What does 'The service is currently experiencing high load' mean in Veo?

A Veo request is a long-running operation: the API accepts it immediately and the actual work is scheduled afterwards. Capacity is checked when that work runs, not when the request is accepted — so the error surfaces one to several minutes in, inside an operation that looked healthy.

That is also why it is not a rate limit. Rate limiting answers 429 at submission time and is handled by rotating to another key. High load happens later, on the machine that was supposed to render your clip, and rotating keys does not help because the shortage is on Google's side, not ours.

The heaviest configurations queue for the scarcest hardware. In our logs the combination that fails most often is 4K output together with a last frame — the interpolation pass on top of a 4K render is the single most demanding thing the model does.

If you keep hitting it on a deadline, the pragmatic route is to generate at 1080p, which on Veo 3.1 and Veo 3.1 Fast costs exactly the same as 720p and almost never queues, and upscale afterwards if you truly need 4K delivery.

The generation is restarted for you, up to four times, reusing the same API key so the operation stays in the same project. The studio shows a live counter — *retry 2 of 4* — so a job that seems to be taking a long time is usually a job that is being retried rather than a stuck one.

If all four attempts hit the same wall, the job is marked failed with the code HIGH_LOAD and the money goes back to your balance in the same transaction. Nothing is left in limbo and there is nothing to claim.

Your part is simply to try again later, or to drop from 4K to 1080p. High load is transient and follows Google's regional load curve; the same configuration that fails four times in a row often succeeds twenty minutes later.