OpenAI and Google throttle Sora and Nano Banana over melting GPUs

If you were hoping for some AI-generated fun over the holiday weekend, you'll have to make every prompt count. Google and OpenAI have both cut generation limits for Sora and Nano Banana Pro, citing overwhelming demand.

Melting GPUs and shrinking daily quotas

Bill Peebles, who leads the Sora team at OpenAI, announced that free users are now restricted to just six video generations per day. His reasoning was short and to the point: "Our gpus are melting."

Unlike previous cutbacks, Peebles did not label the restriction as temporary. Instead, he noted that users can simply purchase additional generations as needed, part of a wider effort to monetize the video platform. Generation caps for ChatGPT Plus and Pro subscribers remain untouched, though exact numbers were not disclosed.

Google is scaling things back as well. Free users on Nano Banana Pro, which rolled out just last week, are now capped at two images a day down from three, according to 9to5Google. Google warned that limits could change frequently and without notice as demand surges. The company also appears to be dialing back free tier access to Gemini 3 Pro.

Why it matters

We think the era of generous, free-for-all access to frontier generative models is slamming directly into the compute wall. Video and cutting-edge image models are pure server burn, and neither company can afford to carry the computing bill for millions of casual users indefinitely.

For everyday creators, six video generations or two images per day gives you zero margin for error. You cannot tweak prompts or experiment with styles at that pace; it effectively turns the free tier into a bare-bones sample platter. OpenAI is smartly turning an infrastructure bottleneck into a direct nudge toward pay-as-you-go revenue. Meanwhile, Google shaving a single image off daily limits looks like a desperate bid to preserve GPU capacity. The door to top-tier free AI tools is closing fast.