Chinese start-up Moonshot AI has suspended new subscriptions for its groundbreaking new model Kimi K3 as surging demand squeezes its computing power, underscoring the challenges Chinese AI labs face in serving their products to global users.
“Over the past 48 hours, demand has pushed close to the limits of our current capacity,” Beijing-based Moonshot AI said in a statement posted on X on Sunday. “Our GPUs [graphics processing units] are feeling it.”
It had “temporarily paused new subscriptions”, adding that existing subscribers were not affected. “We’re adding capacity as fast as we can and will reopen new subscription spots in batches,” said the company.
Late last week, Beijing-based Moonshot released Kimi K3, a 2.8 trillion-parameter large language model. It has outperformed top US models GPT-5.6 Sol and Claude Fable 5 on some benchmarks, including Arena AI’s ranking for front-end code development.
Moonshot is currently serving the massive K3 model through a cloud application programming interface, and plans to release the model’s weights by July 27, which would make it the world’s largest open-weight frontier model so far.
Its frontier capabilities have drawn global attention and added fuel to growing anxiety in the US about the rapidly narrowing lead it holds over Chinese AI labs. However, many users noted that Kimi K3 operated noticeably slower than the top US alternatives.
Chinese AI labs continue to face computing power shortages as the country’s chip industry strives to catch up under US export control measures, which have restricted access to advanced chips and chipmaking equipment.