NVIDIA has partnered with Hugging Face to ensure its role in physical AI development. Source: NVIDIA
The physical AI developers determined not to rely on NVIDIA have had some wins lately. OpenAI shipped its own chip. Anthropic stood up an in-house silicon team with salaries that make FAANG recruiters wince.
Also, Advanced Micro Devices acquired World Labs for $8.2 billion, Qualcomm has collaborated with Liquid AI on embedded agents, and Skild AI launched the S1 foundation model. In addition, Google DeepMind integrated Gemini Robotics 2 into Apptronik‘s Apollo 2 humanoid.
And then, almost politely, NVIDIA said it plans to spend $12.9 billion — the biggest acquisition in the company‘s history — buying Hugging Face, the place where the planet’s developers go to pick which model to run. Cute war. NVIDIA just bought the battlefield.
Chasing chips focuses on the wrong layer
Several headlines have read “Labs race to cut NVIDIA loose.” Adorable framing. Here’s the actual scoreboard: While everyone else designed chips, NVIDIA designed an exit ramp that leads straight back to it, no matter which chip wins. That’s not defense. That’s a company that already read the ending and is waiting, patiently, for the rest of the industry to catch up.
The chip numbers are genuinely impressive, to be fair. OpenAI spent nine months with Broadcom on Jalapeño. It is reportedly the fastest tape-out in the history of advanced semiconductors, shipping by year’s end.
Anthropic followed with its own silicon team, posting hardware roles at up to $485,000, stacked on a multi-gigawatt Tensor Processing Unit (TPU) deal it already had running with Google and Broadcom. Google has had TPUs for a decade.
AWS has Trainium. Meta has MTIA. Everyone is racing to stop writing Jensen Huang a check.
None of these very expensive chips touch the place where the other 200,000 companies that will never build their own silicon actually go to find, fine-tune, and ship a model. That layer — the real on-ramp for the rest of the world’s AI adoption — just got quietly annexed by the company everyone claims to be staging a jailbreak from. NVIDIA didn’t flinch; it bought the prison.
Register now and help us celebrate 20 years of RoboBusiness!‘Sovereign’ a clue to NVIDIA’s strategy
Notice the word NVIDIA has kept reaching for in every earnings call so far this year: “sovereigns.” Not developers first. Not enterprises first. Sovereigns — nation-states spending public money on “AI independence” so they’re not permanently on their knees to a handful of U.S. labs.
No government building sovereign AI wants to bet its national compute on OpenAI’s chip, Anthropic’s silicon, or any single lab that might someday become a geopolitical liability. So NVIDIA stepped in with the pitch: “Don’t worry, we’re neutral. Pick any model, any framework, any cloud you like” — a rousing speech about freedom of choice, delivered by the company that just bought the store where all the choices live.
That’s the actual chess move. OpenAI and Anthropic are optimizing their own cost of goods. NVIDIA is optimizing for sitting underneath everyone’s cost of goods, forever, regardless of who wins the fight it’s graciously letting other people have. Nobody at Nvidia will ever put that in an earnings call. They don’t have to. The acquisition makes the argument for them.
Here’s the part nobody wants to say out loud: Even a great custom chip takes years to match the volume, yield, and software maturity of a platform NVIDIA has spent two decades pouring into CUDA.
Jalapeño’s nine-month tape-out is a real engineering flex. But by every account, it’s still a narrow inference chip built for OpenAI’s own workloads — not a platform anyone else is building a business on top of. Anthropic’s silicon team is, charitably, a hiring page with a mission statement, sitting inside a “multi-chip strategy” that still runs mostly on NVIDIA and Google hardware today.
Meanwhile, NVIDIA just bolted on a distribution channel with Hugging Face that gets more daily developer traffic than either lab’s entire platform combined.
NVIDIA says it has the most repos and the second-most homegrown models on Hugging Face. Source: Hugging Face
Sovereign implications get sharper for robotics
The implication gets sharper in physical AI. A robot cannot wait for a distant API to deliberate while it navigates a warehouse, handles sensitive equipment, or makes a safety-critical decision on a factory floor.
It needs local inference, real-time control, sensor data that stays inside the operation, and a deployment stack that does not collapse because a cloud region hiccupped or a vendor changed its terms. That is precisely where “sovereign” stops being a procurement slogan and becomes an operational requirement.
But the same landlord problem follows the robot onto the factory floor. The company that owns the training stack, simulation environment, edge runtime, developer tools, and model marketplace is not merely selling silicon—it is setting the terms for how physical intelligence gets built and deployed.
Physical AI makes hardware choice matter more, not less. It also makes genuine portability across models, compute, and deployment environments far more valuable.
None of this changes much for the companies shipping product to customers — if anything, it sharpens the point. Airrived last month launched its own sovereign AI platform, positioning itself as the agentic OS on a mission to democratize AI and turn every business user into an AI expert.
That’s the other, more literal definition of “sovereign” — not a government hedging its national compute bet on a chip vendor, but a business user who doesn’t need a data center, a silicon roadmap, or a forward-deployed engineer to own and run their AI.
NVIDIA’s version of sovereignty comes with a landlord attached. The application-layer version doesn’t care whose logo is silk-screened onto the silicon underneath it. It cares whether inference is fast, cheap, and boringly reliable, and it is happy to let $2 trillion-adjacent companies fight over who eats the infrastructure bill. That’s the quiet advantage of building lean above the war instead of funding a side in it: The chip fight is enormously expensive to win and completely free to ignore.
Down where the silicon actually gets baked, this isn’t shaping up to be the underdog story the press releases want it to be. OpenAI and Anthropic are spending billions to buy independence from NVIDIA’s hardware. NVIDIA just spent $12.9 billion buying insurance against ever losing either of them — by owning the exact place their models must show up regardless of what chip runs them.
Call it a war if you want. From here, it looks like NVIDIA collecting rent on both sides of the battlefield, smiling, and asking if anyone else would like to diversify.
About the author
Anurag Gurtu, co-founder and CEO of Airrived, is an entrepreneur and product leader with more than 20 years of experience across cybersecurity and AI. He has built three generative AI companies, including one that reached unicorn status.
Gurtu is known for turning advanced AI into scalable, real-world platforms, and he previously contributed to successful exits, including acquisitions by Splunk, Tripwire, Sun Microsystems, and FireEye, a public company. He holds an M.S. from the University of Southern California and CISSP certification.
Tell Us What You Think!