This week could bring confirmation of one of the sector’s most notable deals: reports say Nvidia is set to acquire Hugging Face for about $13 billion. Hugging Face is a central hub for sharing open-weight AI models and benchmarks and has become a focal point for developers building and deploying large language models (LLMs) outside the biggest proprietary labs — a kind of GitHub for the AI era.
The Nvidia–Hugging Face rumors follow Nvidia’s $6 billion agreement with Poolside, an open-weight model builder whose employees will largely move to the chipmaker. Two weeks earlier, Stripe acquired OpenRouter — a leading provider of open-weight models for businesses — for more than $7 billion.
This represents a substantial flow of capital into a segment that, in principle, is built on sharing models freely, and it highlights some of the current strategic moves in the AI industry.
Why big tech is buying-in
For Nvidia, a key motivation is reducing dependence on deals with hyperscalers and frontier labs. That pressure is heightened as major model builders like OpenAI and Google develop their own inference chips — for example, OpenAI’s recently announced Jalapeño chip. If model developers build chips, Nvidia wants to capture a portion of the model-building and deployment ecosystem.
Nvidia already publishes its own open models under the Nemotron family, but uptake has been limited so far. Acquiring or partnering with the largest U.S. developer and user community for open models would give Nvidia access to many users it could steer toward its hardware and standards.
Cost pressures and alternative models
Rising concerns about the cost of inference have companies looking at cheaper models from Chinese providers such as Moonshot, DeepSeek and Alibaba. Adoption is still relatively small but growing: Ramp’s spending-data survey finds that 6% of companies use open-weight models, while Jellyfish measures usage among software engineers at about 2%.
Nik Albarran, AI product lead at Jellyfish, told TechCrunch that open-weight models are mostly used by companies with repeated inference workloads — for example, firms running customer-service chat systems. Because those tasks involve high volume and repetition, an open-weight model can be fine-tuned to serve answers more cheaply.
That cost-and-efficiency argument is central to how Stripe positioned its OpenRouter purchase. “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,” Patrick Collison, Stripe’s co-founder and CEO, said in a statement.
However, for coding and agentic tasks that require varied prompts and deeper reasoning, frontier models often perform better — partly because proprietary labs make access easier and sometimes subsidize tokens. Albarran suggests that as corporate AI workflows become more mature, switching to open models will be simpler; for now, companies mostly choose open models for control and configurability rather than outright cost concerns.
Providers betting on diversity and specialization
Lin Qiao, CEO of Fireworks — a leading open-weight model router and host for corporate customers often mentioned as an acquisition target — says her company processes 40 trillion tokens per day, more than either Gemini’s or OpenAI’s APIs. Fireworks is betting on model diversity: as LLMs proliferate and improve, companies will increasingly be able to train models specifically for their needs.
Qiao argues that every product company should consider hiring an in-house researcher to build models using product and product-data. She predicts a future of specialized intelligence where each company has bespoke models for distinct use cases, and that this will happen organically as needs mature.
Early market, evolving balance of power
It is easy to forget how early the AI market still is as both a toolset and a business. The current dominance of providers like OpenAI and Anthropic is not guaranteed in perpetuity. As big tech companies hedge their exposure to the largest labs, open technologies and the ecosystems around them are attracting substantial investment and could reshape where control and value sit in the AI stack.
