The robotics story Nvidia tells today did not appear with the humanoid boom. Read the filings. The FY2020 10-K, surfaced via EdgarBeast and filed at sec.gov, lists 'Jetson AGX for robotics and other embedded use' as a product line — years before embodied AI became a keynote staple.
That phrase — 'for robotics and other embedded use' — recurs across subsequent annual reports nearly verbatim, including the FY2024 10-K, which again describes Jetson robotics and other embedded platforms. When a company repeats the same product framing in legal filings for half a decade, it is telling you the position is strategic and durable, not opportunistic.
The engineering reason this matters: a robot's onboard brain is an embedded, power-constrained inference problem. It is not the same as a datacenter GPU and not the same as a phone chip. Jetson is Nvidia's answer to that specific niche — enough AI compute to run perception and control on the robot itself, in a thermal and power budget a mobile machine can carry.
So the durable claim from the filings is that Nvidia intends to own the embedded-inference layer of robotics the way it owns datacenter training. A humanoid, an autonomous mobile robot, or a drone all need an on-device brain, and the 10-Ks have been pointing at Jetson as that brain since before most humanoid startups existed. The arms-dealer position spans both the training cloud and the embedded edge.
Count the years, not the keynotes. The continuity in the sec.gov language tells you Nvidia did not pivot into robotics — it has been positioned at the embedded brain of robotics all along, waiting for the rest of the field to catch up. Filing history via EdgarBeast.