The tech world is buzzing, awaiting confirmation of the week’s most significant tech transaction: A reported $13 billion acquisition by Nvidia of Hugging Face, a prominent platform for sharing open-weight AI models and benchmarks.
Hugging Face, recently in the spotlight as the target of reward-hacking OpenAI agents, stands at the heart of the developer community building and deploying large language models (LLMs) not owned by major frontier labs. It essentially functions as a GitHub for the AI era.
These acquisition rumors follow Nvidia’s earlier $6 billion deal with Poolside, an open-weight model developer, which will see most of its staff joining the chip manufacturing giant. Furthermore, just two weeks prior, Stripe acquired OpenRouter, a leading provider of open-weight models to businesses, for over $7 billion.
This significant investment into a sector that primarily offers its resources freely highlights the evolving trends within the AI industry.
For Nvidia, there’s a strategic imperative to reduce its reliance on agreements with major hyperscalers and frontier labs. This is particularly crucial as leading AI model developers like OpenAI and Google are now developing their own inference chips, such as OpenAI’s Jalapeño, whose capabilities were revealed this week. If model builders are creating chips, Nvidia aims to secure a share of the model-making market.
Nvidia already develops its own Nemotron family of open-weight models, but their adoption has been limited. By gaining control of the largest US developer hub for open models, the company will access a vast user base it can direct towards its chips and standards.
Concerns about the expense of AI inference are also increasing, leading companies to explore more cost-effective models developed by Chinese firms like Moonshot, DeepSeek, and Alibaba. Currently, adoption remains relatively low but is growing – only 6% of companies utilize open-weight models, according to a spending data survey by Ramp, and merely 2% of software engineers polled by Jellyfish, a developer tool provider.
Nik Albarran, Jellyfish’s AI product lead, informed TechCrunch that open-weight models are predominantly used by companies whose products depend on repetitive inference workloads, such as those providing customer service chatbots. For these high-volume, repetitive tasks, an open-weight model can be finely tuned to answer questions economically.
Stripe has certainly articulated its OpenRouter acquisition in this light. “Tokens are the fundamental currency for companies building with AI, and it’s clear that their true economic potential will hinge on efficiently utilizing scarce compute resources,” stated Patrick Collison, Stripe’s co-founder and CEO.
However, for coding and agentic tasks, where requests vary more and require greater reasoning, frontier models often prevail, partly because proprietary labs offer easier access and, in some cases, token subsidies. Albarran suggests that as companies refine their AI workflows, transitioning to open models will become simpler. Nevertheless, the primary drivers for companies considering these models now are control and configurability, rather than cost concerns.
“Few companies are at that stage yet…[but] if prices from frontier labs continue to rise, more companies will be compelled to at least consider it,” Albarran told TechCrunch. “When your AI-driven workflows are considerably more mature, that’s when investing in self-hosting models makes sense.”
Lin Qiao, CEO of Fireworks – a prominent open-weight models router and host for corporate users often cited as a potential target for a tech giant – claims her company processes 40 trillion tokens daily, surpassing the APIs of both Gemini and OpenAI.
Fireworks’ strategy banks on model diversity: as LLMs become more abundant and sophisticated, companies will find it easier to tailor them precisely to their needs. “Every single app company should consider hiring an in-house researcher,” she shared with TechCrunch last week. “They can leverage their product and product data to build their own model. The future is truly specialized intelligence. Literally, every company should have its own model per use case, and that will happen organically.”
It’s easy to overlook how nascent AI’s development is as both a tool and a business. However, the current dominance of OpenAI and Anthropic is not a foregone conclusion. As tech giants diversify their investments beyond the largest labs, the appeal of open technology is proving increasingly compelling.