✦ Robotics & Physical AINarrow moat

Nvidia (NVDA) — the future bets

Nvidia doesn't build robots — it sells every robot maker the brain, and profits whichever body wins.

Nvidia's robotics strategy is its AI strategy with legs: it does not build robots, it sells every robot maker the brain. Jetson Thor, generally available since August 2025, packs 7.5 times the AI compute of the prior generation into a robot-sized computer1, and the Isaac GR00T foundation models — plus an open humanoid reference design — hand the industry a starting point the way CUDA once handed one to AI researchers2. The adopter list is the tell: Amazon Robotics, Boston Dynamics, FANUC, Agility, Hitachi — incumbents and startups alike building on the same stack.

Change in papers citing each chip, 2025 vs 2024Jetson (robots, edge)+33%All NVIDIA chips-13%Air Street Press, State of AI Compute Index v4, June 2025
Research on NVIDIA's robot computer grew 33% in a year when NVIDIA-citing papers overall fell 13%.

The position is the arms dealer's, and it is deliberately agnostic: Tesla bets vertically on Optimus, a dozen startups bet on themselves, and Nvidia profits whichever body wins. The revenue today is a rounding error inside the automotive-and-robotics line — this is the earliest-stage bet on the page, priced accordingly. What would make it real is fleets: humanoids and warehouse robots deployed by the thousand on Thor, not demonstrated by the dozen at trade shows. Watch the deployment announcements and the segment disclosures — the quarter 'robotics' earns its own revenue line in the filings is the quarter this option is in the money.

Moat trajectory: Widening

Widening as the platform standardizes: Thor's general availability plus open GR00T models put Nvidia's stack under most serious robot programs — the CUDA playbook, re-run with legs. Still developer kits more than fleets, which is the gap to close.

The number that tests this moat
Reported
Edge Computing growth excluding the PC slowdown
+27% in Q2 FY2027 and +29% in Q1, led by Blackwell workstations

Robotics revenue is not broken out; it sits in Edge Computing with PCs and workstations. Growth that holds while consumer PCs slow is the only public sign of physical AI adding revenue.

Source: NVIDIA Q1 and Q2 FY2027 CFO commentaries ↗
References
  1. ReportedJetson Thor GA (Aug 2025): 7.5x the AI compute of the prior generation.
    NVIDIA Newsroom — Blackwell-powered Jetson Thor generally available (Aug 2025): 7.5x the AI compute and 3.5x the energy efficiency of AGX Orin; 2,070 FP4 teraflops, 128GB — Aug 2025 · publ. Aug 25, 2025 · source ↗
  2. ReportedGR00T open models + reference designs; Amazon Robotics, Boston Dynamics, FANUC, Agility, Hitachi build on the stack.
    NVIDIA Newsroom — new physical-AI models released as global partners unveil next-generation robots; Isaac GR00T open models and reference designs; adopters incl. Amazon Robotics, Boston Dynamics, Agility, FANUC, Hitachi — 2025–2026 · publ. 2026 · source ↗
Sources
Generated September 18, 2026