Nvidia, Stripe, and others pay billions to acquire open-weight AI companies

Nvidia is reportedly finalizing an agreement to acquire Hugging Face, a platform known for sharing open-weight AI models and benchmarks, for $13 billion. The deal would give Nvidia a central role in the open-weight AI ecosystem, which developers use to build and deploy models not controlled by frontier labs.

Separately, Stripe acquired OpenRouter, a provider of open-weight models to businesses, for more than $7 billion. The company framed the acquisition as part of its focus on optimizing token economics and compute efficiency, stating that tokens are the central currency for AI-driven businesses.

Nvidia also reached a $6 billion agreement with Poolside, an open-weight model builder. Under the deal, most of Poolside’s employees are expected to join Nvidia, signaling the chipmaker’s push into the model-building space.

The flurry of acquisitions follows growing enterprise interest in open-weight models, which are seen as offering greater control and configurability compared to proprietary alternatives. However, adoption remains limited: a survey cited by TechCrunch found that just 6% of companies use open-weight models, and only 2% of software engineers reported using them.

Open-weight models are currently favored for high-volume, repetitive tasks such as customer service chats, where tuning models for specific use cases can reduce costs. For more complex tasks like coding or agentic workflows, frontier models often remain preferred due to easier access and potential token subsidies.

Industry observers suggest that as AI workflows mature, more companies may turn to self-hosted, open models to manage costs and tailor models to their needs. Lin Qiao, CEO of Fireworks—a company processing 40 trillion tokens daily—argued that companies should consider building specialized models for their specific use cases.

The acquisitions also reflect a broader strategic move by Nvidia to diversify beyond its reliance on hyperscalers and frontier labs. This is highlighted by the development of proprietary inference chips by major AI labs, which Nvidia aims to counter by integrating into the model-building process.