OpenAI says it has reached 10 gigawatts of AI compute capacity in the United States, hitting a goal it had originally set for 2029. The company says the milestone arrived several years ahead of schedule, underscoring how quickly the infrastructure race behind artificial intelligence is moving.
A compute target reached ahead of time
The company said in a blog post that it has reached its 10 gigawatt goal for AI compute capacity in the United States. That target had been set for 2029, meaning OpenAI now says it has achieved the benchmark well before its planned timeline.
The scale is significant because AI compute capacity is not only about chips or software. It also depends on power, data centers, contracts and the ability to bring large infrastructure online fast enough to support demand.
The source gives one comparison for the size of the commitment: one gigawatt can power roughly 750,000 U.S. homes at once. By that measure, 10 gigawatts represents a very large power footprint tied to OpenAI’s AI infrastructure ambitions.
The last 90 days mattered
OpenAI said three of the 10 gigawatts were contracted in the last 90 days alone. That detail suggests the company’s progress was not only the result of long-running commitments, but also of recent dealmaking.
One named supplier in that recent push is Amazon, which accounted for 2 gigawatts of the capacity contracted during that 90-day period. The source does not provide further details on the structure of the Amazon arrangement, so the key fact is the size of the commitment and its timing.
For OpenAI, reaching the goal early gives the company a stronger infrastructure story at a time when AI systems require large amounts of computing power. For the broader market, it shows how quickly major AI companies are trying to secure the energy and data center capacity needed to train and run advanced systems.
Expansion continues, but not every project is moving forward
OpenAI plans to keep expanding its compute capacity in the coming years. Reaching the 10 gigawatt target does not appear to be the end of its infrastructure buildout, based on the company’s stated plans.
At the same time, the source article notes that OpenAI has pulled back from several projects. That makes the milestone more complicated than a simple story of uninterrupted expansion.
The reported setbacks include:
- An expansion of the Stargate data center in Texas was rejected.
- A project in the UK was paused because of high energy costs.
- A site in Norway was dropped entirely.
Those examples show that AI compute growth is shaped by more than ambition. Energy costs, site decisions and local project outcomes can affect where capacity is built and which plans survive.
The source does not say these setbacks prevented OpenAI from reaching its 10 gigawatt goal. Instead, the picture is one of rapid contracting alongside selective retreats from particular locations or expansions.
Stargate remains part of the backdrop
The Stargate project is one of the key infrastructure efforts mentioned in the source. It was announced in early 2025 as a $500 billion initiative with Oracle and SoftBank.
That figure gives a sense of the scale of the infrastructure commitments being discussed around AI compute. It also places the rejected Texas expansion in a broader context: large AI data center plans may be enormous in ambition, but individual pieces can still face practical limits.
The source does not provide new details about the full status of Stargate beyond the rejected expansion in Texas and the original announcement. For readers, the important point is that OpenAI’s compute expansion involves both major partnerships and difficult site-by-site execution.
Why the 10 gigawatt milestone matters
AI systems depend on compute capacity to operate at scale. When a company like OpenAI says it has secured 10 gigawatts of capacity in the United States, it is pointing to the physical infrastructure behind digital products and services.
The milestone also highlights how energy has become central to the AI business. The source’s comparison to U.S. homes makes the number easier to grasp: gigawatts are not abstract in this context, because they represent power that must be available for large-scale computing.
OpenAI’s announcement shows two things at once. First, the company says it has moved faster than expected toward a major U.S. AI compute goal. Second, the setbacks in Texas, the UK and Norway show that scaling compute is not simply a matter of announcing more projects.
The next phase, based only on what the source states, is continued expansion. OpenAI says it plans to keep growing its compute capacity in the coming years, even after reaching the 10 gigawatt target ahead of schedule.