{"slug": "kimi-k3-is-so-popular-moonshot-ran-out-of-gpus-and-paused-new-sign-ups", "title": "Kimi K3 is so popular Moonshot ran out of GPUs and paused new sign-ups", "summary": "Moonshot AI paused new sign-ups for its Kimi K3 model after demand pushed its GPU cluster to capacity. The Chinese lab said on X that demand over 48 hours \"pushed close to the limits of our current capacity,\" adding \"Our GPUs are feeling it.\" Existing subscribers are unaffected, and new slots will reopen in controlled batches. The 2.8-trillion-parameter open-weight model, which topped a major coding leaderboard within a day of its July 16 launch, will not have its weights released until July 27, forcing all demand onto Moonshot's servers until then.", "body_md": "Moonshot AI has a problem most startups would envy. Its new Kimi K3 model proved so popular that it ran out of computing power to serve everyone. So the Chinese lab has hit pause on new sign-ups.\n\nOver the past 48 hours, demand “pushed close to the limits of our current capacity,” the company [said on X](https://x.com/kimi_moonshot/status/2078855608565207130). Put more plainly: “Our GPUs are feeling it.”\n\n## What Moonshot changed\n\nNew subscriptions are paused for now. Existing subscribers are not affected, and Moonshot says it is pointing all its spare compute at them to keep the service stable. New slots will reopen in stages, in controlled batches rather than all at once.\n\nThe company also split its plans in two. One tier, a Kimi Membership, covers the web, app, and Work products. A separate Kimi Code Membership handles programming. Moonshot says the split lets it spread computing power more evenly across different kinds of users.\n\nThat distinction matters. Coding and agentic tasks are heavy work for a model. They involve long chains of reasoning and large chunks of context, so each user can eat far more compute than a casual chat. Ring-fencing the coders is a way to stop them draining the pool.\n\nThe suspension hit consumer plans with immediate effect, [TechNode reported](https://technode.com/2026/07/20/kimi-k3-overwhelms-capacity-just-days-after-launch-suspends-new-consumer-subscriptions/). User requests in the two days after launch ran far past Moonshot’s own projections, it said, taking the compute cluster close to full.\n\n## The open-weight paradox\n\nHere is the twist. [Kimi K3](https://thenextweb.com/news/moonshot-kimi-k3-largest-open-model) is billed as the world’s largest open-weight model, at 2.8 trillion parameters. Open weights are supposed to ease the strain on any single provider, because anyone can download the model and run it themselves.\n\nFor now, they cannot. Moonshot will not release the K3 weights until 27 July. Until then, the only way to use the model is through Moonshot’s own apps and API. All that demand lands on one company’s servers.\n\nAnd this is not a model you run on a laptop. Kimi K3 needs a multi-GPU rig, on the order of eight H100 or H200 chips just to serve it, [by one estimate](https://cryptobriefing.com/kimi-k3-gpu-capacity-crunch-moonshot-ai/). It also carries a one-million-token context window and handles text and images, features that make each session hungrier still. That makes the hosted service the easy option, and the bottleneck. “So much for the idea that open source cuts computing needs,” The Decoder noted.\n\nThe 27 July weight release is the pressure valve. Once the files are public, big customers and cloud providers can host K3 themselves and skip Moonshot’s queue. Casual users, though, will still lean on the apps, so the crunch may ease rather than vanish.\n\n## Turning a viral moment into money\n\nThe timing is not lost on anyone. Kimi K3 launched on 16 July and [topped a major coding leaderboard](https://thenextweb.com/news/kimi-k3-reactions-sacks-khosla-marcus-ai-race) within a day, drawing alarm from Washington and Wall Street alike.\n\nMoonshot is now converting that attention into paying customers. Annual recurring revenue hit $300 million in June, up from $200 million in April. The rush also arrives as the firm prepares a Hong Kong listing that could value it at [more than $30 billion](https://thenextweb.com/news/moonshot-ai-ipo-six-months-30-billion-kimi-k3). A model too popular to sell is an unusual pitch to investors, but a persuasive one.\n\nMoonshot is racing to add capacity. Once the weights are public, larger customers will be able to run K3 on their own hardware and skip the queue entirely. The [scramble for GPUs](https://thenextweb.com/news/ai-memory-crunch-boom-bust-2028), though, is not going away.\n\n## Everyone is short of compute\n\nMoonshot is not alone in the squeeze. In the same week, Anthropic cut usage limits on its Fable 5 model for subscribers, citing demand it called hard to manage. Rivals are circling the moment, too. Alibaba rushed out an open-weight Qwen model at a steep discount, aimed squarely at Kimi K3.\n\nThe pattern points somewhere bigger. A wave of capable [Chinese models](https://thenextweb.com/news/a-chinese-ai-startup-is-about-to-hit-1bn-in-sales-while-giving-its-best-models-away-for-free) is arriving at prices that undercut US labs. The barrier to using them is no longer capability. It is whether anyone has the chips to keep them running.\n\n## Get the TNW newsletter\n\nGet the most important tech news in your inbox each week.", "url": "https://wpnews.pro/news/kimi-k3-is-so-popular-moonshot-ran-out-of-gpus-and-paused-new-sign-ups", "canonical_source": "https://thenextweb.com/news/moonshot-kimi-k3-subscriptions-paused-gpu-capacity", "published_at": "2026-07-20 10:44:15+00:00", "updated_at": "2026-07-20 13:27:01.035740+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-infrastructure", "ai-startups", "ai-products"], "entities": ["Moonshot AI", "Kimi K3", "Anthropic", "Fable 5", "TechNode", "The Decoder", "Cryptobriefing", "H100"], "alternates": {"html": "https://wpnews.pro/news/kimi-k3-is-so-popular-moonshot-ran-out-of-gpus-and-paused-new-sign-ups", "markdown": "https://wpnews.pro/news/kimi-k3-is-so-popular-moonshot-ran-out-of-gpus-and-paused-new-sign-ups.md", "text": "https://wpnews.pro/news/kimi-k3-is-so-popular-moonshot-ran-out-of-gpus-and-paused-new-sign-ups.txt", "jsonld": "https://wpnews.pro/news/kimi-k3-is-so-popular-moonshot-ran-out-of-gpus-and-paused-new-sign-ups.jsonld"}}