All eyes are on Nvidia as it is expected to announce this week’s most intriguing tech acquisition: A purported $13 billion purchase of Hugging Face, a platform dedicated to the sharing of open-weight AI models and benchmarks.

Currently recognized as a prime target for a squad of reward-hacking OpenAI agents, Hugging Face sits at the core of the community of developers focused on creating and implementing LLMs outside the realm of frontier labs. Consider it the GitHub equivalent for the AI age.

Speculations regarding this acquisition follow Nvidia’s $6 billion deal with Poolside, a builder of open-weight models, which will result in the majority of its workforce transitioning to the semiconductor titan. Additionally, two weeks prior, Stripe obtained OpenRouter, the leading supplier of open-weight models for businesses, for upwards of $7 billion.

This influx of investment into a sector that thrives on sharing resources highlights emerging trends in the AI landscape.

For Nvidia, minimizing reliance on partnerships with major hyperscalers and frontier labs is crucial. This is especially relevant as significant AI model developers like OpenAI and Google are simultaneously creating their own inference chips, such as OpenAI’s Jalapeño, which was unveiled this week. If model developers are producing chips, Nvidia aims to secure a piece of the model development pie.

Nvidia has its own Nemotron line of open-weight models, but their adoption has been limited. By annexing the largest developer community in the U.S. focused on open models, the firm will gain access to a large user base that can be directed toward its chips and standards.

Moreover, there are escalating concerns about AI inference costs, prompting companies to investigate more affordable models crafted by Chinese companies such as Moonshot, DeepSeek, and Alibaba. Although current adoption rates stand at a modest level, with only 6% of companies leveraging open-weight models, according to spending data compiled by Ramp, and just 2% of software engineers assessed by Jellyfish, which develops tools for coders.

Nik Albarran, the AI product lead at Jellyfish, informed TechCrunch that open-weight models are mainly utilized by firms whose offerings depend on repeated inference tasks, like customer service chat functionalities. Given that these tasks involve high volumes and frequent repetition, an open-weight model can be optimized for cost-effective responses.

This is notably the framing Stripe has adopted in discussing its OpenRouter acquisition. “Tokens serve as the main currency for companies developing AI solutions, and it’s evident that the tangible economic potential hinges on efficiently managing limited computing resources,” Patrick Collison, Stripe’s co-founder and CEO, remarked in a statement.

In contrast, for coding and agent-related tasks, varying requests and deeper reasoning indicate that frontier models often excel, partly because proprietary labs provide easier access, and sometimes offer token subsidies. Albarran notes that as organizations refine their AI workflows, migrating to open models will become more feasible. Still, the primary motive for companies exploring these models now is for control and adaptability, rather than cost concerns.

“There aren’t many companies that find themselves in that situation yet… [but] if prices continue to rise from frontier labs, an increasing number of companies will have no choice but to at least contemplate it,” Albarran shared with TechCrunch. “When your AI-driven processes are significantly more developed, that’s when investing in self-hosted models truly makes sense.”

Lin Qiao, the CEO of Fireworks, a prominent router and host for open-weight models aimed at corporate clients, is frequently mentioned as a possible acquisition target for a major tech firm. Qiao stated that her company manages 40 trillion tokens daily, surpassing both Gemini’s and OpenAI’s APIs in volume.

Fireworks focuses on model diversity: As LLMs expand and enhance, it will become increasingly simpler for organizations to tailor them to their specific requirements. “Every single app company should contemplate bringing on an internal researcher,” she remarked to TechCrunch last week. “They can utilize their product and the data it generates to develop their own model. The future will truly be about specialized intelligence. Every company ought to have its own model tailored for each use case, and this will occur organically.”

It’s easy to overlook how early we are in AI’s evolution as both a tool and a commercial venture. However, the supremacy of OpenAI and Anthropic is not a foregone conclusion. As tech giants seek to hedge their investments in the largest labs, the charm of open technology is proving to be difficult to resist.