Sam Altman warns OpenAI capacity crunch will delay releases

AI adoption is moving faster than data centers can handle. OpenAI CEO Sam Altman admitted on X that immense demand for ChatGPT's new image generation tool is triggering severe capacity constraints, which will cause upcoming product releases to be delayed.

The feature driving the crunch can mimic hand-drawn aesthetics like Studio Ghibli. Altman said the team hasn't been able to catch up since launch, with engineers working nights and weekends just to keep things online. At one point on Monday, ChatGPT added one million users in a single hour.

Usage metrics show how fast the platform is snowballing. ChatGPT now reports 500 million weekly active users and 20 million paying subscribers, up from 300 million and 15.5 million at the end of 2024.

Triage mode in the data center

To keep the lights on, OpenAI is actively rationing access. The company paused rolling out the new image generator to free ChatGPT accounts and temporarily shut off video generation access for new Sora users.

Altman warned users to brace for more delays, occasional service slowdowns, and glitches while the team scrambles to bring infrastructure up to par.

Why it matters

If you were expecting OpenAI's next big model drops to arrive right on schedule, it is time to recalibrate. Even existing users should expect performance hiccups during peak hours.

We think this crunch highlights the brutal reality of multimodal AI infrastructure. Text queries are cheap, but visual generation is an insatiable compute sink. Forcing OpenAI to restrict Sora access reveals that compute shortages remain the chief bottleneck across the industry. The real race is no longer just training better models, but figuring out how to serve millions of simultaneous generation requests without crashing the grid.