Two founders can't agree on who called whom.
Jensen Huang says Clem Delangue came to him, "as he considered the next chapter of Hugging Face." Co-founder Thomas Wolf tells a different story on LinkedIn: Huang made the offer, pitching Delangue on building the hub into an open, hardware-neutral platform. Small discrepancy, but it points at the thing that matters — this deal is a story both sides need told a specific way, and neither version is really about the money.
Which is the place to start, because the money makes no sense on its own.
Nvidia — a company reportedly famous for pricing discipline, for squeezing margin out of every rack and gigawatt — just agreed to pay $12.9 billion for Hugging Face, a platform with roughly $150 million in annual revenue. That's 86 times sales. For comparison, when Microsoft bought GitHub in 2018, the closest analog anyone can name, it paid $7.5 billion.
So the world's most disciplined hardware company overpaid, by a lot, for a software platform it swears in the same breath it will never tilt in its own favor. "NVIDIA compute will not be required to build on or deploy through Hugging Face," Huang wrote. The platform stays "open and neutral," Nvidia's Justin Boitano promised in the press briefing, "interoperates with every cloud provider, every hardware company in the world."
When a hardware company pays 86x for software it vows not to weaponize, the thing it's buying isn't on the invoice.
Here's the counterintuitive part: take the neutrality pledge at face value. It's probably true, in this author's estimation. And that's exactly what makes the deal work.
The instinct is to assume Nvidia bought the App Store of AI to steer 18 million developers toward its chips. But it doesn't need to. Open models already land on Nvidia hardware by default — anyone hitting an API from OpenAI or another provider has, historically, mostly ended up on Nvidia silicon. Huang gave Axios the whole thesis in one line: "Free AI should be great for hardware."
If the flywheel already spins in your direction, you don't rig the wheel. You buy the instrument that tells you which way it's turning.
That instrument is the point. Forrester's Naveen Chhabra put it plainly: Nvidia now gets to see "which models are trending, what datasets customers are downloading, and the architectures that are gaining traction weeks before they hit mainstream tech news." And, he added, "data-driven foresight into what next-generation chip architectures they need to build based on developer habits."
Read that twice. Nvidia didn't buy a store. It bought a sensor — a live feed on the open ecosystem's design choices, wired directly into the group that decides what the next generation of chips should accelerate.
Nvidia didn't buy a store. It bought a sensor — a live feed on the open ecosystem's design choices, wired directly into the group that decides what the next chips should accelerate.
Neutrality isn't the price Nvidia pays to keep developers happy. Neutrality is what keeps the sensor calibrated. The moment Hugging Face visibly favors Nvidia toolchains, developers migrate, the data curdles, and the $12.9 billion buys a corrupted readout. The business model requires the platform to stay honest. That's the strongest kind of promise — the kind aligned with the buyer's self-interest.
Now widen out, because the sensor is defending against a specific threat.
The customers Nvidia can't design around are the giants building their own silicon. Google has its TPUs, Amazon its Trainium and Inferentia, OpenAI is designing accelerators, and now Anthropic is too. Every one of those chips is a future gap in Nvidia's revenue.
The open crowd is different. Open models run across many clouds, inside companies, at universities, in government agencies — a customer base that doesn't build its own chips, that lives on rented GPUs and shared repositories. That is the base Nvidia just fortified.
Buying Hugging Face isn't offense against the closed labs. It's insulation. Nvidia is bolting down the one large, growing, multi-cloud constituency that has no incentive to defect to custom silicon — and making sure it can watch that constituency's every move.
The compute side of the deal reinforces the read. Hugging Face doesn't just host models; it rents the GPUs to run them, plugging into Nvidia's Lepton marketplace, which routes jobs to partners like CoreWeave, Lambda, and Nebius.
Nvidia tried owning the rental layer directly once, with DGX Cloud, and pulled back rather than poach customers from its biggest buyers. Lepton is the compromise: Nvidia as the router, not the landlord. A giant developer hub feeding jobs into that router is a convenient new sales channel — especially with a reported $36 billion in commitments to cloud partners whose terms ease as they resell capacity.
And the landlords themselves are consolidating fast. Crusoe reportedly raised over $3 billion at a roughly $30 billion post-money valuation, co-led by Atreides and Valor, with Mubadala in, according to sources cited by Bloomberg. Days apart, it reportedly signed a five-year, roughly $13 billion deal to supply Jane Street with GPUs and infrastructure, according to sources cited by Bloomberg.
Sit with that pairing. A trading firm is now a multibillion-dollar anchor tenant for AI compute, and a compute provider is reportedly worth $30 billion. The plumbing of open AI — models, hosting, financing — is settling into a few specialized owners and their financier tenants. Neutral, in this market, increasingly means "owned by someone big."
So what should a developer actually do with this?
First, understand what "open and neutral" is and isn't. It's a blog post and a briefing quote. It has no enforcement mechanism — no independent board, no structural firewall, no penalty if it erodes. It rests entirely on Nvidia's continued judgment that neutrality pays better than favoritism. That judgment is sound today. Judgment changes.
Second, watch the deal's actual timeline, not its press cycle. According to The Decoder citing the SEC filing, it won't close until the first half of 2027, pending regulatory approval. There's also an equity retention program of up to $1 billion to keep Hugging Face staff on board — which tells you Nvidia knows the people are the asset, and that people leave.
So run two clocks. The retention clock: do the engineers who built the neutral public square stay long enough to defend it? And the routing clock: does Lepton quietly become the path of least resistance, the default that turns "you can use any cloud" into "but this one is one click and pre-configured"?
Neutrality dies not by decree but by default settings. Watch where the friction goes.
The honest summary is uncomfortable. The single most important open platform in AI — 3 million models, 500,000 datasets, 200,000 companies — would, once the deal closes, have one owner, and that owner is the company with the most to gain from watching everything that moves across it.
The good news for developers is real: today, the incentives line up. Nvidia genuinely profits more from an open, multi-cloud, thriving Hugging Face than from a captured one. Free AI really is great for its hardware.
The bad news is that "the incentives line up" is not the same as "you are protected." It's a snapshot, not a contract. The 86x multiple isn't generosity and it isn't a mistake — it's the cost of buying surveillance over the ecosystem that keeps the flywheel spinning, plus insurance against the customers who are learning to build without you.
Don't read the pledge. Read the retention numbers and the routing defaults. That's where you'll find out what "neutral" actually meant.