NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·2h ago
Cloud & Infrastructure
The New Stack·14d ago·5 min read
Moonshot launched Kimi K3. Then demand shut down subscriptions in 48 hours.
TL;DR
Moonshot AI launched its Kimi K3 model, but demand was so high that it had to halt new subscriptions within 48 hours due to GPU capacity limits.
✦ Why It Matters
Evaluate your infrastructure capacity to ensure it can handle unexpected demand spikes.
Key Takeaways
- Monitor demand trends for AI models to anticipate infrastructure needs.
- Consider batch subscription models to manage capacity effectively.
- Optimize software to reduce resource consumption during inference.
- Stay informed about regional chip supply constraints affecting AI development.
Full Summary
Kimi K3, a new AI model from Moonshot AI with 2.8 trillion parameters, quickly gained popularity, resulting in a subscription pause just two days post-launch. The surge in demand highlighted a significant gap between AI inference demand and available infrastructure, as companies struggle to keep up with the resource needs of complex coding tasks.
Moonshot's official statement acknowledged the unexpected demand and indicated plans to increase capacity and reopen subscriptions in batches. The model's open weights allow for widespread deployment, but hosting it incurs substantial costs, particularly for longer coding tasks that require more GPU resources.
This situation is exacerbated for Chinese AI developers like Moonshot, who face chip supply constraints due to US export controls, forcing them to rely on older technology and optimize software efficiency.