The operational shift here is custody, not code. Before this deal, Hugging Face functioned as the closest thing the open-model economy had to a neutral registrar: a hub where more than 13 million developers, per figures cited by 36Kr, uploaded, benchmarked, and downloaded models regardless of who built them or what chips they ran on. After the deal — $12.9 billion in cash according to The Information and CNBC, with 36Kr reporting a roughly $14 billion total package including a $1 billion retention pool for core R&D staff — that registrar answers to the world’s dominant GPU vendor.
Nvidia’s $12.9 billion purchase of Hugging Face is less a bet on open-source ideology than a move to own the scoreboard that ranks whose open models win.
That scoreboard problem isn’t hypothetical. Hugging Face’s own ‘state of open models’ report, published August 14, 2026 and cited by Fortune and PYMNTS, showed Alibaba’s Qwen family pulling more than 3 billion downloads in six months — dwarfing Google’s 418 million and Meta’s 227 million. Nvidia has spent the past month rallying 80-plus companies around an open-letter defense of open-weight models and launching its SAFE incident-sharing initiative, positioning itself as open source’s institutional champion against both closed US labs and, implicitly, unaligned Chinese ones. Owning the platform that produces the industry’s authoritative download and leaderboard data hands Nvidia a lever over that narrative it didn’t have when it was merely an investor.
What actually moves
Concretely: the 2,000-plus enterprises paying for Hugging Face’s Hub, per 36Kr’s reporting, now sit inside Nvidia’s commercial orbit rather than a vendor-neutral one — a soft lock-in vector toward CUDA and Nvidia inference stacks that no amount of open-standards messaging fully offsets. Notably, Hugging Face rejected a $500 million Nvidia investment offer earlier this year, per 36Kr, before full acquisition talks began after a rival suitor showed interest, according to CNBC’s sourcing of The Information’s reporting — a sequence suggesting price, not principle, closed the gap from a $4.5 billion 2023 valuation to nearly $14 billion now.
Watch three things: whether US antitrust regulators, flagged by 36Kr as a likely reviewer given the software-hardware vertical integration, force behavioral commitments on model-hosting neutrality; whether Hugging Face’s leaderboard methodology or enterprise defaults shift measurably toward Nvidia-optimized formats; and whether rival chipmakers or hyperscalers building custom silicon route around Hugging Face entirely rather than fund a competitor’s storefront.
“Open models let startups, businesses, universities and public institutions build on advanced capabilities without training every model from scratch,” Nvidia CEO Jensen Huang said in the press release announcing the deal. “AI advances faster when people can build together.”
— WIRED
