AI-Economy

Kimi K3: Moonshot stops new subscriptions due to GPU shortage

3 min read
A technician monitors a rack of glowing GPU servers at night, a laptop in the foreground shows the Kimi chat interface with Moonshot AI branding Image generated with GPT Image 2
A technician monitors a rack of glowing GPU servers at night, a laptop in the foreground shows the Kimi chat interface with Moonshot AI branding

TL;DR Too Long; Didn’t read

Moonshot AI stopped new registrations for its AI model Kimi K3 on July 19, 2026, after demand overwhelmed GPU capacities within 48 hours. Existing subscribers retain access according to the company, new users will only follow in groups. The complete model weights for local use are still expected to be released by July 27.

Key takeaways

  • Within 48 hours of the Kimi K3 launch, server requests reached Moonshot's capacity limit.
  • New registrations are stopped, existing subscriptions continue unchanged according to the company.
  • Moonshot will separate membership for web and app from a separate Kimi Code membership.
  • The open model weights for self-hosting are still expected to be released on July 27.
  • Kimi K3 processes 2.8 trillion parameters and a context window of one million tokens.
  • Reports estimate Moonshot's valuation at over 20 billion dollars after a two billion round.

Moonshot AI has temporarily halted new registrations for its AI model Kimi K3. The chatbot provider is responding to a wave of demand that pushed its own graphics processors to their limits within 48 hours of the launch on July 16. Existing subscribers are not affected by the suspension.

Registration Stop Aims to Protect Existing Customers

The interest in Kimi K3 has significantly exceeded its own expectations, Moonshot AI announced on July 19 on the platform X, and the existing graphics processors have thus reached their capacity limits. The model was only launched on July 16 and had caused unusually high access numbers in a very short time. To secure ongoing operations for paying customers, the company will not open new subscriptions for the time being. Existing memberships will continue to function unchanged, as the available computing power remains primarily reserved for them.

New interested parties will be activated in controlled groups in the future as soon as additional server capacity becomes available. The company does not specify a fixed date for reopening. At the same time, Moonshot announced that it will split the existing pricing model: one membership will cover usage via the website, the app, and the office application Kimi Work, while a second, standalone membership is reserved for the programming environment Kimi Code. This separation is intended to prevent compute-intensive programming tasks from consuming the capacity for regular chat requests.

Five Tiers Range from Free to $159

Once registration reopens, a five-tier pricing structure will apply according to the official Kimi pricing page. The free tier Adagio allows simple chats, file uploads, and web searches, but only a limited credit for agent functions. The paid tiers Moderato, Allegretto, Allegro, and Vivace cost $15, $31, $79, and $159 per month, respectively, when billed annually, and unlock more simultaneous tasks, a higher agent credit, and faster Kimi Code usage. Those accessing the model through their own application via an interface are charged separately based on tokens: three dollars per million input tokens, with caching only 30 cents, and $15 per million output tokens. The registration stop does not change these prices; they remain the same for all pricing tiers.

Whether and how Moonshot will officially offer the service in Germany or the European Union remains open. The company does not mention any regional restrictions on its pages but also does not explicitly refer to a market launch in the EU. Access is currently possible via the website, an app, and the mentioned programming interface.

Kimi K3 Had Already Moved the Markets

Kimi K3 had already caused a stir before the capacity bottleneck: the open model with 2.8 trillion parameters and a context window of up to one million tokens triggered a brief stock price drop at Nvidia immediately after its launch and briefly helped Apple achieve the highest market value in the world, as beckmann.ai reported. The current registration stop reads to observers as evidence that demand for the affordable open model is real, not just on paper.

According to consistent reports from Yahoo Finance and Crypto Briefing, Moonshot is now valued at over $20 billion, following a recent funding round of around $2 billion. The annual revenue figures from the two outlets differ: Yahoo Finance cites over $300 million, while Crypto Briefing puts it at over $200 million — neither figure is independently verified. For developers who want to run the model themselves, Moonshot is sticking to the original timeline: the complete model weights are expected to be released by July 27.

It will be crucial whether Moonshot can expand its server capacity quickly enough to deliver on the promise of a swift reopening. If the registration stop drags on for weeks, the open-weight release on July 27 could become the real safety valve: anyone with their own compute would no longer depend on Moonshot’s infrastructure at all.

Frequently asked questions

Can existing customers continue to use Kimi K3?

Yes. According to Moonshot, access for already running subscriptions remains unchanged, only new registrations are affected.

When will Moonshot reopen registration?

The company has not yet provided a fixed date. New users are to be gradually unlocked in groups as additional server capacity becomes available.

What does Kimi K3 cost once registration is possible again?

The rates range according to the official pricing page from a free tier to 159 dollars per month with annual billing. Those using the API pay separately starting at three dollars per million input tokens.

Is Kimi K3 officially available in Germany or the EU?

There is no explicit confirmation for Germany or the EU. Moonshot does not mention any regional restrictions, and access is via website, app, and API.

Why did the shortage occur at all?

According to the company, usage increased so significantly within 48 hours of launch that the existing graphics processors reached their capacity limit.


← Back to the blog