Nvidia’s hugging face buy raises neutrality, antitrust fears

Hugging Face matters because it sits above any single model as the place where open AI work is organized. Varindia describes it as a “central marketplace and collaboration platform for open AI” where developers “discover, share, optimize and deploy models and datasets,” while FOMO研究院電子報 likens it to an “AI 模型的 App Store 加 GitHub,” a starting point where users search models, inspect ratings and downloads, test performance, find datasets and tools, then move into fine-tuning, testing and deployment rather than simply pulling files from storage.

Its scale reinforces that role as the distribution and collaboration layer for open AI. FOMO研究院電子報 says, “目前平台約有 1,300 萬名使用者、超過 200 萬個公開模型、50 萬個公開資料集。這代表它已經成為許多 AI 專案的起點,” while TBPN cites “over 18 million developers, 200,000 companies using the product, three million models and over half a million data sets,” describing a nexus where people upload, discover, test and modify models, with provenance records that show who uploaded them, what data was used and how versions changed.

Nvidia’s logic is not that Hugging Face owns the best models, but that it sits at the moment when model selection becomes hardware usage. Decrypt says Nvidia “reportedly agreed to buy Hugging Face for $12.9 billion,” a deal that would put “the world's largest open-model hub” under the ownership of its largest GPU maker, while Forbes says Nvidia is “reportedly paying $12.9 billion for Hugging Face to secure the layer where developers choose which models to run” — the point where the next wave of compute demand lands, because “the 2.96 million repositories come along as inventory.”

That makes the acquisition a distribution bet on open weights: Nvidia can steer workloads toward its chips without owning the models, because the value lies in shaping the default path developers follow after discovery. Forbes notes that although 85.6% of repositories have been downloaded fewer than 200 times, just 1.5% account for 99.2% of downloads, so influence at the top of the funnel matters disproportionately; as export controls pushed buyers toward “open weights they could hold themselves,” 20VC’s Harry Stebbings summed up the payoff bluntly: “NVIDIA’s $12.9B Hugging Face Deal… drives open-weights dominance,” because “Open source is good for compute salespeople.”

Hugging Face spent years cultivating value as a neutral layer rather than an arm of any one infrastructure company, and that posture was explicit in both its financing and its public language. TBPN says that in August of 2023, Hugging Face raised $235 million at a $4.5 billion valuation, then “Nvidia came in to offer 500 million late 2025 at a 7 billion valuation, but they turned it down,” while TechCrunch adds that the company “turned down a $500 million investment offer from Nvidia earlier this year” because “They said they didn't want just one investor controlling everything, right?”

That resistance mattered because Hugging Face had positioned itself as what TBPN called “the Switzerland of AI,” promising to be “chip agnostic” even while counting Nvidia, AMD, Intel and Qualcomm among its backers. TBPN described the platform as willing to host models that run well on AMD and models from Google or Amazon alike, turning it into “neutral territory” across competing chips and clouds; in that framing, the company’s usefulness depended less on exclusive alignment than on the trust that “no single company could control the platform.”

The fight is not over whether Hugging Face can stay open in name, but whether Nvidia can keep its open-source AI hub neutral in practice while nudging developers toward its own stack. After Nvidia confirmed on September 3, 2026 that it would acquire Hugging Face in a $12.9 billion deal—described as Nvidia’s second-biggest purchase—critics argued the asymmetry itself was the warning sign: as shattered.io quoted Cramer, “If AMD were announced they were buying Hugging Face, if Broadcom announced they were…” markets would react differently, because a dominant GPU vendor owning the hub creates an obvious conflict over defaults, recommendations, and support priorities.

That concern sharpens because the levers are subtle: Jensen Huang said on Nvidia’s Q2 FY2027 call that “the world will need both closed models and open models… and nearly all open models run on NVIDIA” and “Our position in open models is very good, because the CUDA ecosystem is literally everywhere,” language critics read as strategic intent rather than neutrality. Latent.Space noted fears that tooling direction could tilt “CUDA and away from AMD/portable GPU backends,” while AOL proposed three checkable triggers over the next two quarters, and antitrust scrutiny remains live, with closing not expected until the first half of 2027, subject to clearance.