Kimi K3 got so popular Moonshot had to stop taking new subscribers
Three days after launch, demand for Kimi K3 pushed Moonshot's GPU capacity to its limit within 48 hours. New subscriptions are paused while existing users get priority and capacity expands.
Kimi K3 launched, demand hit Moonshot's GPUs so hard they had to stop taking new subscribers.
Three days after we covered Kimi K3's launch, Moonshot AI suspended new subscriptions to the model on July 19. Demand pushed the company's GPU capacity to its limit within 48 hours of launch.
What Moonshot actually said
Moonshot's own statement frames this as a deliberate choice to protect service quality: all available compute is now dedicated to existing subscribers rather than spreading thinner across a growing queue, while the company accelerates capacity expansion. New subscription slots are set to reopen gradually as that capacity comes online, not all at once.
Why this is the more interesting number
A 2.8 trillion parameter open weight model release is one story. A launch that overwhelms a major lab's GPU capacity within 48 hours is a different, more concrete one: it is a direct read on how much real demand exists for open weight frontier models right now, measured in the thing that actually constrains AI companies, compute, not in benchmark scores or social media reaction.
Why a build studio cares
Anyone evaluating Kimi K3 for a build now has an availability question sitting on top of the technical one. Capacity-driven subscription freezes are a recurring pattern across this AI cycle, and they are a real operational risk to plan around if a client build depends on a specific model's API being reliably available, not just performant.
Next step: see Global Times and BeInCrypto's coverage of the suspension. If model availability risk is a live concern for a build you're planning, write to us at hello@gattyworks.com.