership was reported to include an investment and compute access expected to increase SSI's compute by an order of magnitude, but the companies did not disclose the investment amount. 中 SR003, SR005, SR009
CR009 Assigned reviews describe SSI using Google Cloud TPUs before its next scaling phase with Nvidia hardware. 中 SR003, SR005
CR010 Assigned sources report that Daniel Gross left SSI in 2025, after which Ilya Sutskever became CEO and Daniel Levy became president. 中 SR003, SR005, SR009
CR011 GetLatka's page, last updated September 30, 2025, records $2 billion raised across one round and states approximately 13 employees as of 2026. 低 SR006
CR012 Vorp Labs separates a roughly $3 billion confirmed-and-reported equity base from an undisclosed 2026 Nvidia investment amount. 中 SR005
CR013 Value Add VC estimates roughly $8 billion of lifetime funding by September 25, 2026 by including a reported $5 billion Nvidia amount, which differs in period and inclusion basis from the lower funding records. 低 SR004
CR014 The absence of iterative public deployment can reduce external real-world stress testing and market-feedback evidence, although it does not establish that SSI lacks internal testing. 中 SR005, SR008, SR009
CR015 The assigned evidence does not establish SSI's litigation, enforcement, licensing, IP-ownership, privacy, regulatory-incident, or deployment-approval posture. 低
CR016 SSI's documented compute and financing relationships create material counterparty dependencies but do not independently validate the safety of its research outputs. 中 SR001, SR005, SR009
CV001 Safe Superintelligence Inc. was valued at $32 billion in an April 2025 financing round led by Greenoaks Capital and maintained that reference valuation into late 2026. 中 SV001, SV002, SV003
CV002 Nvidia completed a strategic investment in SSI announced in July 2026, reported by Bloomberg at $5 billion, without publicly disclosing official transaction terms. 中 SV001, SV002, SV004, SV007
CV003 SSI has raised approximately $8 billion in total capital across its $1 billion seed, $2 billion April 2025 round, and the July 2026 Nvidia financing. 中 SV001, SV003
CV004 SSI operates as a pre-revenue research laboratory with zero reported commercial revenue, zero commercial APIs, and no publicly shipped products as of late 2026. 中 SV001, SV004
CV005 SSI employs approximately 50 personnel, implying an exceptional paper valuation of roughly $640 million per employee at the $32 billion mark. 中 SV001
CV006 SSI's strategic partnership with Nvidia supplies next-generation Vera Rubin compute infrastructure designed to expand the lab's computational capacity by an estimated 10 times. 中 SV004, SV007
CV007 A single-source rumor of an August 2026 model release did not materialize on official channels, leaving SSI without public models or technical papers. 中 SV001, SV004
CV008 Co-founder Daniel Gross departed SSI in mid-2025 following a rejected acquisition bid by Meta, leaving Ilya Sutskever as Chief Executive Officer and Daniel Levy as President. 中 SV001, SV008
CV009 SSI's founding philosophy commits to a straight-shot development path that rejects intermediate product monetization or commercial distraction prior to achieving safe superintelligence. 中 SV004, SV008
CV010 Secondary market indications for SSI equity have traded through broker estimates near $93.64 per share on platforms like Hiive, though transacted liquidity remains limited. 中 SV001
CV011 A base valuation scenario for SSI anchors at $32 billion, whereas downside bear cases model significant liquidation preference overhang and down-round compression if technical milestones stall. 中 SV001, SV002
CV012 Bull case market projections model cumulative capital requirements reaching $22 billion by 2028 and prospective step-up valuations ranging between $60 billion and $200 billion upon proof of scalable alignment. 中 SV002
CV013 The diligence stance recommends a research-more rating with an expensive valuation assessment, reflecting binary founder optionality and extreme capital intensity rather than operating fundamentals. 中 SV001, SV002
CV014 Key thesis-break triggers include sustained compute unavailability, core researcher departures, or external frontier labs achieving verifiable safe reasoning breakthroughs ahead of SSI. 中 SV001, SV004, SV008
来源
编号出版方标题引文
SO001 Safe Superintelligence Inc. Safe Superintelligence Inc. We have started the world’s first straight-shot SSI lab, with one goal and one product: a safe superintelligence.
