Read Nvidia's own SEC filing and its own product announcements from the same week, and you find the company arguing both sides of a fight with itself.
In one document, Nvidia warns investors that a White House crackdown on Chinese-origin models — restrictions that could limit its ability to support apps built on "DeepSeek, Qwen or Kimi" — would have a material impact on its business.
In the other, per CNBC, it ships day-zero optimizations reportedly for those exact models, touting a "local AI initiative" tuned for DeepSeek's V4 Flash and Alibaba's Qwen 3.8.
That is not a company hedging. That is a company doubling down on the thing it just told the SEC could blow up.
To see why, start with the number that broke this week. Nvidia has agreed to buy Hugging Face for $12.9 billion, per The Information — against annualized revenue of roughly $150 million. That is about 86 times sales, per TechSpot, though other reports cite roughly 80x.
You cannot get there with a spreadsheet. "A price of $12.9 billion against annualised revenue of approximately $150 million is not a software valuation; it is a strategic premium on distribution," said Sanchit Vir Gogia, chief analyst at Greyhound Research.
His sharper framing is the one to hold onto: "NVIDIA already supplies much of the road beneath AI. Hugging Face is the junction where developers choose the road."
NVIDIA already supplies much of the road beneath AI. Hugging Face is the junction where developers choose the road.
So the price is not an acquisition multiple. It is a toll booth on a road every AI developer already drives — 13 million users, more than two million public models, the de facto GitHub of AI, per Hugging Face's own stated 2025 community figures.
Now the strategic reveal. Why does the world's most valuable chipmaker need to own the on-ramp?
Because the customers it depends on most are trying to build a bypass. OpenAI, Anthropic, Google, and Amazon are all developing their own silicon to escape Nvidia's leverage — and this week OpenAI reportedly released benchmarks for its in-house "Jalapeño" chip, claiming it beats Nvidia's flagship Blackwell.
The closed frontier labs are the exact accounts most motivated to leave. So Nvidia's counter is to make sure there is always a thriving alternative to those labs — open models that anyone can download, fine-tune, and self-host, all of which still have to run on somebody's GPUs.
That is why Nvidia is spending a separately reported $26 billion over five years, per ainvest.com, to build open-weight models and give them away free. The give-away is the distribution mechanism; the chip underneath is the product.
And it is why Nvidia is deliberately propping up the open-model ecosystem — including the Chinese models that currently lead it. The math is cold: the more open models thrive, the harder it is for any single closed lab to sever demand for Nvidia hardware.
The employee framing CNBC got, per CNBC, states the case plainly. "Every model should run best on the U.S. technology stack," an unnamed Nvidia employee said, adding that because "developers will choose the stack that the model is optimized for," optimizing for the American stack is "critical."
Notice what that logic requires. It arguably requires Nvidia to optimize Chinese models — DeepSeek, Qwen, Z.ai's GLM — precisely so those models pull developers onto CUDA rather than onto Huawei's Ascend chips, which are chasing the same optimizations.
Which brings us back to the contradiction that makes this whole flywheel fragile.
The same Chinese-model support that fuels the strategy is the risk Nvidia flags in its own 10-Q. Washington is reportedly weighing rules that would limit Nvidia's ability to support "third-party applications and models built on open-source foundation models originating in China."
That is not an abstract macro risk. It is the concrete, named contingent exposure — the DeepSeek/Qwen/Kimi language — that the August 26 filing puts in black and white, and it is the disclosure worth tracking quarter to quarter.
Nvidia is not fighting it alone. It joined Microsoft, Meta, Palantir and more than 20 companies in a July letter urging policymakers to avoid "premature restrictions" on open-weight models, and Jensen Huang personally signed on.
But lobbying is a tell, not a shield. You do not organize a 25-company coalition against a risk you think is remote.
And there is a national-security counter-argument gaining ground in Washington that no letter erases. "If Chinese AI technology becomes the default for developing countries, those countries may be more likely to align themselves politically with Beijing," said Daniel Remler of the Center for a New American Security.
So Nvidia's pitch — every model on the American stack keeps the world on America — is one framing. The other is that Nvidia is optimizing and distributing Chinese models at global scale, which is not obviously what a crackdown-minded administration wants to hear.
Then layer the acquisition on top of the export risk, and you get the second vulnerability: trust, and the regulators who police it.
Hugging Face's entire value came from neutrality — it sat above the hardware fight, supporting AMD, AWS, and Google alongside Nvidia. That is why developers used it as the honest broker.
Recall that Hugging Face rejected a $500 million Nvidia investment late last year at a $7 billion valuation because it "didn't want a dominant investor that could sway its decisions."
The tell is that the concern hasn't changed — only the ownership stake has. A single influential backer worried them; full ownership is that worry taken to its logical end.
And there is a ghost in this deal. Nvidia's proposed $40 billion Arm acquisition collapsed in 2022 under global antitrust pressure, per TechSpot.
The mechanism to watch isn't a ban. It's gravity. Gogia's warning is that the platform "does not need to remove a rival; it only needs to let gravity make one path progressively easier."
He draws the line cleanly: "Legal openness is measured by rights. Practical openness is measured by meaningful downstream choice." Search ranking, default routes, and optimization can steer developers without changing a single license.
Which lands on what this actually means for you.
If you own Nvidia, the flywheel spins — that much is not in dispute; this is a company that, per TechSpot, just posted a $96.2 billion quarter. The open question is whether Washington or a regulator jams it, and the DeepSeek/Qwen/Kimi disclosure is where you'll see the first crack.
If you run production workloads on Hugging Face, the exposure is more immediate, and the analysts covering enterprises are unusually blunt about it.
"Hugging Face earned enterprise trust by sitting above the hardware fight," said Bhupendra Chopra of Kanerika. "Owned by one of those hardware vendors, that changes."
His action item is the one to steal: teams "running production workloads should be getting written commitments now on model portability and support parity." Not after the deal closes — now, while you still have leverage.
Because the strategy doesn't need to lock you in with a wall. It only needs to make the Nvidia path easier every quarter until, one day, it's the only road you can still see.
Get the written exit path before gravity picks the road for you.