OpenAI’s compute scale plan
- OpenAI announced plans to scale compute to 30 GW by 2030 and has already secured over 8 GW of its 2025 target. - The disclosure frames massive hardware commitments as necessary support for future intelligent systems and large models. - That compute expansion will intensify demand for data-center capacity, suppliers, and long-term capital across the AI stack (x.com).
OpenAI says it is now planning for 30 gigawatts of compute by 2030, a threefold increase from its 10-gigawatt target announced in January 2025. (x.com) The company said on April 22 that it has already identified more than 8 gigawatts of the 10 gigawatts it committed to secure for 2025. In power terms, one gigawatt is roughly enough electricity for a small U.S. city. (x.com) (cdn.openai.com) OpenAI has been assembling that capacity through Stargate, the infrastructure effort it launched with SoftBank, Oracle and MGX on January 21, 2025, with a stated goal of investing $500 billion in U.S. artificial-intelligence infrastructure over four years. (openai.com) By July 2025, OpenAI and Oracle said they had 4.5 gigawatts of additional U.S. capacity under agreement, bringing Stargate to more than 5 gigawatts under development, including the Abilene, Texas, site. Oracle had already begun delivering Nvidia GB200 racks there, and OpenAI said it had started early training and inference workloads. (openai.com) The buildout widened again in September 2025, when OpenAI, Oracle and SoftBank said five new U.S. sites would bring Stargate to nearly 7 gigawatts of planned capacity and more than $400 billion of investment over three years. OpenAI added a 1-gigawatt Michigan campus in November 2025. (openai.com 1) (openai.com 2) OpenAI tied the bigger 2030 target to demand for “intelligent systems,” its term for AI products that do more than answer prompts and instead run longer tasks across software and services. In a March 31 funding announcement, the company said durable access to compute improves research, products, access and delivery costs at scale. (x.com) (openai.com) That demand has already pulled in new capital and new suppliers. OpenAI raised $122 billion in committed capital on March 31, 2026, then added a $38 billion infrastructure partnership with Amazon Web Services in November 2025 and a January 2026 partnership with SoftBank-backed SB Energy that included a 1.2-gigawatt Texas data-center lease. (openai.com 1) (openai.com 2) (openai.com 3) The numbers also show how much of the AI race is shifting from model releases to physical buildout. OpenAI’s own infrastructure blueprint argued in September 2025 that advanced chips, data centers, power plants and transmission capacity now sit on the critical path for AI growth. (cdn.openai.com) A 30-gigawatt target does not mean 30 gigawatts will be online tomorrow. It means OpenAI is signaling to utilities, chipmakers, builders, financiers and cloud partners that it expects to keep buying power, land and hardware at a scale closer to national infrastructure than to a normal software company. (x.com) (openai.com)
