Nvidia has reportedly agreed to buy Hugging Face, the open-source AI model repository used by more than 13 million developers, for $12.9 billion. The news broke late on August 27, 2026, when The Information cited a person with direct knowledge of the talks, and was quickly matched by CNBC, TechCrunch, Fortune, Tom’s Hardware and SiliconANGLE. If it closes, it would be Nvidia’s largest acquisition ever, nearly double what the chipmaker paid for Mellanox in 2020.
One caveat matters here: this is not a signed, announced deal. Multiple outlets, including Business Insider’s reporting summarized by TechCrunch, describe the talks as ongoing and note a signed agreement had not been reached as of publication. Both companies declined to comment. So treat the $12.9 billion figure as a strong, multi-sourced report rather than a closing bell. That distinction shapes everything from the stock reaction to the antitrust conversation that’s already started.
What’s actually being reported
According to SiliconANGLE, Nvidia is buying the platform that hosts open-source AI models and datasets used across the industry, from research labs to solo developers fine-tuning small models on a laptop. Hugging Face’s flagship product, a GitHub-style hub for uploading and sharing model weights, launched in 2020 and grew into one of the default places developers go to find a pretrained model instead of training one from scratch.
CNBC and Fortune both frame the reported price as a sharp reversal from where talks stood less than a year ago. Nvidia had proposed a $500 million investment in Hugging Face that would have valued the company at around $7 billion. Hugging Face turned it down, reportedly because the founders didn’t want a single chipmaker holding outsized influence over a platform meant to stay neutral. Now Nvidia is said to be paying nearly double that valuation to own the whole thing outright.
Hugging Face was founded in 2016 by Clément Delangue, Julien Chaumond and Thomas Wolf, initially as a teen chatbot app before pivoting to open-source machine learning tools. Nearly a decade later it sits at the center of the open-weight ecosystem: a place where a model released by Meta, Mistral, or a university lab gets downloaded, benchmarked and deployed. TechCrunch reports Hugging Face had been fielding interest from more than one potential acquirer before the Nvidia talks accelerated.
Why Nvidia wants the model layer, not just the chip layer
Nvidia’s core business is still GPUs, and it just posted another blowout quarter, with shares jumping roughly 7% on the earnings print according to CNBC’s coverage of the same week. But GPU demand doesn’t exist in a vacuum. It’s downstream of decisions developers make about which models to run and where to run them. Hugging Face sits at exactly that decision point.
Own the hub where millions of developers pick a model, and you can shape which hardware and software stack that model is optimized for by default. That’s the logic analysts have floated in the hours since the report broke: this isn’t really a bet on Hugging Face’s own revenue, it’s a bet on steering the top of the funnel that eventually turns into GPU orders. It also marks a return to a business Nvidia had reportedly pulled back from roughly a year earlier: renting out cloud infrastructure directly rather than just selling chips to cloud providers.
There’s a defensive angle too. Google, Amazon and Microsoft are all shipping their own AI silicon at increasing volumes, chipping away at the assumption that every serious AI workload needs an Nvidia GPU underneath it. Locking in a closer relationship with the platform that distributes open models gives Nvidia leverage that’s harder to route around than a spec sheet comparison.
The cloud angle is worth spelling out too. Nvidia had scaled back its own direct cloud rental ambitions roughly a year before this report surfaced, choosing instead to sell chips to hyperscalers rather than compete with them for cloud customers directly. A Hugging Face acquisition partially reverses that retreat. Hugging Face already runs paid Inference Endpoints and Spaces hosting, meaning Nvidia would inherit an existing, if smaller, cloud-services business layered directly on top of a developer community it doesn’t have to build from scratch.
The reported numbers at a glance
| Detail | Reported figure | Source |
|---|---|---|
| Reported acquisition price | $12.9 billion | The Information, via CNBC/TechCrunch/Fortune |
| Nvidia’s earlier rejected offer | $500M investment at ~$7B valuation | TechCrunch |
| Hugging Face founded | 2016 | Multiple outlets |
| Flagship model-hosting product launched | 2020 | SiliconANGLE |
| Developers on the platform | 13M+ | SiliconANGLE |
| Nvidia’s largest prior deal (Mellanox, 2020) | ~$7 billion | Tom’s Hardware |
| Deal status as of Aug 27, 2026 | Reported, not signed | Business Insider, via TechCrunch |
| Nvidia stock move around earnings week | +7% | CNBC |
Historical context: this isn’t Nvidia’s first vertical bet
Nvidia’s M&A history is short but pointed. The $7 billion Mellanox deal in 2020 gave it control of the networking gear that connects GPUs inside data centers, a purchase that looked expensive at the time and now reads as prescient given how much of Nvidia’s data-center revenue depends on high-speed interconnects. The pattern has been consistent: buy the layer just above or below the chip, rather than diversify into unrelated businesses.
