OpenAI has announced a way for enterprise customers to guarantee compute resources for the company’s AI services through a new offering called Guaranteed Capacity.
Capacity has become less assured as leading-edge AI models are more widely adopted by large-scale enterprises, prompting some to seek greater certainty. With the new commitments, customers can choose one-, two-, or three-year plans, with discounts for longer terms.
On the form, OpenAI lets customers choose upward of 1 billion tokens per minute of capacity.
“Customers are increasingly asking us for certainty on capacity. As models get better, we expect that the world will be capacity-constrained for some time,” said OpenAI CEO Sam Altman on X.
Guaranteed Capacity will be available on a first-come, first-served basis, according to OpenAI, with sales continuing until capacity sells out. It may look to do more of these capacity sales in the future if there is enough interest and available capacity.
The battle for the enterprise market
OpenAI has been trying to win back the enterprise lead after Anthropic reportedly surpassed it in enterprise customers and enterprise revenue run rate. To that end, it made the GPT-5.5 update focused on several key benchmarks that appeal to users of the API, agentic tools, and coding tools, with the consumer version primarily aimed at reducing hallucinations.
It also embedded many more tools into Codex AI, making it a “super app” for developers, while launching a mobile app to go alongside it.
It is not the only AI vendor trying to grab some of Anthropic’s lucrative enterprise and coding market, with a good portion of Google’s I/O announcements dedicated to agentic and coding services. Elon Musk’s xAI also recently launched Grok Build, the first coding tool developed by the research lab.
At first glance, it is surprising that OpenAI is so willing to guarantee capacity, given its struggles to acquire enough compute power to train its AI models and provide service, although we do not know how much capacity is on offer. Anthropic has been even more hampered by downtime from excessive demand for its AI services.
OpenAI early buildout paying dividends
That said, OpenAI is in a stronger place to offer this guaranteed capacity. It has received far more funding than Anthropic, enabling it to sign up more capacity. It has also made larger deals with hyperscalers such as Amazon Web Services, Oracle, and Microsoft.
That is on top of the Stargate project, backed by SoftBank and Nvidia, which, even with the recent pullback, is still scheduled to provide OpenAI with more than 7 gigawatts of capacity by the end of the decade.
But the entire industry looks to be sandwiched between serving customers and improving its own models. Google, which has its own cloud computing division, has reportedly begun scrutinizing employee use of compute resources, with many projects receiving sufficient space only if they can improve the Gemini AI model.
At the same time, Alphabet CEO Sundar Pichai said it could be making far more from its Google Cloud division if it were not keeping some of its compute capacity for its own benefit.
Also read: Elon Musk plans to appeal after a jury rejected his OpenAI lawsuit as untimely.