Hours after NVIDIA printed its record quarter, the open layer absorbed a cap-table shock. The Information reports the chipmaker has agreed to buy Hugging Face for $12.9 billion; Bloomberg and Business Insider report serious talks in recent weeks over a deal valuing the platform at more than $13 billion, with no agreement reached. Neither NVIDIA nor Hugging Face has confirmed. If the number survives even as a negotiation, it is the largest consolidation of the open-weights ecosystem’s infrastructure into the hands of the company whose revenue runs on renting the silicon those weights execute against — and it lands the same evening today’s earnings re-answered the custom-silicon fight.
The open hub’s landlord changes — reported, not confirmed
What happened. The Information, citing a person with knowledge of the deal, reports Nvidia has agreed to buy Hugging Face for $12.9 billion. Business Insider, citing a person familiar with the matter, frames the same arc as serious acquisition conversations over a deal valuing the platform “more than $13 billion,” adding that the companies have not reached a deal and the talks could still fall apart. It dropped hours after NVIDIA guided Q3 to $108 billion and said it has $18 billion committed to equity investments for the rest of the fiscal year. Context: Hugging Face last raised in 2023 at a $4.5 billion valuation, and reportedly turned down a $500 million NVIDIA investment at a $7 billion valuation last year, citing the risk of a single dominant investor.
Why it matters. Hugging Face is where open weights live for most operators — the registry, Spaces, datasets, the model IDs a supply chain already keys on. A closed-vendor owner moves single-vendor coupling from the model layer to the distribution layer: the GPUs and the hub that hosts what runs on them end up on one balance sheet. The weights themselves survive in git history and mirrors, so this is not a hunting license on open models; it is a neutrality problem. Plan for a platform whose routing, Spaces, and export choices are made to optimize one stack, and keep weight provenance and object-storage redundancy independent of the hub no matter which report is right.
What I’m watching
Whether the deal is real and at which number: The Information says agreed at $12.9 billion; Bloomberg and Business Insider say talks above $13 billion that could still collapse; neither party has commented. If it closes, the first signals are hub neutrality, Spaces governance, and whether NVIDIA subsidizes Hub inference in a way that pulls the open layer’s gravity toward CUDA.
Source: theinformation.com, businessinsider.com, bloomberg.com, techcrunch.com