NVDA
Not just another chatbot binge
Nvidia keeps showing up everywhere in AI land, but this one’s a little different: the next big customer for its Vera Rubin GPUs isn’t a cloud titan or an app company. It’s Figure AI, a humanoid robotics startup that wants enough compute to teach robots how to deal with the messy, non-text world we all live in.
Figure’s new partnership with AI cloud provider Nscale is the headline-grabber. The plan calls for up to 100,000 GPUs, with an initial $3.5 billion compute commitment and a path that could climb past $6 billion. That’s not pocket change; that’s “we need a bigger truck just to carry the server bill” money.
Why this matters for Nvidia
The market has mostly treated Nvidia like the toll booth for generative AI: if hyperscalers and chatbot builders are spending, Nvidia gets paid. But robotics changes the script. Humanoid robots need constant training on visual, behavioral, and real-world data, which means they can chew through compute like a teenager raiding the fridge.
Jensen Huang is framing this as a “robotics flywheel”: train on Vera Rubin, validate in Isaac Sim, then deploy on Nvidia-powered robots. Translation: Nvidia isn’t just selling chips; it’s trying to own the entire physical-AI stack.
The bigger investor takeaway
If Figure’s bet is a preview of where AI infrastructure spend is headed, Nvidia’s customer base could get a lot less dependent on chatbots and a lot more diversified. That’s good news for durability, because physical AI may become one of the next long-running demand engines for Nvidia’s platform.
Big picture: the AI trade isn’t just about software that talks. It might soon be about machines that walk, lift, and maybe occasionally trip over your dog.