Moonshot launched Kimi K3. Then demand shut down subscriptions in 48 hours.

This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).

Moonshot AI's Kimi K3 launch exposed a growing industry problem: inference capacity can't keep up with demand. Within 48 hours of release, the company froze new subscriptions after GPU resources were exhausted. At 2.8 trillion parameters, K3 is one of the largest open-weight models and performs competitively against GPT-5.6 Sol and Claude Fable 5 in coding benchmarks. The crunch is worsened for Chinese AI companies by US export controls limiting access to Nvidia's advanced chips, forcing reliance on older hardware and domestic alternatives. Analysts note that agentic workloads — which continuously generate and process tokens — consume far more GPU resources than simple chatbot queries, shifting the bottleneck from training compute to server memory. Token pricing pressure and massive data-center investment from Alibaba and ByteDance underscore how the AI infrastructure race is intensifying, with implications for developers who can no longer assume cheap, unlimited API access.

5m read timeFrom thenewstack.io
Post cover image
Table of contents
Inference demand outpaces supplyOpen weights, closed capacityChina’s chip constraints compound crunchToken economics under pressure
15 Impressions