Main主页 / companies / public / nvidia
company
NVIDIANVIDIA NVDA
Country国家 United States美国
Bottlenecks瓶颈 P04 Edge inference边缘推理 · P08 Simulation & synthetic data仿真与合成数据 · P10 Predictive world models预测式世界模型 · P11 Generalisable VLA policies可泛化的 VLA 策略
Market cap市值 Live, via TradingView.实时数据，来自 TradingView。
Jetson/Thor edge compute, Isaac simulation, the GR00T humanoid model, the Cosmos world model, Omniverse and the CUDA runtimes.
Jetson／Thor 边缘算力、Isaac 仿真、GR00T 人形模型、Cosmos 世界模型、Omniverse 与 CUDA 运行时。
What it doesJetson/Thor edge compute, Isaac simulation, the GR00T humanoid model, the Cosmos world model, Omniverse and the CUDA runtimes.
在干什么Jetson／Thor 边缘算力、Isaac 仿真、GR00T 人形模型、Cosmos 世界模型、Omniverse 与 CUDA 运行时。
Name sheetNVIDIA
Why it mattersThe de facto compute, simulation and model tooling layer. But it is not the whole robot operating system: Isaac, ROS 2, controllers and safety systems occupy different layers and must not be conflated.
为何重要事实上的算力、仿真与模型工具层。但它并非完整的机器人操作系统：Isaac、ROS 2、控制器与安全系统分处不同层级，不应混为一谈。
Caveat
Return +18.04%, confidence 85. Robotics is still not separately disclosed and remains immaterial against data centre: hold it as free optionality, never as a robotics pure play.
The reporting structure now invites exactly the error that rule forbids: beginning Q1 FY2027 NVIDIA combined the former Gaming, Professional Visualization and Automotive segments into a single Edge Computing segment, which also captures revenue from devices for agentic and physical AI, AI-RAN base stations and robotics. Edge Computing growth is mostly PCs, consoles and telecom base stations; it is not a robotics number and must never be quoted as one.
Jetson AGX Thor delivers 2,070 FP4 TFLOPS with 128GB at 40–130W, and the company states 7.5x the AI performance and 3.5x the efficiency of AGX Orin, but the volume T3000 and T2000 modules announced in July 2026 do not ship until Q1 2027, so any 2026 P04 revenue thesis is about design wins, not shipments.
风险提示
回报 +18.04%，置信度 85。机器人仍未单独披露，且相对数据中心并不重大：应作为免费期权持有，而非机器人纯标的。
当前的报告结构恰恰容易诱发这条规则所禁止的错误：自 FY2027 Q1 起，英伟达把原游戏、专业可视化与汽车三个分部合并为单一的“边缘计算（Edge Computing）”分部，其中同时包含代理式与物理 AI 设备、AI-RAN 基站与机器人的收入。边缘计算的增长主要来自 PC、游戏主机与电信基站，它不是机器人数字，绝不可引作机器人数字。
Jetson AGX Thor 提供 2,070 FP4 TFLOPS、128GB、40–130W，公司称其 AI 性能为 AGX Orin 的 7.5 倍、能效 3.5 倍；但 2026 年 7 月发布的量产型 T3000／T2000 模组要到 2027 年 Q1 才供货——任何以 2026 年为期的 P04 收入论点，讲的都是定点而非出货。
ProductJetson/Thor, Isaac, GR00T, Cosmos, Omniverse, CUDA runtimes and tooling
Value accrualCUDA and developer ecosystem can become a cross-robotics compute and tooling layer
Key riskQ1 FY2027 total revenue $81.6bn; robotics is not separately disclosed and remains immaterial versus data center, so do not use NVDA as a pure robotics comp.
Bottlenecks瓶颈 P04 Edge inference边缘推理 · P08 Simulation & synthetic data仿真与合成数据 · P10 Predictive world models预测式世界模型 · P11 Generalisable VLA policies可泛化的 VLA 策略
- Class
- public
- Country
- US
- HQ
- Santa Clara, United States
- Layer
- Compute / Simulation / Foundation models
- Systems
- blood · brain
- Caveat stance
- pass
- Ticker
- NVDA
- Exchange
- Nasdaq
- Purity
- Low / highly diversified
- Valuation tier
- 1
- Last verified
- 2026-08-12
- Confidence note
- High facts / Moderate robotics exposure
- Sleeve
- A
- Sources
- S-PUB-003
- Translated fields
- what_it_does_en · why_it_matters_en · caveat_en
- Chain position
- upstream
COMPANY RESEARCH公司研究 Commercial evidence商业证据 Expand +展开 + Collapse −收起 −
- Customer and task
- 客户与任务
- Not yet assessed in this structured record.本结构化记录尚未评估。
- Deployment and payment
- 部署与付款
- Not yet assessed in this structured record.本结构化记录尚未评估。
- Task results and human assistance
- 任务结果与人工协助
- Not yet assessed in this structured record.本结构化记录尚未评估。
- Cost and repeat purchases
- 成本与复购
- Not yet assessed in this structured record.本结构化记录尚未评估。
- What the evidence cannot establish
- 证据的局限
- Not yet assessed in this structured record.本结构化记录尚未评估。