Moonshot halts new Kimi K3 subscriptions as capacity tightens
Moonshot said demand for its new Kimi K3 model reached the limits of available compute within 48 hours, prompting a pause on new paid subscriptions.
By Wei-Lin Zhao · AI Correspondent
· 2 min read
Moonshot has paused new subscriptions for Kimi K3 after demand for the recently released AI model pushed close to the limits of the company’s available capacity within 48 hours, according to a company post on X. The company did not disclose pricing changes, subscriber numbers, GPU inventory, or the amount of compute now reserved for existing customers.
The pause does not affect current subscribers, Moonshot said. New subscription slots are expected to reopen gradually, though the company did not give a timetable or say what capacity additions would be required before sales resume.
For AI operators, the move is a reminder that open model distribution does not remove the serving burden for companies trying to package models as hosted products. Kimi K3 has been positioned as an open model, but Moonshot’s commercial subscription flow still depends on enough inference capacity to handle usage spikes. The company’s statement frames the issue as demand management rather than a product withdrawal.
Moonshot splits Kimi subscriptions into two products
Moonshot also said it is separating its subscription offering into two plans. A “Kimi Membership” will cover web, mobile app and work-related features, while a “Kimi Code Membership” will be used for programming workflows.
The company said the change is intended to distribute computing power more evenly and maintain a stable user experience. It did not say whether the two memberships will have different usage caps, prices, model access, priority levels, or performance guarantees.
The split is notable because coding workloads can be particularly compute-intensive and can create different usage patterns from general chat or workplace tasks. Moonshot did not provide enough detail to assess whether the new structure is mainly a capacity control, a packaging change, or an early step toward pricing heavier users separately.
Alibaba moves with Qwen 3.8
Moonshot’s capacity pause comes as Alibaba is promoting Qwen 3.8, a competing model. Alibaba’s rival system will be open weight for the first time in a while, according to the report, and a paid preview version is already available at a steep discount.
That timing matters for the Chinese AI model market, where product access, benchmark positioning and infrastructure availability are moving together. A model can attract attention quickly, but sustained adoption depends on whether users can reliably get access through hosted products, APIs or downloadable weights.
Moonshot has said only that demand has strained its current capacity and that it will reopen new slots gradually. Until the company provides capacity, pricing or usage data, the scale of the Kimi K3 bottleneck remains unclear.
This story draws on original reporting from The Decoder.