Moonshot AI pauses Kimi K3 subscriptions

The GPU capacity crunch for Kimi K3 highlights that even for heavily backed AI startups, serving massive multi-trillion parameter open-weights models at scale is an operational bottleneck that compromises user acquisition.

* Compute limits are the ultimate equalizer: Moonshot AI's infrastructure was rapidly overwhelmed by the massive inference requirements of Kimi K3's 2.8 trillion parameter architecture.

* Defensive prioritization: Temporarily blocking new subscribers is a necessary trade-off to maintain service reliability and retain existing high-value customers.

* The hidden cost of open weights: While releasing model weights democratizes access, hosting those models with low latency and high availability remains a highly capital-intensive challenge.