Nvidia has confirmed an open-weight AI acquisition that reframes what the chip giant actually is: a definitive agreement to buy Hugging Face for a purchase price of approximately $11.9 billion, plus up to $1.0 billion in equity-based retention payments for Hugging Face employees who join Nvidia. The transaction is expected to close in the first half of 2027, subject to regulatory approvals.
The headline figure of roughly $13 billion makes this Nvidia’s second-largest deal on record, according to CNBC, trailing only its $20 billion purchase of chip assets from Groq last December. Hugging Face itself is no small trophy: more than 18 million developers, researchers, and creators use the platform to share over 3 million models, and more than 200,000 companies use it to discover and deploy AI.
Think of it as GitHub for the AI era, right down to the fact that Nvidia is promising to keep it open. As part of the deal, Nvidia has committed to continuing to let model makers, developers, and users upload and download models and datasets freely, and to supporting other silicon vendors. Whether that commitment survives a decade of corporate ownership is a question worth bookmarking.
The Open-Weight AI Acquisition Logic
The Hugging Face deal doesn’t exist in isolation. It lands alongside Nvidia’s separate $6 billion arrangement with Poolside, which is structured not as a straight acquisition but as a non-exclusive licence for Poolside’s Model Factory software, plus a $1 billion equity investment at a $12 billion pre-money valuation, according to Reuters. Poolside’s three co-founders are staying put and the company continues to operate independently.
The snippet characterised the Poolside deal as one where most employees would move to Nvidia. The reality is more targeted: 109 employees specifically involved in developing the Laguna model family were offered positions, per a letter Poolside sent investors. Poolside called it explicitly ‘not an acquisition and not an acquihire.’
The Wall Street Journal reports that Nvidia views the Poolside arrangement as a way to build one of the world’s most powerful open-weight models to compete with Chinese rivals including DeepSeek and Kimi K3, and that Nvidia believes the US has fallen behind in open AI partly because American labs have focused on proprietary models. Nvidia, as it happens, is already the largest contributor of open models and data to the Hugging Face platform. So the Hugging Face acquisition gives it the shop window; the Poolside deal gives it the product.
Then there’s Stripe. Its acquisition of OpenRouter, confirmed on the Stripe newsroom as the company’s largest-ever deal, came in at $7.5 billion according to the New York Times, with $1.5 billion going to founders and $6 billion to investors. That compares with an OpenRouter valuation of roughly $1.3 billion as recently as May 2026, a step-up of approximately 5.8x in about three months. Stripe reportedly outbid Databricks to get there.
Why Tokens Are the New Currency
OpenRouter is worth understanding before wondering whether Stripe overpaid. The platform provides a single gateway to more than 400 AI models from over 80 providers, letting developers route each request to whichever model fits best on price, speed, and reliability. It processes more than 10 trillion tokens a day and serves over 10 million developers and businesses.
Patrick Collison, Stripe’s cofounder and CEO, framed the logic plainly: ‘Tokens are the central currency for companies building with AI, and it’s clear that the real-world economic potential will depend on making good use of scarce compute resources.’ Stripe is, at heart, a payments infrastructure company. Owning the routing layer for AI tokens is not such a strange extension of that identity.
Adoption of open-weight models is still modest: just 6% of companies use them, according to spending data surveyed by Ramp, and only 2% of software engineers in a Jellyfish survey. Nik Albarran, Jellyfish’s AI product lead, told TechCrunch that open-weight models currently suit high-volume, repetitive inference workloads, like customer service, where a tuned model can handle questions cheaply. Frontier models still win on coding and agentic tasks, partly because proprietary labs offer easier access and occasional token subsidies.
Lin Qiao, CEO of Fireworks, an open-weight model router and host often discussed as a future acquisition candidate, puts it more boldly. Her company processes 40 trillion tokens a day. ‘Every single app company should consider hiring an in-house researcher,’ she told TechCrunch. ‘The future is actually specialised intelligence. Literally, every single company should have their own model per use case.’
Albarran’s view is that price pressure from frontier labs will eventually force more companies to make that move: ‘If the prices continue to go up from the frontier labs, more and more companies will be forced to at least consider it.’ That moment hasn’t arrived yet. But judging by the capital now flowing into this open-weight AI acquisition wave, a few very large companies are betting it’s coming sooner than the adoption numbers suggest.
