Moonshot AI trains Kimi on 20,000 Nvidia GPUs

// 2h agoINFRASTRUCTURE

Moonshot AI trains Kimi on 20,000 Nvidia GPUs

Chinese artificial intelligence startup Moonshot AI trained its flagship Kimi model leveraging a massive 20,000 Nvidia GPU cluster provided by Alibaba Cloud. The arrangement underscores Alibaba's central role as a key cloud infrastructure provider supplying high-performance compute resources to frontier AI firms navigating global semiconductor supply chain constraints.

// ANALYSIS

Massive GPU compute clusters remain the foundational bottleneck and primary differentiator in training competitive large-scale AI models.

–Partnering with Alibaba for a 20,000-chip cluster highlights the growing reliance of top-tier AI startups on major cloud providers for hardware access.
–The scale of compute signals Moonshot AI's ambition to match top global labs in LLM performance and scaling efficiency.
–Cloud-hosted hardware allocations will increasingly dictate which AI developers can iterate rapidly on frontier models.

// TAGS

aimoonshotkimialibabanvidiagpusinfrastructurecloud

DISCOVERED

2h ago

2026-07-31

PUBLISHED

6h ago

2026-07-31

RELEVANCE

8/ 10

AUTHOR

gk1

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

OPEN SOURCE49m ago

Fish Audio open-sources sub-90ms Fish Speech

Fish Audio has open-sourced Fish Speech, a text-to-speech model supporting 83 languages with an impressive sub-90ms latency to first audio. Designed to deliver high-quality voice synthesis at one-sixth the price of established platforms like ElevenLabs, Fish Audio aims to demonstrate that the voice AI market is far from settled and that open, affordable models can challenge proprietary market leaders.

RESEARCH58m ago

Alibaba Unveils Qwen-UI-Agent Cross-Platform GUI Framework

Alibaba's MAI-UI team has released the technical report for Qwen-UI-Agent, a next-generation foundation GUI agent designed for autonomous real-world task execution across mobile, desktop, and web platforms. Trained on extensive multi-platform interaction datasets, the model targets fundamental challenges in visual grounding, action space unification, and long-horizon workflow execution.

RESEARCH1h ago

Explorative Modeling unlocks a third pretraining scaling axis alongside parameters and data by factoring the training loop to achieve end-to-end generative AI with dramatic compute and sample efficiency gains.

Generative modeling has traditionally struggled with true end-to-end training due to the challenge of handling multi-modal data distributions, relying instead on factored multi-step generation procedures that tend to blur modes. Explorative Modeling (XMs) introduces a new paradigm by factoring the training loop rather than the generation procedure. By evaluating K candidate predictions per step and training on the best candidate match, XMs force predictions to commit to distinct modes and establish exploration as a third pretraining scaling axis alongside model size and dataset volume across vision, video, and language domains.