Hugging Face fits that pattern but sits further from silicon than anything Nvidia has bought before. It’s not a networking company or a systems vendor, it’s a developer platform and community, closer in spirit to GitHub than to a chip supplier. That’s part of why the reported price raised eyebrows: Nvidia isn’t just buying infrastructure, it’s buying a piece of open-source culture that has resisted single-vendor control specifically because Hugging Face’s founders said no to a much cheaper offer a year earlier for that exact reason.
It also lands in the same window as a broader wave of AI infrastructure consolidation. Databricks bought MosaicML to bring model training in-house. Google closed its own multibillion-dollar security shift with the $32 billion Wiz acquisition earlier in 2026, reshaping how hyperscalers think about owning adjacent layers of the stack rather than partnering for them. Nvidia’s own recent moves, including the AWS deal adding 2 million GPUs and steady expansion into networking and software, point the same direction: control more of the stack, not just the chip at the bottom of it.
Competitive landscape: who this squeezes
The companies with the most to think about aren’t necessarily Hugging Face’s direct rivals, since it doesn’t really have a like-for-like competitor at its scale. It’s the companies that depend on Hugging Face staying neutral.
| Player | Why the deal matters to them |
|---|---|
| AMD | Competes with Nvidia on AI accelerators; a Hugging Face under Nvidia’s roof could tilt default tooling and optimization toward CUDA over ROCm |
| Google / Microsoft | Both ship in-house AI chips (TPUs, Maia) and would lose neutral-ground access to the platform that distributes the open models their customers run |
| CoreWeave and other GPU clouds | Rely on open-model distribution to drive rentable compute demand; ownership concentration adds a new dependency risk |
| Databricks | Already vertically integrated via MosaicML; a Hugging Face tie-up with Nvidia raises the bar for owning the training-to-deployment pipeline |
| Independent AI startups | Many build directly on Hugging Face’s hub, spaces and inference tooling; vendor lock-in risk rises if Nvidia optimizes the platform for its own stack |
None of this means Hugging Face stops being open-source under Nvidia. The reported plan doesn’t include closing the platform or restricting access to competing hardware. But ownership changes incentives even when the license doesn’t change, and that’s the concern regulators are likely to focus on if the deal moves toward a formal filing.
The antitrust question nobody’s answered yet
Nvidia already controls the overwhelming majority of the AI accelerator market. Layering ownership of the largest open-model distribution hub on top of that is the kind of vertical combination that draws regulatory attention, even when the acquired company itself isn’t a chipmaker. The concern isn’t that Hugging Face’s catalog disappears, it’s that a chip vendor gets to see, and potentially shape, which models the rest of the industry standardizes on.
The EU and US have both shown more appetite for scrutinizing AI infrastructure deals over the past two years than they did earlier in the boom. A $12.9 billion price tag isn’t large enough on its own to guarantee a deep antitrust review, but the strategic position Hugging Face occupies, sitting between every open model and the hardware it runs on, makes this a different kind of scrutiny than a straightforward horizontal merger. Expect questions about interoperability commitments and whether Nvidia would agree to keep the platform hardware-agnostic as a condition of clearance, assuming the deal reaches that stage at all.
What’s still unconfirmed
A lot of the deal’s shape hasn’t leaked. Reporting so far doesn’t specify whether the $12.9 billion would be paid in cash, stock, or some mix of both. There’s no confirmed timeline for signing or closing, and neither company has issued an official statement beyond declining to comment when reporters asked. Business Insider’s account, relayed by TechCrunch, is the most cautious of the bunch, describing the two sides as still in talks at a valuation above $13 billion rather than at a finalized number.
That gap between “agreed” and “in talks” matters for anyone tracking this closely. The Information’s sourcing points to a firmer stage of negotiation; other outlets are more conservative. Both can be true at once if terms were verbally agreed but not yet papered, which is a common stage for deals of this size before a public announcement.
Market impact: why this is bigger than the price tag
$12.9 billion is real money, but it’s a rounding error against Nvidia’s cash generation from AI chip sales this year. The market impact isn’t really about Nvidia’s balance sheet. It’s about what the deal signals for how the next phase of AI infrastructure gets built: fewer independent layers, more vertically owned stacks.
If it closes, expect three immediate effects. First, competing chipmakers will accelerate their own developer-platform investments rather than relying on Hugging Face as neutral ground, which could fragment where open models get published. Second, cloud providers building in-house silicon will have another reason to route around Hugging Face for anything strategically sensitive. Third, expect a wave of “is this still neutral” scrutiny from the open-source AI community itself, the same community whose skepticism led Hugging Face to reject Nvidia’s cheaper offer a year ago.
