Envision Launches World's Largest AI Compute Super-Block for Million-GPU Clusters
机器之心 · wechat · 2026-08-10
Envision Group has officially launched the "Envision Star River Base" in Ulanqab, Inner Mongolia, unveiling the world's largest AI compute super-block spanning over 120,000 square meters. With a total planned capacity of 2GW, the base aims to solve the physical infrastructure bottlenecks of the million-GPU era and can support the simultaneous operation of one million cards.
Facing the power and cooling challenges of ultra-large-scale clusters, the facility reimagines the AI power system. By directly supplying electricity from dedicated wind and solar farms (over 80% green energy), combined with a meteorological large model for forecasting and an energy large model for scheduling, alongside multi-time-scale energy storage and an 800V DC architecture, it transforms fluctuating green electricity into stable compute power. Envision's entry into the AIDC market from the energy side allows clients to simply bring their GPUs for deployment, signaling a shift towards AI data centers designed as system engineering projects where energy and compute are co-designed.
More from Infra
- NVIDIA Nemotron 3.5 Lightning Hits DeepInfra with 1M Token Context — gharik · 2026-08-11
- China's DRAM Leader CXMT Joins MSCI China Index, Set to Lure Massive Fund Inflows — pstAsiatech · 2026-08-11
- OpenAI Pledges to Support New Power Generation and Grid Infrastructure in Texas — pstAsiatech · 2026-08-11
- MiniMax H3 Video Generation Benchmark: RTX 5090 Takes Under 5 Minutes — gabxav · 2026-08-11
- Open-Source Semantic LLM Cache PromptCache Cuts Costs by 80% with Sub-Millisecond Latency — tom_doerr · 2026-08-11
- Nvidia's HBM Reduction Is an Emergency Measure, Not a Compute Breakthrough — JOBhakdi · 2026-08-11