SO002 Safe Superintelligence Inc. Updates - Safe Superintelligence Inc. I am now formally CEO of SSI, and Daniel Levy is President. The technical team continues to report to me.
SO003 Vorp Labs Research Safe Superintelligence (SSI): Funding, NVIDIA Partnership, Verifiable Record Confirmed and reported equity totals about 3 billion dollars; larger circulating figures are unverified
SO004 HokAI Safe Superintelligence: Funding, Founders & Products (2026) SSI is a private research lab led by CEO Ilya Sutskever and President Daniel Levy, working from Palo Alto and Tel Aviv. It sells no chatbot, API or model and publishes no research.
SO005 MindStudio Safe Superintelligence Explained: Ilya Sutskever's SSI and Its 2026 Plans Most specifics about SSI’s architecture and methods are speculation based on limited public statements.
SO006 AI Wiki Safe Superintelligence Inc **Safe Superintelligence Inc.** (abbreviated **SSI**) is an [AI safety](https://aiwiki.ai/wiki/ai_safety) research company that is building a single product, a safe [superintelligence](https://aiwiki.ai/wiki/superintelligence), and nothing else.
SO007 Value Add VC $32B SSI Valuation Implies $640M Per Employee, $0 Revenue A missed rumor is not the same as a broken promise — SSI never confirmed Baker's timeline in the first place, so there is no company statement to hold it to.
SO008 GetLatka Safe Superintelligence Inc. Funding 2026: $2B Raised Safe Superintelligence Inc. employs approximately 13 people as of 2026.
SM001 AI Market Watch Safe Superintelligence Inc. - AI Startup Profile | AI Market Watch Safe Superintelligence Inc. is a dedicated AI research lab with a singular focus on building safe superintelligence, insulated from short-term commercial pressures.
SM002 Exa Websets Safe Superintelligence Inc. Funding & Key Investors As of March 2025, Safe Superintelligence Inc. employs around 20 people, focusing its research and development on advancing superintelligent AI in a safe and controlled manner.
SM003 SiliconANGLE Safe Superintelligence CEO Daniel Gross joins Meta’s new AI lab Until last week, Gross (pictured) was chief executive of Safe Superintelligence Inc., a well-funded AI startup. He co-founded the company last year with former OpenAI researchers Ilya Sutskever and Daniel Levy.
SM004 Value Add VC $32B SSI Valuation Implies $640M Per Employee, $0 Revenue $32 billion is what Safe Superintelligence is worth on paper as of September 25, 2026 — the same number investors set in April 2025, now backed by $8 billion of cash raised, zero revenue, zero shipped products, and roughly 50 employees
SM005 Vorp Labs Safe Superintelligence (SSI): Funding, NVIDIA Partnership, Verifiable Record | Vorp Labs As of the August 15, 2026 check there is no released model, developer documentation, pricing page, or product surface on any official SSI channel, and no SSI model appears in public model catalogs
SM006 GetLatka Safe Superintelligence Inc. Funding 2026: $2B Raised Safe Superintelligence Inc. reached a $32B valuation in 2025, set during its Seed round.
SM007 GetLatka Safe Superintelligence Inc. Alternatives & Competitors Safe Superintelligence Inc. has 13 employees. Every one of the 20 alternatives below is compared with Safe Superintelligence Inc. on revenue, funding, customers, team size and founding year.
SM008 MindStudio Safe Superintelligence Explained: Ilya Sutskever's SSI and Its 2026 Plans Unlike most AI labs, SSI has released no consumer products, no APIs, and no benchmarks. Its entire public output has been a short mission statement and occasional comments from Sutskever and investors.
SM009 AI Wiki Safe Superintelligence Inc SSI describes itself as the world's first "straight-shot" superintelligence lab, and in April 2025 it was reportedly valued at $32 billion despite having no public product, no published research, and a headcount under 100.
