Nvidia's reported $13 billion Hugging Face deal heads a wave of open-weight acquisitions

Nvidia is said to be buying Hugging Face for $13 billion

The tech industry is waiting on Nvidia to confirm what TechCrunch describes as the most notable tech deal of the week: a reported $13 billion acquisition of Hugging Face, the platform where developers publish open-weight AI models and benchmark results. Neither company has announced the transaction, so the price and terms rest on the report rather than an official statement.

Hugging Face plays roughly the role in machine learning that GitHub plays in software at large. It is the central gathering point for developers building and deploying large language models that are not controlled by the frontier labs such as OpenAI, Google and Anthropic.

A cluster of open-weight deals

The reported purchase would be the third large transaction in the space within a short window. Nvidia has already struck a $6 billion agreement with Poolside, an open-weight model developer, under which most of Poolside's employees will move to the chipmaker. Roughly two weeks before TechCrunch's report, Stripe acquired OpenRouter, the leading supplier of open-weight models to businesses, for more than $7 billion.

That is a lot of capital flowing into companies whose core product is freely available technology, and TechCrunch reads it as a sign of where the AI sector is heading.

Nvidia's motivation

Part of the logic is defensive. Nvidia's business depends heavily on its relationships with hyperscalers and frontier labs, several of which are designing their own inference chips — OpenAI's chip, called Jalapeño, had its capabilities detailed in the same week as the report. If model builders are becoming chipmakers, the reasoning goes, Nvidia wants a position in model-making itself.

Nvidia already maintains its own Nemotron family of open-weight models, but according to TechCrunch their uptake has been limited. Controlling the largest US developer space for open models would give the company a large audience it can direct toward its chips and its technical standards.

Open-weight adoption is still niche

For all the deal money, actual usage remains small. Citing spending data from Ramp, TechCrunch reports that 6% of companies use open-weight models, while Jellyfish, which builds tooling for developers, measures adoption at 2% of software engineers.

Nik Albarran, AI product lead at Jellyfish, told TechCrunch that open-weight models mostly serve companies whose products depend on repeated, high-volume inference, such as customer-service chat, where a tuned model can answer common questions cheaply. For coding and agentic work, varying requests and heavier reasoning still tend to favor frontier models, partly because proprietary labs offer easier access and sometimes subsidize tokens.

Albarran said the current appeal of open-weight models is control and configurability rather than cost, though that could shift. If frontier-lab prices keep climbing, more companies will at least weigh the alternative, and mature AI workflows are the point where self-hosting starts to make sense. Rising inference costs are also pushing some companies to look at cheaper models from Chinese developers such as Moonshot, DeepSeek and Alibaba, according to TechCrunch.

Stripe's CEO Patrick Collison framed the OpenRouter acquisition in similar efficiency terms, saying in a statement that the economic potential of AI depends on making good use of scarce compute resources.

Fireworks and the specialist-model thesis

Lin Qiao, CEO of Fireworks, an open-weight router and hosting provider often mentioned as a takeover candidate, told TechCrunch her company processes 40 trillion tokens a day, a figure she says exceeds the APIs of both Gemini and OpenAI.

Her argument is that as models proliferate and improve, companies will increasingly train models tuned to their own needs. Every app company should consider hiring an in-house researcher, she said, because product data can feed a proprietary model and the future is specialized intelligence, with a model per use case at each company.

Why it matters

If the reported figure holds, the Hugging Face purchase would rank among the largest AI deals ever, and it would confirm that open-weight platforms — communities, routers and model builders — have become prime acquisition targets for tech giants.

The buyers are not purchasing current revenue: adoption sits in the single digits. They are buying positioning against a future in which frontier labs control both models and increasingly their own silicon. For Nvidia, owning the main hub of the open-model ecosystem would tie a large developer population to its hardware and standards, while deals like Stripe's OpenRouter purchase point to inference economics as the other driving force. The broader signal, as TechCrunch notes, is that the dominance of OpenAI and Anthropic is not being treated as inevitable, and open technology is proving hard for the giants to resist.