Tech giants race to acquire open-weight AI startups as Nvidia, Stripe make big bets

Nvidia is reportedly in talks to acquire Hugging Face for $13 billion, following its $6 billion deal with Poolside and Stripe's $7 billion purchase of OpenRouter. The acquisitions reflect a push by chipmakers and payments firms to control the open-weight AI ecosystem, which offers cheaper inference for high-volume tasks. Open-weight models are currently used by only a small fraction of companies, but their adoption is growing as firms seek cost savings.

The reported $13 billion Hugging Face deal would give Nvidia control over the primary hub where developers share open-weight models, benchmarks, and tools. This follows Nvidia’s $6 billion Poolside agreement and Stripe’s $7 billion OpenRouter purchase, signaling a strategic pivot toward the open-weight ecosystem. These models currently serve a niche—only 6% of companies use them, per Ramp data—but they excel in high-volume, repetitive tasks like customer service chats, where tuning can cut inference costs. Fireworks, a rival router, processes 40 trillion tokens daily, more than major frontier APIs, highlighting the scale possible.

Open-weight adoption is driven less by cost concerns today and more by control and configurability, according to Jellyfish’s AI lead. As frontier labs raise prices and workflows mature, more firms may shift to self-hosted models. Nvidia’s interest also stems from frontier labs building their own chips, like OpenAI’s Jalapeño, pushing the chipmaker to secure a user base for its hardware and standards.