SM010 HokAI Safe Superintelligence: Funding, Founders & Products (2026) Unlike other frontier labs, SSI has never shipped a product, an API, a chatbot, a model card or a research paper. Its one stated output is safe superintelligence itself, an AI system that exceeds human ability across essentially all cognitive tasks while staying reliably aligned with human intent.
SP001 AI Landscape Safe Superintelligence Inc. (SSI) — Alternatives & Similar Tools | AI Landscape Safety-focused AI lab dedicated to building safe superintelligence
SP002 Global VC Megadeal Briefing Safe Superintelligence, Growth $5B SSI competes for talent and compute against OpenAI, Anthropic, Google DeepMind, xAI, and Meta Superintelligence Labs (which Daniel Gross has since joined) — rivals with substantially greater capital and, in most cases, existing products and distribution.
SP003 MindStudio Safe Superintelligence Explained: Ilya Sutskever's SSI and Its 2026 Plans Most specifics about SSI’s architecture and methods are speculation based on limited public statements.
SP004 Value Add VC $32B SSI Valuation Implies $640M Per Employee, $0 Revenue SSI carries 2.4x the valuation of Thinking Machines Lab and roughly six times that of World Labs, and is the only one of the three with no shipped product.
SP005 Vorp Labs Safe Superintelligence (SSI): Funding, NVIDIA Partnership, Verifiable Record | Vorp Labs As of the August 15, 2026 check there is no released model, developer documentation, pricing page, or product surface on any official SSI channel, and no SSI model appears in public model catalogs, including the OpenRouter catalog checked the same day, consistent with the company's stated plan.
SP006 GetLatka Safe Superintelligence Inc. Funding 2026: $2B Raised Safe Superintelligence Inc. employs approximately 13 people as of 2026.
SP007 Distill Intelligence Safe Superintelligence Inc. (SSI Inc.) Competitors List of Safe Superintelligence Inc. (SSI Inc.) competitors
SP008 AI Security and Safety OpenAI vs Safe Superintelligence Inc. (2026): Which AI Safety Org OpenAI is a company based in San Francisco, CA, USA, working on alignment, red teaming, responsible ai, evaluation, and content safety. Safe Superintelligence Inc. is a company based in Palo Alto, CA, USA, with a focus on alignment and responsible ai.
SI001 Startup Fortune Safe Superintelligence is a case study in how venture capital prices silence Within months it reportedly raised $1 billion at a $5 billion valuation despite having no public product, no revenue, and no clear go-to-market motion.
SI002 Silicon Valley Investclub Safe Superintelligence - Funding, Valuation & Leadership No publicly reported financial figures beyond funding and valuation were provided in the notes.
SI003 Trending Topics Nvidia Pours Billions Into Safe Superintelligence - a Startup With No Product and No Revenue The official announcement stays vague on the numbers. Nvidia has made a “substantial” investment, it says, with no financial details disclosed.
SI004 Value Add VC $32B SSI Valuation Implies $640M Per Employee, $0 Revenue $32 billion is what Safe Superintelligence is worth on paper as of September 25, 2026 — the same number investors set in April 2025, now backed by $8 billion of cash raised, zero revenue, zero shipped products, and roughly 50 employees, with the model an investor teased for August still nowhere in sight.
SI005 Vorp Labs Research Safe Superintelligence (SSI): Funding, NVIDIA Partnership, Verifiable Record The size of NVIDIA's 2026 investment is undisclosed by both companies; Bloomberg reported it at 5 billion dollars and TechCrunch's source described it as multiple billions, neither confirmed, so reported-but-unconfirmed figures would lift the total well past the 3 billion dollar confirmed-and-reported equity base.
SI006 GetLatka Safe Superintelligence Inc. Funding 2026: $2B Raised Safe Superintelligence Inc. has raised $2B in total funding across 1 round, most recently a $2B Seed round in 2025.
SI007 MindStudio Safe Superintelligence Explained: Ilya Sutskever's SSI and Its 2026 Plans Unlike most AI labs, SSI has released no consumer products, no APIs, and no benchmarks.
