OpenAI hits its 10 GW compute target ahead of schedule

OpenAI has hit one of the most ambitious infrastructure targets in the AI industry ahead of schedule: the company has already secured contracts for 10 GW of compute capacity. The milestone was originally planned for 2029, but the last 3 GW were contracted in just 90 days.

The plan accelerated sharply

For OpenAI, this is not just a nice number for reports and presentations. In the era of large language models, compute has become a strategic resource on par with data, engineers and money. The earlier a company locks in future compute, the lower the risk that releases of new models, agent services and enterprise products will stall on a plain shortage of infrastructure. In effect, OpenAI is booking room to grow in advance, while competitors keep fighting for the same data centers, GPUs and power connections.

The pace of recent months is especially telling. The 10 GW goal once looked like a marker for the second half of the decade; now it has been met well ahead of schedule. Three additional gigawatts in 90 days is a very aggressive tempo even by the standards of a market where such deals usually drag on for months. A jump like this reflects not only OpenAI's own ambition but also the willingness of suppliers, investors and infrastructure partners to quickly reserve large blocks of capacity for it.

Contracts versus reality

It is important not to confuse signed contracts with all of the capacity going live at once. The agreements alone do not mean OpenAI can load all 10 GW with model training and inference today. Some capacity will come online in stages: in places sites still need to be finished, in others hardware has to be delivered, and in others the company must wait for grid connections. At this scale, delays and phased launches are almost unavoidable, so "securing access" and "getting everything at once" are formally two different stories.

Even so, the result changes the picture. The company is reducing in advance the risk that growing demand for ChatGPT, the API and future products runs into a shortage of hardware, electricity or available sites. For the whole industry, it is a signal that the AI race is moving further from demos and releases toward energy, construction and contracts.

The practical value of such agreements for OpenAI is already clear:

• Headroom for training larger models
• The ability to scale inference for ChatGPT and the API
• Resources for new agent and enterprise services
• A stronger position in negotiations with suppliers and investors

Money and allies

Another important nuance is who pays for and backs these deals. OpenAI often raises money from investors while relying on partners who then provide compute services or related infrastructure. The result is a tight link between capital and capacity: belief in the company's future growth translates directly into access to data centers and energy. The more trust in OpenAI's business, the easier it is to reserve new capacity in advance, and the lower the chance of running out of resources at a critical moment.

The arrangement is rational for partners too. They get a customer with almost guaranteed multi-year demand and can expand infrastructure for workloads that are already confirmed. For investors, it is a bet not only on the next model but on a future share of a critical part of the AI market. That is why the 10 GW news reads less like an accounting update and more like an indicator that the contest between OpenAI, Google, Anthropic and other players is increasingly about energy, sites, racks and the speed of data center construction.

What this means

The main bottleneck in AI today is not only talent and algorithms but infrastructure. Whoever locks in energy and compute first gains an advantage that lasts not for one model release but for several years.