Moonshot AI's Kimi K3 got so popular in its first 48 hours that the company had to stop taking new subscribers, and it hasn't said when signups reopen.
Moonshot AI just proved that even a hit AI model backed by a fresh multibillion dollar valuation can run into a wall it didn't build: not enough GPUs.
Kimi K3 subscriptions are paused. Moonshot AI said this week that demand for its new coding focused model overwhelmed the GPU capacity behind it within 48 hours of launch, forcing the company to stop signing up new subscribers so existing customers keep their service. "Kimi K3 has received far more love than we expected, and our GPUs are feeling it," the company wrote on X. Existing subscribers aren't affected. New ones are locked out, with no firm reopening date.
The model earned that attention. Kimi K3 is a 2.8 trillion parameter open weight model, the largest of its kind released to date, and Arena ranked it first in blind frontend coding tests, ahead of Anthropic's Claude Fable 5. The timing was brutal for chip stocks. The Financial Times reported that the Philadelphia Semiconductor Index had its worst week since April 2025, while Reuters reported that Apple overtook Nvidia on July 17 as investors reassessed the AI trade.
The Squeeze Nobody Flagged
Here's the detail nobody flagged in that first wave of coverage: the model that spooked chip investors can't actually serve everyone who wants it. Moonshot says it's expanding capacity in batches. It also plans to split its single membership in two: a Kimi web and app plan, and a separate Kimi Code Membership built specifically for coding workflows, so it can match compute to demand more precisely. That's a company redesigning its subscription structure around a shortage it didn't see coming.
That's not a marketing problem. It's a supply problem.
Moonshot's fundraising history makes the squeeze more striking. TechCrunch reported in May that the company had raised about $2 billion at a $20 billion valuation, up from $4.3 billion at the end of 2025. The South China Morning Post reported in June that Moonshot was seeking a round that could value it at as much as $30 billion. Money isn't the constraint here. Physical GPU allocation is, and Moonshot, founded by Yang Zhilin in 2023 and named after Pink Floyd's The Dark Side of the Moon, is learning that a viral product and a chip supply chain don't move on the same timeline. Yang, a Carnegie Mellon graduate, built a company now valued in the tens of billions. GPUs still don't care about valuation.
What It Means If You're Building on Kimi K3
If you picked Kimi K3's API for a coding tool because it's priced at $3 per million input tokens and $15 per million output tokens, or because it topped a benchmark, this week is a warning label. A model you depend on for production traffic just got rationed by its own maker, with existing users protected and everyone else told to wait for the next batch. That's a single point of failure dressed up as a growth story. Full open weights aren't due until July 27, which means you're building on infrastructure whose own operator can't yet guarantee capacity.
None of this makes Kimi K3 a bad bet. The benchmark results and the pricing genuinely matter, and Moonshot is doing the sensible thing by protecting paying customers instead of quietly degrading service for everyone. But OpenAI, Anthropic and Google spent years building redundant capacity for exactly this reason, so a viral moment doesn't turn into a waitlist. Moonshot is doing that same work in public, in real time, under a spotlight it didn't fully choose.
Moonshot hasn't given a date for when new signups reopen. For now, the only way in is to have gotten there before the rush.
Also read: Alibaba Rushes Out Qwen3.8 Two Days After Kimi K3 Rattled Chip Stocks • Moonshot AI Pauses New Kimi K3 Subscriptions After Demand Overwhelms Its GPUs • US Companies Now Run Nearly Half Their AI Traffic on Chinese Models