SI008 Multiples Safe Superintelligence Valuation, Funding & Investors As a pre-commercial AI research lab focused on safety, SSI has disclosed no revenue or financial metrics.
SE001 Welcome.AI Safe Superintelligence Inc. - Technology The specific technologies powering Safe Superintelligence Inc. are not detailed on the website. However, the company's approach involves revolutionary engineering and scientific breakthroughs aimed at achieving safe superintelligence.
SE002 Wikipedia Safe Superintelligence Inc. The money will be used to build up more computing power and hire top individuals in the field.
SE003 AI Wiki Safe Superintelligence Inc The company has not articulated intermediate milestones in public, although the founders have suggested that progress will be reviewed internally on a multi-year horizon.
SE004 Value Add VC $32B SSI Valuation Implies $640M Per Employee, $0 Revenue It didn't happen: as of this update, seven weeks past that window, SSI has issued no model, no paper, and no statement about either the rumor or a revised date.
SE005 Vorp Labs Safe Superintelligence (SSI): Funding, NVIDIA Partnership, Verifiable Record As of the August 15, 2026 check there is no released model, developer documentation, pricing page, or product surface on any official SSI channel, and no SSI model appears in public model catalogs, including the OpenRouter catalog checked the same day, consistent with the company's stated plan.
SE006 Latka Safe Superintelligence Inc. Funding 2026: $2B Raised Safe Superintelligence Inc. employs approximately 13 people as of 2026.
SE007 MindStudio Safe Superintelligence Explained: Ilya Sutskever's SSI and Its 2026 Plans Most specifics about SSI’s architecture and methods are speculation based on limited public statements.
SE008 HokAI Safe Superintelligence: Funding, Founders & Products (2026) Unlike other frontier labs, SSI has never shipped a product, an API, a chatbot, a model card or a research paper.
SU001 Enterprise DNA Nvidia's $5B Bet on Safe Superintelligence Explained The gap between frontier research and enterprise adoption is widening. SSI is working on problems that won’t reach your business for years.
SU002 NVIDIA and Safe Superintelligence Inc. Ilya Sutskever’s Safe Superintelligence Inc. and NVIDIA Announce Long-Term Strategic Partnership Safe Superintelligence Inc. (SSI) and NVIDIA today announced a long-term partnership to rapidly accelerate SSI’s strategic growth. NVIDIA has additionally made an investment in SSI.
SU003 MindStudio Safe Superintelligence Explained: Ilya Sutskever's SSI and Its 2026 Plans Unlike most AI labs, SSI has released no consumer products, no APIs, and no benchmarks.
SU004 AgentMarketCap SSI at $32B: Ilya Sutskever's Slow-and-Safe Bet Against the AI Speed Race That company — Safe Superintelligence Inc. (SSI) — has shipped exactly nothing since its founding in June 2024. It has published no models, posted no benchmarks, and given no product demos.
SU005 Value Add VC $32B SSI Valuation Implies $640M Per Employee, $0 Revenue $32 billion is what Safe Superintelligence is worth on paper as of September 25, 2026 — the same number investors set in April 2025, now backed by $8 billion of cash raised, zero revenue, zero shipped products, and roughly 50 employees, with the model an investor teased for August still nowhere in sight.
SU006 Vorp Labs Safe Superintelligence (SSI): Funding, NVIDIA Partnership, Verifiable Record As of the August 15, 2026 check there is no released model, developer documentation, pricing page, or product surface on any official SSI channel, and no SSI model appears in public model catalogs, including the OpenRouter catalog checked the same day, consistent with the company's stated plan.
SU007 Latka Safe Superintelligence Inc. Funding 2026: $2B Raised Safe Superintelligence Inc. has raised $2B in total funding across 1 round, most recently a $2B Seed round in 2025.
SU008 IBM Listen to episode 8 of Mixture of Experts I’ve a friend who’s observing, like, you start out trying to deliver on AGI, and then you find yourself being like, we got to do B2B SaaS.
SR001 cvam.sight Safe Superintelligence: A Research Map Current alignment techniques help with pieces of this problem. They do not compose into an end-to-end proof.
