Nvidia and Stripe lead $20 billion wave of acquisitions targeting open-weight AI platforms
Nvidia, Stripe and other tech firms are spending billions to acquire companies that host open-weight large language models, aiming to reduce reliance on proprietary AI labs and capture a growing developer market.
Nvidia is expected to confirm a reported $13 billion purchase of Hugging Face, a platform that hosts open-weight AI models and benchmarks, positioning it as a central hub for developers building large language models outside the major labs. The move follows Nvidia's recent $6 billion deal with Poolside, an open-weight model builder, and Stripe's acquisition of OpenRouter for more than $7 billion, underscoring a surge of capital into a sector built on freely shared models.
Analysts say Nvidia seeks to lessen its dependence on hyperscaler and frontier-lab partnerships, especially as rivals such as OpenAI and Google develop their own inference chips. By controlling the largest U.S. developer community for open models, Nvidia hopes to steer users toward its hardware and standards.
Open-weight models currently serve a modest share of the market—about 6 % of companies and 2 % of software engineers, according to surveys by Ramp and Jellyfish. Nik Albarran, AI product lead at Jellyfish, noted that these models are attractive for high-volume, repetitive tasks like customer-service chat, where they can be fine-tuned to reduce compute costs.
"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," said Patrick Collison, Stripe co-founder and CEO, referring to the strategic rationale behind the OpenRouter purchase.
While proprietary models still dominate complex coding and reasoning tasks, Albarran expects broader adoption of open models as firms mature their AI workflows and seek greater control and configurability. He added that rising prices from frontier labs could push more companies to consider self-hosting.
Fireworks, an open-weight model routing and hosting service for enterprises, processes roughly 40 trillion tokens daily, according to CEO Lin Qiao. She argues that as model diversity expands, companies will increasingly train specialized models in-house, stating that “every single app company should consider hiring an in-house researcher” to build custom intelligence.
The wave of acquisitions reflects a broader industry trend: major technology players are hedging against the dominance of OpenAI and Anthropic by investing in open AI infrastructure, which they view as a long-term growth opportunity.