Altman thanks Huang as Nvidia expands $30B AI infrastructure push across cloud platforms

Jensen Huang, the chief executive of Nvidia, stood before investors at the Morgan Stanley Technology, Media & Telecom Conference earlier this month and made a declaration that rippled through the AI industry: Nvidia would invest $30 billion in OpenAI, framing the bet as one of the last chances to back a transformative company before it reaches the public markets. The announcement crystallized what has been quietly happening behind the scenes for months—a massive, coordinated effort to ensure that OpenAI's computing infrastructure keeps pace with the explosive growth of ChatGPT and the systems built on top of it.

Sam Altman, OpenAI's chief executive, responded publicly on X with a note of gratitude directed at Huang. "Very grateful to Jensen for working to expand Nvidia's capacity at AWS so much for us!" Altman wrote. The comment was brief but revealing: it acknowledged not just the investment itself, but the grinding operational work required to make it real—the expansion of graphics processing units across Amazon Web Services, Microsoft Azure, and Oracle Cloud Infrastructure. These three platforms now form the backbone of OpenAI's computational reach, and Nvidia has been, in Huang's own words, working "like mad" to ensure the company has the GPU resources it needs to support its rapidly expanding AI systems.

The scale of this infrastructure play extends beyond OpenAI alone. Nvidia is simultaneously expanding its support for other leading AI companies, including Anthropic and Meta Platforms, as the competition to control the computing foundation for next-generation AI models intensifies. What was once a relatively straightforward market—Nvidia selling training chips to AI labs—has become a complex, multi-layered competition for dominance across the entire AI stack.

But the investment and capacity expansion are only part of the story. Nvidia is also engineering a fundamental shift in its own business strategy. The company is developing a new processor specifically designed for AI inference computing—the type of processing that allows trained AI models to actually respond to user queries in real time. This marks a departure from Nvidia's traditional strength in training chips, the domain where it has built near-total market dominance. Analysts estimate Nvidia controls more than 90 percent of the GPU market for training, a position built on its Hopper, Blackwell, and Rubin GPU series. But as the AI industry matures, the bottleneck is shifting. Training happens once; inference happens millions of times. GPUs, designed with training in mind, are not optimally suited for this workload, and Nvidia recognizes that controlling the inference layer is essential to maintaining its position.

OpenAI has already agreed to become one of the largest customers for this new inference processor, according to reporting from The Wall Street Journal. The announcement is expected at Nvidia's GTC developer conference in San Jose next month. To build this chip, Nvidia is drawing on technology from Groq, a chip startup it acquired in a roughly $20 billion deal late last year. The acquisition, which seemed significant at the time, now appears to have been a strategic move to acquire the engineering talent and intellectual property needed to compete in the inference space. Huang's $30 billion commitment to OpenAI, combined with the new chip development and the multi-cloud infrastructure expansion, represents a bet that Nvidia's future depends not just on selling hardware, but on becoming the indispensable partner in every stage of how AI systems are built, trained, and deployed. For OpenAI, the arrangement ensures that computational capacity will not be a constraint on growth. For Nvidia, it is a way to lock in a dominant position in what may be the most consequential technology shift of the decade.