SR002 Constellation Research Safe Superintelligence Inc. launches: Here's what it means It is unclear in which direction the company goes. I will be surprised if they themselves are clear about their milestones
SR003 AI Wiki Safe Superintelligence Inc SSI has published no papers and made no benchmark claims through 2026, leaving outside observers to speculate.
SR004 Value Add VC $32B SSI Valuation Implies $640M Per Employee, $0 Revenue A missed rumor is not the same as a broken promise — SSI never confirmed Baker's timeline in the first place, so there is no company statement to hold it to.
SR005 Vorp Labs Safe Superintelligence (SSI): Funding, NVIDIA Partnership, Verifiable Record SSI publishes no comparable document, so its safety approach cannot be externally compared on thresholds, reviews, or update cadence.
SR006 GetLatka Safe Superintelligence Inc. Funding 2026: $2B Raised Safe Superintelligence Inc. employs approximately 13 people as of 2026.
SR007 Scribd user upload Safe Superintelligence (Ssi) SSI’s primary differentiation is the prioritization of safety architecture from the early stages of development.
SR008 TokenRing AI via FinancialContent The $32 Billion Stealth Bet: Ilya Sutzkever's Safe Superintelligence and the Future of AGI However, critics argue that the lack of iterative deployment makes it difficult to stress-test safety measures in real-world scenarios.
SR009 DigiDAI Ilya Sutskever and Safe Superintelligence's Evidence Gap That design has a potential advantage: fewer incentives to ship capability before controls are ready. It has a corresponding accountability problem: without products, papers, evaluations, or detailed milestones, outsiders have little evidence with which to judge progress.
SV001 Value Add VC $32B SSI Valuation Implies $640M Per Employee, $0 Revenue $32 billion is what Safe Superintelligence is worth on paper as of September 25, 2026 — the same number investors set in April 2025, now backed by $8 billion of cash raised, zero revenue, zero shipped products, and roughly 50 employees
SV002 FutureSearch What will Safe Superintelligence Inc.'s (SSI) valuation be at the end of 2027? As of August 2026, Safe Superintelligence Inc. (SSI)'s last clearly confirmed public valuation remains the $32B post-money mark from its ~$2B round reported in early 2025
SV003 UpMarket Safe Superintelligence Inc. Stock for Accredited Investors | Pre-IPO Shares In July 2026, Safe Superintelligence Inc. received a $5 billion investment from Nvidia at a $32 billion post-money valuation (TechCrunch and Reuters/Bloomberg).
SV004 Vorp Labs Safe Superintelligence (SSI): Funding, NVIDIA Partnership, Verifiable Record On July 27, 2026, NVIDIA announced a long-term partnership with and investment in Ilya Sutskever's Safe Superintelligence Inc., a lab that has released no model, no API, and no published research.
SV005 Latka Safe Superintelligence Inc. Funding 2026: $2B Raised Safe Superintelligence Inc. reached a $32B valuation in 2025, set during its Seed round.
SV006 Jewish News Syndicate Israel-raised pioneer’s Safe Superintelligence startup raises $2b Artificial intelligence researcher Ilya Sutskever, 38, has become the primary reason venture capitalists are investing approximately $2 billion into his secretive company Safe Superintelligence (SSI), based in Tel Aviv and Palo Alto, Calif., The Wall Street Journal reports, citing people familiar with the matter.
SV007 Tanipyon Research Institute AI News Summary: NVIDIA Invests in 'Safe Superintelligence', Market Wary of OpenAI's Massive Data Center Plan NVIDIA and Safe Superintelligence Inc. (SSI) have announced a long-term strategic partnership. NVIDIA will invest in SSI and supply its next-generation 'Vera Rubin' systems. This is expected to increase SSI's computing power by approximately 10 times. The investment amount has not been disclosed.
SV008 AI Wiki Safe Superintelligence Inc SSI describes itself as the world's first "straight-shot" superintelligence lab, and in April 2025 it was reportedly valued at $32 billion despite having no public product, no published research, and a headcount under 100.