Nvidia to invest $26 billion in developing open AI models

Nvidia is crossing the Rubicon: the world's largest maker of chips for artificial intelligence has announced its intention to invest $26 billion in developing its own open-weight AI models. The decision, recorded in officially filed documents, could fundamentally rewrite the logic of competition in the AI market and take the company far beyond its traditional role as a supplier of computing infrastructure.

A few years ago, Nvidia's position looked both completely unassailable and neutral: the company earned money from every participant in the AI race, selling GPUs to OpenAI and its rivals alike. It was the classic "sell shovels in a gold rush" strategy: while everyone else hunted for gold, Nvidia skimmed the profits from the hunt itself. The H100 and A100 chips became the de facto standard for training large language models, and the company's market capitalization soared. But that very success created a new vulnerability: dependence on the hardware business at a time when competitors, from Google with its TPUs to AMD and a whole range of startups, are relentlessly attacking Nvidia's monopoly in AI accelerators.

The decision to invest $26 billion in open models should be read in exactly this context. Open weights are not just a technical choice; they are a political and economic statement. Open-weight models, like those Meta releases under its Llama project, or those from China's DeepSeek, create whole ecosystems of developers who adapt, fine-tune and embed them in their own products. If Nvidia manages to secure a significant place in this ecosystem, it will stop being merely a hardware supplier and become the architect of the software environment on which the industry runs. That is a qualitatively different level of influence and, importantly, a qualitatively different source of revenue.

The scale of the announced investment is telling in itself. Twenty-six billion dollars is comparable to the annual R&D budgets of the world's largest technology corporations. For comparison: figures like this point to serious ambition, not pilot experiments. The bet on openness also looks tactically sensible: unlike the proprietary models of OpenAI or Anthropic, Nvidia's open models could attract the academic community, independent developers and corporate customers who on principle avoid dependence on closed platforms. This sharply widens the potential audience and speeds up adoption of the technology.

The consequences of this step for the market could be considerable. Until now, OpenAI and Anthropic have operated in a world where Nvidia was a partner, not a rival. That is changing: the largest supplier of computing power is becoming a player in the models themselves.

This creates a potential conflict of interest that will inevitably affect business relationships. Will companies that compete with Nvidia at the model level still be able to trust it as a chip supplier? The question is already hanging in the air in Silicon Valley.

In addition, Nvidia's entry into open models increases pressure on DeepSeek and other open initiatives: they now face a player with practically unlimited resources and deep expertise in hardware optimization.

Strategically, Nvidia is playing the classic game of vertical integration, only for the AI era. By controlling both the hardware and the models, the company can optimize the entire stack, from silicon to application, in a way none of its current competitors can. Open models trained and optimized specifically for Nvidia chips will run better on them, which gives developers a strong incentive to stay within the company's ecosystem. This is not just business diversification; it is an attempt to become indispensable at every level of the technology stack at once.

The AI industry is entering a new phase in which the lines between hardware makers, model developers and platform players are finally blurring. Nvidia's decision to invest $26 billion in open AI models is not just corporate news; it is a signal that the rules of the game are changing right now. A company that only yesterday was a neutral arbiter of the technology race is now lining up at the start as one of the main contenders.

*Meta has been designated an extremist organization and banned in Russia.