Sam Altman, CEO of OpenAI, has announced that the company will boost its computing power to over one million GPUs by the end of 2025 — an impressive figure that’s five times the GPU capacity of Elon Musk’s xAI, which utilizes Nvidia’s H100 graphics cards.
According to reports, Altman proudly praised his team, writing: “I’m proud of my team, but now they have to figure out how to increase this number by 100 times.” While this might sound humorous, given Altman’s track record, it appears more like a serious plan than a joke.
OpenAI to Surpass One Million GPUs by End of 2025
In early 2025, Altman admitted that OpenAI had to delay the release of GPT-4.5 due to a shortage of GPUs. The global scarcity of Nvidia chips became a wake-up call for him. Since then, expanding computational capacity has become Altman’s top priority.
Altman has embarked on national-scale projects that go far beyond simple IT upgrades. Reaching one million GPUs this year is not just a number — it would make OpenAI the world’s largest consumer of AI computing power.
Altman’s bigger goal is to reach 100 million GPUs — a value estimated at around \$3 trillion, roughly equivalent to the United Kingdom’s GDP. Under current conditions, achieving this scale is out of reach due to factors such as energy supply constraints.
Nevertheless, as seen in OpenAI’s Texas data center — the largest single-site facility in the world, currently using 300 megawatts of power and expected to reach 1 gigawatt by mid-2026 — Altman is building the infrastructure designed to realize AGI (Artificial General Intelligence).
In addition to Microsoft Azure, OpenAI is working with Oracle to build dedicated data centers and is even considering Google’s TPU accelerators. This initiative is part of a broader race joined by companies like Meta and Amazon, which are developing custom chips and investing in HBM memory.
Crossing the one million GPU milestone is a major and historic achievement for OpenAI. In a world where GPUs are the driving force behind AI models, Altman’s ambitious vision is pushing the boundaries of computational limits and shaping the future of AI infrastructure.
