Hugging Face in Talks to Sell to Nvidia for $13B — After Rejecting Its $500M Offer
By LW Forge — maintainer of LLM Scout · Updated August 29, 2026
Hugging Face, the open-source hub that hosts most of the model weights, datasets, and Spaces the open-weight LLM ecosystem runs on, is reportedly close to being sold — and the buyer circling the deal is the same company it turned down less than a year ago.
What's actually been reported
On August 24, 2026, Business Insider reported that Hugging Face was in talks to be acquired at a valuation of roughly $13 billion, without naming the buyer. Two days later, The Information reported that the acquirer is Nvidia, at a price closer to $12.9 billion. As of this writing, neither company has confirmed anything publicly, and both outlets are explicit that no signed agreement exists — Business Insider's own reporting notes talks "could still fall apart." Treat this as a developing story, not a done deal, until Hugging Face or Nvidia say otherwise.
The reversal: Nvidia was rejected once already
The twist is what happened in late 2025. Nvidia offered roughly $500 million for a stake in Hugging Face that would have valued the company at about $7 billion — and Hugging Face said no. According to TechCrunch's reporting, the company turned down the extra capital specifically because it didn't want a single investor with outsized influence over its decisions. Hugging Face's role — a neutral place where OpenAI, Meta, Google, DeepSeek, Alibaba, and Nvidia itself all publish models side by side — depends on that neutrality being credible.
Less than a year later, the number on the table has nearly doubled and the same company is reportedly the one at the negotiating table. That jump says something about how fast the ground moved under Hugging Face's business, not just about Nvidia's appetite.
Why the number nearly doubled
A few data points explain the trajectory, per Business Insider's and TechCrunch's reporting:
- Hugging Face's last priced round was in 2023: a $235 million Series D led by Salesforce Ventures, with Alphabet, GV, and IBM Ventures participating, at a $4.5 billion valuation.
- The company's annualized revenue is reported around $150 million, and CEO Clem Delangue has said Hugging Face is close to profitability and is optimizing for long-term sustainability rather than a maximized fundraise.
- The wider infrastructure layer of the AI stack has been getting bid up all year — Stripe's roughly $7 billion acquisition of OpenRouter earlier in August 2026 is the closest comparable, another deal for a company that sits between developers and a catalog of models rather than a model vendor itself.
A $13 billion price on ~$150 million of revenue is a bet on Hugging Face's position as infrastructure, not on any single model it hosts.
Why Nvidia specifically is a strange fit
Delangue has spent much of 2026 publicly defending the case for open-weight models, including an appearance on CBS's Face the Nation where he warned about the risk of Chinese labs (DeepSeek, Alibaba's Qwen) dominating open-source AI if U.S. companies don't compete on openness. Nvidia has separately backed open-source initiatives of its own — it publishes the Nemotron family of open-weight models, which we broke down the real hosting and self-hosting cost of — so a Nvidia-Hugging Face combination isn't as incongruous as "GPU vendor buys neutral hub" first sounds.
It is, however, a real tension for the neutrality argument. Meta, Google, and DeepSeek all currently publish models on Hugging Face that compete directly with Nvidia's own AI ambitions and with the GPUs Nvidia sells to run them. Whether that neutrality survives new ownership — and whether rival labs keep publishing there at the same pace if it doesn't — is the open question analysts will be watching regardless of how the deal closes.
There's also a strategic-buyer logic that has nothing to do with model neutrality. Owning the hub that developers use to discover and deploy every major open-weight model gives Nvidia a direct read on where inference demand is heading before it shows up in GPU orders — and a natural on-ramp to sell more compute to exactly the teams building on those models. That's a different motive than "we like open source," and it's the one deal skeptics are likely to focus on if talks progress toward a signed agreement.
What it means for your bill, today: nothing yet
No pricing, hosting terms, or API access on Hugging Face has changed, and nothing in the reporting suggests an announced change is imminent even if the deal signs. If you're running open-weight models through Hugging Face-hosted endpoints or Inference Endpoints, your current rates hold. What's worth tracking is the medium-term picture: if you self-host open-weight models, the GPU cloud pricing comparison and GPU cloud cost calculator are the tools to check whether renting hardware still beats a hosted endpoint for your volume. If you're weighing open weights against a frontier API, the DeepSeek cost comparison and the token calculator cover the other side of that decision.
We'll update this post if Nvidia and Hugging Face confirm a deal, if the price or structure changes, or if the talks collapse the way the 2025 investment did.
Frequently asked questions
Is the Nvidia-Hugging Face deal confirmed?
No. As of this post's publication, neither company has confirmed the acquisition publicly, and both Business Insider and The Information report that no signed agreement exists — talks could still fall through, as Nvidia's earlier investment attempt did in 2025.
Why did Hugging Face reject Nvidia's earlier offer?
In late 2025, Hugging Face turned down roughly $500 million from Nvidia (a $7 billion valuation) because it didn't want a single investor with outsized influence over its decisions, given its role hosting models from Nvidia's own competitors.
Will my Hugging Face API costs change because of this?
Not based on anything reported so far. No pricing or hosting terms have changed, and neither company has announced plans to change them even if the acquisition closes.
Facts verified against TechCrunch's August 26–27, 2026 reporting citing The Information, and TechCrunch's August 24, 2026 reporting citing Business Insider.