OpenAI mulls selling compute capacity down the road, but not anytime soon

OpenAI is sitting on a massive and rapidly growing pile of computing infrastructure. The natural question: will it start selling access to some of that horsepower? The answer, at least for now, is a polite “not yet.”

In an August 20, 2025 interview, OpenAI CFO Sarah Friar made clear that the company isn’t actively pursuing plans to sell compute capacity or offer infrastructure services to outside customers. The priority remains internal: feeding the voracious appetite of OpenAI’s own AI workloads, which continue to scale at a pace that leaves little room for sharing.

The AWS parallel everyone keeps bringing up

The comparison is inevitable. Amazon Web Services started as internal infrastructure that Amazon built for its own e-commerce operations, then realized it could rent spare capacity to the rest of the world.

Friar acknowledged this parallel explicitly, noting that monetizing infrastructure expertise is a prospective future business avenue. The logic is straightforward: as OpenAI’s hardware ages and newer, more powerful chips come online, older equipment doesn’t have to sit idle or get scrapped. It could be repurposed for external customers whose workloads don’t demand cutting-edge silicon.

A compute empire under construction

OpenAI’s infrastructure footprint has grown dramatically through a web of strategic partnerships. The company maintains compute relationships with Microsoft, AWS, Oracle, AMD, and Nvidia, a diversified portfolio designed to ensure it never becomes too dependent on any single provider.

The most ambitious piece of this puzzle is the Stargate project, which targets over 10 gigawatts of compute capacity.

The $38 billion partnership with AWS underscores just how seriously OpenAI is investing in raw infrastructure. Training and running frontier AI models requires an almost obscene amount of computing power, and that demand shows no signs of plateauing.

This buildout strategy also extends to custom hardware development. Rather than relying solely on off-the-shelf chips from Nvidia, OpenAI has been working on optimizing its infrastructure stack from the ground up.

The economics of eating your own cooking

There’s a financial tension at the heart of OpenAI’s infrastructure story. The company is spending enormous sums on compute, and those costs have been rising roughly in lockstep with revenue growth.

If OpenAI eventually offers infrastructure services to external customers, it creates a new revenue stream that could help offset the staggering cost of maintaining its compute empire. The company would essentially be double-dipping: using its infrastructure for internal AI development while simultaneously renting out any slack capacity or older hardware to paying customers.

Reports from early to mid-2026 indicate that OpenAI remains firmly in accumulation mode, with no active external sales efforts announced.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.