1 day ago
53
Moonshot AI launched Kimi K3 on July 16, 2026, and within two days, the model was too popular for its own good. Demand exceeded the startup’s available GPU capacity so quickly that the company had to pause new subscriptions entirely, prioritizing service quality for users already on the platform.
What Kimi K3 actually is
Kimi K3 is a 2.8 trillion-parameter open-weight model built for serious workloads: long-horizon coding, complex reasoning, multimodal inputs, and agentic tasks that require the model to operate with minimal human hand-holding.
The context window runs up to 1 million tokens. In practical terms, that means the model can process the equivalent of several long novels worth of text in a single session, which makes it genuinely useful for enterprise-grade knowledge work rather than casual chatbot interactions.
Running Kimi K3 is not a laptop-friendly proposition. The model requires multi-GPU setups, with a minimum of around eight H100 or H200 equivalents just for inference.
Moonshot is responding by scaling its compute infrastructure and plans to bring new subscribers back in controlled batches rather than reopening the floodgates all at once. The full open weights for Kimi K3 are expected to be publicly available on July 27, 2026, which will let organizations run the model on their own hardware and sidestep the subscription bottleneck entirely.
The company behind the model
Moonshot AI is a Beijing-based startup that has moved fast by any standard. The company raised approximately $2 billion in a recent funding round, pushing its valuation above $20 billion.
Earlier in 2026, Moonshot reported over $200 million in annual recurring revenue.
The pricing structure for Kimi K3 reflects a competitive strategy clearly aimed at the enterprise market. Input tokens are priced at $3 per million, output tokens at $15 per million, with caching discounts available for operations that benefit from stored context.
What this means for the AI market and crypto-adjacent investors
For crypto and Web3 investors, the GPU capacity crunch at Moonshot carries a specific relevance. The same H100 and H200 chips that Kimi K3 requires for inference are the chips that AI-focused crypto projects, decentralized compute networks, and GPU-backed tokens have been betting on for the past two years. When a single model launch at a single company can stress-test GPU availability enough to pause subscriptions within 48 hours, it reinforces the core thesis behind decentralized compute: centralized infrastructure has real ceilings.
Moonshot’s decision to release open weights on July 27 adds another dimension. Open-weight releases tend to accelerate adoption, create derivative models, and put pressure on closed competitors to justify their pricing. The combination of a capable open model and a company with $2 billion in fresh capital is a competitive posture that OpenAI and Anthropic will need to account for in their own roadmap decisions.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·