China's First 100MW Grid-Flex Data Center Test: AI Compute Makes Room for the Grid
创业邦 · wechat · 2026-09-22
Chuangyebang's research center released a report on compute-power coordination (算电协同) in China. Key facts: in August 2026, 11 data centers in Jibei shed over 100MW within 10 minutes — China's first 100MW-scale test of data centers acting as flexible grid resources; Shanghai ran a similar test days earlier (16 centers, 97.8MW). Global data center power use will hit 950B kWh by 2030; Chinese AI compute hit 2185 EFLOPS while wind/solar curtailment tripled in two years.
Key claims: US hyperscalers are betting $725B of 2026 capex on nuclear; China bets on green-power dispatch with electricity at 1/3–1/4 of US prices, making power the key controllable cost lever (power is 10–25% of full token cost). AIDC architectures are being rewritten (800V DC, solid-state transformers at 98.5% efficiency). From 2024–2026, 307 funding rounds worth ¥65.1B went to the sector, but 97% chased "power" assets while the compute-scheduling "interface layer" got only 9% — the report argues the surest opportunity lies in that interface layer (EMS, liquid cooling, VPP, dispatch trading platforms).
More from Infra
- SemiAnalysis: Engram DRAM offloading delivers up to 50% better perf, upstreamed to vLLM — bookwormengr · 2026-09-23
- liuliu warns: claimed 4x-10x speedups over MLX or llama.cpp on Apple hardware are noise — teortaxesTex · 2026-09-23
- DeepSeek DSec cluster BoM estimated at ≤$20M; $1B could buy 50 clusters and 19M concurrent sandboxes — teortaxesTex · 2026-09-23
- B200 Rental Prices Hit All-Time High at $7.88 per GPU-Hour — sudoraohacker · 2026-09-23
- ASML shipped zero new systems to European customers in Q2; Korea took 43% — FinanceYF5 · 2026-09-23
- Custom fused sampler kernel adds 15% tok/s for DiffusionGemma on vLLM — imjustnewatai · 2026-09-23