There’s also a talent dimension that tends to get overlooked in deal-price headlines. Hugging Face’s engineering team built and maintains widely used open-source libraries, including Transformers and Diffusers, that plug into nearly every major machine learning framework. Retaining that team matters as much to Nvidia as the platform’s user base, since tooling maintained by disengaged or departed engineers tends to stagnate fast in a field moving as quickly as AI infrastructure. Any acquisition agreement is likely to include retention packages for the founders and core maintainers, similar to how Databricks structured its MosaicML purchase.
How this compares to prior AI infrastructure deals
| Deal | Buyer | Reported/confirmed value | Status |
|---|---|---|---|
| Nvidia – Hugging Face | Nvidia | $12.9B | Reported, unsigned as of Aug 27, 2026 |
| Nvidia – Mellanox | Nvidia | ~$7B | Closed, 2020 |
| Google – Wiz | $32B | Closed, 2026 | |
| Databricks – MosaicML | Databricks | ~$1.3B | Closed, 2023 |
| AWS – GPU/DuckDB expansion | Amazon | 2M GPUs added | Announced, 2026 |
Set next to those, the Hugging Face number is smaller in dollars than the Wiz deal but arguably higher-stakes strategically, since it touches the distribution layer for every open model rather than a single company’s security posture. It also dwarfs Databricks’ MosaicML purchase, which was aimed at owning model training rather than model distribution at Hugging Face’s scale.
What developers should actually watch for
If you build on Hugging Face today, the near-term reality is nothing changes until there’s an official announcement, and even then, integration usually takes months. The things worth tracking:
- Whether Nvidia commits publicly to keeping the platform hardware-neutral for inference and hosting
- Any changes to Hugging Face’s Inference Endpoints pricing or default hardware options
- Whether competing chipmakers respond by backing an alternative model hub
- Regulatory filings in the US and EU, which would confirm the deal is moving past “reported” status
- Founder statements from Delangue, Chaumond or Wolf, none of which have surfaced publicly as of this writing
Predictions: where this goes from here
- A formal announcement, if it comes, is more likely to land within weeks than months, given how many outlets already have matching sourcing on the price.
- Regulators in at least one major jurisdiction will request more information before clearing the deal, given Nvidia’s existing chip market share.
- AMD and at least one hyperscaler will publicly reaffirm investment in alternative open-model distribution channels within the next two quarters.
- Hugging Face’s core product will stay free and open-source at the license level, even if infrastructure defaults quietly shift toward Nvidia hardware over time.
- Expect at least one more billion-dollar-plus AI infrastructure acquisition announced before the end of 2026, continuing the vertical-integration trend seen with Wiz and this deal.
Frequently asked questions
Has Nvidia officially confirmed buying Hugging Face?
No. As of August 28, 2026, neither company has issued an official statement. The $12.9 billion figure comes from The Information’s sourcing, matched by CNBC, TechCrunch, Fortune, Tom’s Hardware and SiliconANGLE, but Business Insider’s reporting says no signed agreement had been reached and talks could still fall apart.
How much did Nvidia previously offer Hugging Face?
Nvidia reportedly proposed a $500 million investment last year that would have valued Hugging Face at around $7 billion. Hugging Face turned it down.
Who founded Hugging Face and when?
Clément Delangue, Julien Chaumond and Thomas Wolf founded Hugging Face in 2016. It started as a teen chatbot app before becoming an open-source machine learning hub.
Would this be Nvidia’s biggest acquisition ever?
Yes, if it closes at the reported $12.9 billion price. Nvidia’s previous largest deal was Mellanox in 2020 at roughly $7 billion.
Would Hugging Face stop being open-source?
Nothing reported so far suggests Hugging Face’s open-source licensing model would change. The concern raised by analysts is about hardware neutrality and platform incentives, not the licenses attached to hosted models.
Why would Nvidia want a software platform instead of more chip capacity?
Owning the platform where developers choose and download models gives Nvidia influence over the layer just above hardware, where decisions about optimization and default deployment targets get made, ultimately steering demand back toward its own GPUs.
Could regulators block the deal?
Analysts cited in coverage of the deal say it’s likely to draw antitrust scrutiny given Nvidia’s dominant position in AI chips, though a $12.9 billion price alone doesn’t guarantee an extended review. A formal filing would be the first real signal of how deep that scrutiny goes.
When could the deal close?
No timeline has been disclosed. Given the deal is still described as unsigned by some sources, any closing date remains speculative until an official announcement.
Related Coverage
- HBM4 Memory Hits 80% Yield, Powers Nvidia Rubin
- HPE Stock Hits 52-Week High on Nvidia CPU
- Jalapeño Chip: OpenAI Targets Nvidia’s 75% Margin
- AWS-NVIDIA Deal Adds 2M GPUs as AWS Buys DuckDB
- Google Closes $32B Wiz Deal: Cloud Security Reset
For more coverage of chips, GPUs and AI infrastructure economics, see the Hardware & Chips section.
