Moonshot pauses new Kimi K3 subscriptions as demand outruns GPU capacity

Moonshot AI paused new subscriptions to Kimi K3 on July 19, two days after launching the 2.8-trillion-parameter open-weight model, saying a demand surge had pushed it close to the limits of its GPU capacity.

“Over the past 48 hours, demand has pushed close to the limits of our current capacity,” the company said on its official Kimi account, adding that “Kimi K3 has received far more love than we expected, and our GPUs are feeling it.” Existing subscribers are unaffected; Moonshot said it would reopen new slots in batches as it adds infrastructure.

The pause is a capacity problem, not a retreat. Kimi K3 is an open-weight model that Moonshot describes as the “world’s first open 3T-class model,” built on what it calls Kimi Delta Attention and Attention Residuals, with native vision and a 1-million-token context window. Serving a model that large at consumer scale is costly, and the crunch shows how fast inference demand can outrun a lab’s hardware when a release catches on.

Moonshot is also splitting membership into two tiers to better match compute to workload: a general Kimi Membership covering its web, app and work products, and a separate Kimi Code Membership for programming. The company said it still plans to publish Kimi K3’s full open weights around July 27.

Whether the batched reopening keeps pace with demand – or the wait for open weights simply pushes users to self-host – will be the test of how durable the launch momentum proves.