Nvidia's acquisition of the open-source model hub for $12.9 billion — after rejecting the startup's own $500 million offer just a year earlier — signals that the biggest AI infrastructure companies now absorb potential kingmakers before they ever ring a Nasdaq bell.
For years, the theory of the case for open-source AI infrastructure startups was straightforward: build the platform millions of developers depend on, grow revenue, and eventually go public on the strength of your neutrality — the one thing no cloud giant or chipmaker could credibly claim. Hugging Face, the seven-year-old repository that hosts three million models, a million applications and half a million datasets used by more than 18 million developers, according to TechCrunch, was the closest thing the industry had to that story.
Instead, Nvidia is buying it for $12.9 billion, as PitchBook reports, confirmed by CNBC on Aug. 27 citing The Information. The deal caps a run in which Nvidia rejected — or rather had rejected — a $500 million offer for Hugging Face just a year earlier, according to the Financial Times as cited by TechCrunch, before the price climbed roughly 25x. That trajectory is the market map in miniature: the company that used to be a mere GPU vendor to AI developers is now the buyer of last resort for the platforms those developers can’t live without, and the public markets are increasingly optional for the sellers.
a price Nvidia barely notices
$12.9 billion sounds enormous until it’s set against Nvidia’s balance sheet. Reuters Breakingviews calculated the deal at roughly 0.2% of Nvidia’s approximately $5.5 trillion market capitalization, according to inspenet.com — a rounding error that explains why Nvidia can write a check whose direct financial return, by the outlet’s own framing, is still difficult to calculate. Hugging Face, for its part, was clocking $150 million in annualized revenue as of last month, per The Information, and told TechCrunch in July it was getting “close to profitability.” The company had raised just over $395 million total, according to Crunchbase, with its last round in 2023 — $235 million led by Salesforce Ventures, joined by Google, Amazon, IBM and Nvidia itself — valuing it at $4.5 billion, per inspenet.com. Nvidia, in other words, was already an investor before it became the owner.
who gains, who pays
The winners are legible: Hugging Face’s prior backers cash out at nearly triple the 2023 valuation, Nvidia’s Jensen Huang gets a direct pipeline into the developer community that decides which hardware open models run on, and CEO Clem Delangue gets what he called the ability to “think long-term,” telling EE Times the deal would accelerate Hugging Face’s goal of 100 million builders from its current 18 million. The less obvious payer is the open-source ecosystem’s claim to neutrality itself. Nvidia is Hugging Face’s single biggest contributor, having released more than 500 models and 250 open datasets on the platform, Huang said in a blog post cited by TechCrunch, in which he pledged that “Nvidia compute will not be required to build on or deploy through Hugging Face.” Nvidia VP Justin Boitano, fielding pointed questions about neutrality at the deal’s press conference, said Nvidia wants developers able to “access and connect the best open models into and across every cloud provider, every neocloud” — a commitment that is also, unavoidably, a commercial strategy: an open ecosystem that runs well on Nvidia hardware sells more Nvidia hardware.
the defensive logic
The acquisition reads as much as a hedge as a land grab. OpenAI, Google and Anthropic are all developing their own chips and infrastructure to reduce Nvidia dependence, inspenet.com reported — a concentration risk that, if a handful of firms end up controlling the market, could strengthen their leverage to negotiate down or route around Nvidia’s components. Owning the platform that connects thousands of smaller developers and companies to compute is a counterweight: better, from Nvidia’s position, to sell infrastructure broadly than depend on a few giants who might not need it. Siddy Jobe, a fund manager at Eonopolis Exponential Technologies, told CNBC’s Squawk Box Europe that Hugging Face “fits perfectly” into what he described as Nvidia’s five-layer stack ambition, “going from energy to foundational models and also to applications.” The deal follows a pattern: Nvidia’s $20 billion licensing agreement with chip startup Groq and a $6 billion deal with coding startup Poolside to build open models, per the Wall Street Journal as cited by TechCrunch, plus more than $50 billion the company says it has infused into AI frontier labs overall.
what to watch
The neutrality pledge is testable, not just asserted. Watch whether Nvidia-specific runtimes crowd out open alternatives like vLLM and SGLang that Boitano cited as coexisting today, whether Hugging Face keeps folding in non-Nvidia-aligned labs such as France’s Mistral and Germany’s Black Forest Labs as Delangue promised, and whether Hugging Face’s cybersecurity posture — the platform was breached earlier this year, with Delangue blaming “engineering mistakes” and separately disclosing that an unreleased OpenAI model had also breached the platform, per TechCrunch — holds up under new ownership. The bigger structural question for the data and AI-infrastructure economy is whether Hugging Face’s exit becomes the template: strategics with trillion-dollar balance sheets absorbing category-defining platforms at valuations no public market discipline ever tested, leaving fewer neutral infrastructure companies to go public at all.
Reuters Breakingviews points out that the approximately $12.9 billion represents about 0.2% of a market capitalization of roughly $5.5 trillion. This comparison helps to understand why Nvidia can undertake an acquisition whose direct financial return is still difficult to calculate.
