Everyone is waiting for Nvidia to announce the most interesting tech deal of the week. It’s a reported $13 billion acquisition of Hugging Face, a platform that shares indifference-weighted AI models and benchmarks.
Currently best known as the target of a team of OpenAI agents hacking rewards, Hugging Face is at the center of an ecosystem of developers building and deploying LLMs not owned by Frontier Labs. Think of it like GitHub for the AI era.
Rumors of the deal come after Nvidia signed a $6 billion deal with random model manufacturer Poolside, which will see most of its employees move to the chip manufacturing giant. And two weeks ago, Stripe acquired OpenRouter, a leading provider of open weight models for enterprises, for more than $7 billion.
This reflects the latest trends in the AI field, with significant capital being poured into areas based on the free provision of goods.
For Nvidia, it is necessary to avoid further reliance on contracts with major hyperscalers and Frontier Labs. That’s especially true when major AI model builders like OpenAI and Google also build their own inference chips, like OpenAI’s Jalapeño, whose capabilities were announced this week. If model builders are making chips, Nvidia wants a piece of the model manufacturing business.
Nvidia has already built its own Nemotron family of openweight models, but their adoption hasn’t been huge. By controlling the largest open model development space in the United States, the company will have access to a large number of users that it can drive to its chips and standards.
There are also growing questions about the cost of AI inference, with companies considering cheaper models built by Chinese companies such as Moonshot, DeepSeek and Alibaba. Adoption is currently relatively small, but on the rise. Only 6% of companies use an open-weight model, according to a study of spending data by Ramp, and just 2% of software engineers surveyed by developer tools developer Jellyfish.
Nick Albarran, AI product lead at Jellyfish, told TechCrunch that the open-weight model is primarily used by companies whose products rely on repetitive inference workloads, such as companies that provide customer service chat. These are repetitive and high-volume tasks, so open-weight models can be tuned to answer questions cheaply.
That’s certainly what Stripe planned to do with its OpenRouter acquisition. “Tokens are the central currency for companies building on AI, and it’s clear that real-world economic potential depends on making effective use of scarce computing resources,” Stripe co-founder and CEO Patrick Collison said in a statement.
However, for coding and agent tasks, different requests and more inferences mean that frontier models often win. One reason for this is that proprietary labs facilitate access and, in some cases, provide token subsidies. As companies dial in to AI workflows, the transition to an open model will become easier, Albarrán said. Still, the primary reason businesses are turning to these models now is for ease of control and configuration, rather than expense concerns.
“There aren’t many companies like that yet…[but]if prices from Frontier Labs continue to rise, more and more companies will be forced to at least consider it,” Albarrán told TechCrunch. “As AI-driven workflows become more mature, it makes sense to invest in a self-hosted model.”
Lin Qiao is the CEO of Fireworks, a company that provides open-weight model routers and hosts for enterprise users and is often discussed as a potential acquisition for major technology companies. Qiao says her company processes 40 trillion tokens a day, which is more than Gemini and OpenAI’s APIs.
Fireworks is betting on model diversity. As LLMs become more popular and improved, it becomes easier for companies to train LLMs to suit their needs. “Every app company should consider hiring in-house researchers,” she told TechCrunch last week. “Companies can use their products and product data to build their own models. In fact, the future is specialized intelligence. Literally every company should have a unique model for each use case, and it will happen automatically.”
It’s easy to forget how early we are in the development of AI as a tool and a business. However, OpenAI and Anthropic’s dominance is not inevitable. The allure of open technology is proving difficult to resist as tech giants seek to avoid betting on the biggest labs.
If you make a purchase through links in our articles, we may earn a small commission. This does not affect editorial independence.
