Starcluster
summary
A planetary‑scale mesh of decentralized AI compute claims to let independent node keepers run data‑center grade racks powered by renewables and join a global cloud. Phase one launches 2025 with high‑density racks (one GB 2000 NVL 72 rack draws ~120 kW and holds 72 advanced AI chips), requiring about 8,300 racks to reach one gigawatt. Virtual machines in August and confidential computing in September appear as immediate tooling, with telco edge in 2026 and homes and schools beginning 2027 for low‑latency inference and large‑scale training.
AI-written summary · about
related talks
28mThe Future of AI Won’t Be Centralized
51mSustainable & Smart AI Factories in Europe
46mChallenges of Making Large AI Clusters Reliable
35mBuilding a 5-Exaflop Supercomputer for Meta-AI Research and...
25mOpening Keynote
06mThe AI Compute Crisis & Web3 Opportunity
05mOpen Source Infrastructure for the AI Native Era
31mThe Future of AI Isn't One-Size-Fits-All