Why Have Open-Weight Model Companies Become Acquisition Targets in Silicon Valley?

The artificial intelligence sector is witnessing a striking shift in investment trends, as the attention of major tech companies in Silicon Valley turns toward firms specializing in open-weight models. With news leaking about massive billion-dollar deals, a fundamental question emerges: What makes a sector fundamentally built on the principle of «free sharing» suddenly transform into the most valuable acquisition target?

A Wave of Massive Deals

Rumors dominate the headlines about Nvidia's intention to acquire the Hugging Face platform for an estimated value of around $13 billion. This platform is the beating heart of the developer community that builds and releases large language models not owned by the leading labs, to the extent that many describe it as the «GitHub» of the AI era.

This deal was not isolated but came as part of a series of successive moves:

• Nvidia's agreement with Poolside, a company specializing in building open models, valued at $6 billion, with most of its employees moving to the chip-making giant.
• Stripe's acquisition of the OpenRouter platform, the most prominent provider of open models for companies, at a value exceeding $7 billion.

This enormous amount of capital flowing toward a sector built on the principle of free availability reflects the latest trends in the AI industry.

Nvidia's Motives for Acquisition

Through these deals, Nvidia seeks to reduce its reliance on partnerships with major cloud infrastructure providers and leading labs. This concern grows more pressing as major companies like OpenAI and Google move toward manufacturing their own chips for inference, similar to the «Jalapeño» chip that OpenAI recently unveiled.

The logic is clear: if model-building companies are making their own chips, then Nvidia in turn wants a share of the model-building market. Despite owning a family of open models called «Nemotron», its adoption has not been significant. By controlling the largest space for developing open models in the United States, the company ensures access to a huge base of users who can be directed toward its chips and standards.

The Question of Cost and Adoption

Questions are mounting about the cost of inference in AI, pushing companies to explore cheaper models built by Chinese companies such as Moonshot, DeepSeek, and Alibaba. Nevertheless, the adoption rate remains limited though in continuous growth; data indicates that only about 6% of companies use open models, while the percentage drops to 2% among software engineers.

Specialists explain that open models are primarily used by companies whose products rely on recurring inference workloads, such as chat services for customer support. These high-volume, highly repetitive tasks allow an open model to be tuned to answer their questions at low cost.

Control Before Savings

The statements of Patrick Collison, co-CEO of Stripe, reveal this vision when he says that «tokens are the central currency for companies building with AI», emphasizing that the real economic value will depend on the good utilization of scarce computing resources.

As for programming tasks and agentic tasks that require deeper reasoning and diverse requests, proprietary frontier models still often prevail. Experts believe that the main reason companies turn to open models today is the need for control and customization capability, not merely cost reduction. However, they expect that as the prices of leading labs rise, more companies will be forced to consider this option.

The Future of Specialized Intelligence

Companies like Fireworks — which is among the most prominent platforms for routing and hosting open models for companies — are betting on the principle of «model diversity». Its CEO indicates that the company processes about 40 trillion tokens daily, exceeding the APIs of both Gemini and OpenAI.

The company believes that as language models proliferate and improve, it will become easier for companies to train them according to their own needs, to the point where it suggests that each application employs an in-house researcher to build its own model using its product's data. The future, according to this vision, is one of «specialized intelligence», where each company owns its own model for each use case.

It's easy for us to forget that we are still at a very early stage in the development of AI as a tool and as a business. And while the dominance of OpenAI and Anthropic currently appears clear, it is not an inevitable fate. As tech giants seek to diversify their bets away from the largest labs, it becomes evident that the appeal of open technology has become too difficult to resist.

✦ بقلم فريق دروب أيديا