Moonshot AI's Kimi K3 with 2.8T parameters halts subscriptions amid demand surge
New open-source giant Kimi K3 sparks AI cost panic and chip sanctions accusations...
Chinese startup Moonshot AI released its Kimi K3 open-source model on July 27, boasting 2.8 trillion parameters and performance reportedly near that of leading US models like GPT-4. The model's efficiency attracted massive user interest, quickly overwhelming Moonshot's computing resources. As a result, the company halted new consumer subscriptions, citing demand far exceeding available GPU capacity. The incident underscores the immense compute requirements of large-scale AI models and the operational challenges for emerging players.
Beyond operational hiccups, Kimi K3 reignited investor fears about AI infrastructure costs, contributing to a broader sell-off in tech stocks. The White House Office of Science and Technology Policy Director Michael Kratsios further escalated tensions by accusing Moonshot AI of training the model on restricted Nvidia chips (likely A100 or H100 variants subject to US export controls) and engaging in model distillation from US-developed AI. These allegations highlight ongoing geopolitical friction over AI technology and chip access, casting a shadow over open-source AI collaboration.
- Kimi K3 has 2.8 trillion parameters and matches US model performance.
- Moonshot AI paused subscriptions due to huge demand and compute shortages.
- White House accused Moonshot AI of using restricted Nvidia chips and model distillation.
Why It Matters
Raises concerns on AI chip sanctions, infrastructure costs, and open-source security.