登録して招待リンクを共有すると、動画再生報酬と紹介報酬を獲得できます。

Schulz Duggan
@Schulz_Research
AI and chip news as it happens, explained simply. Research on optics and memory from Asia and US filings. Weekly industry explainers. Econ & finance background.
参加 January 2026
87 フォロー中    3.3K ファン
$INTC China's daily AI token volume jumped roughly 5,000x in 30 months from 0.1 trillion in early 2024 to ~500 trillion by June 2026 per China's National Data Administration and CAICT, yet greenfield data centers still take 24+ months of civil work before a single rack powers on. $INTC China and local grid partners are compressing that deployment cycle by ~88% to 9-13 weeks by parking containerized compute pods right next to western solar and wind farms. At Shanghai's 9th CIIE (Nov 5-10), Intel China Chairman Wang Zhicong debuted the "Green Power to Green Token" pod architecture slated for initial commercial rollouts in 1H 2027. Long-distance UHV lines lose meaningful energy hauling western power thousands of kilometers to coastal cities. Converting solar and wind generation into inference tokens on site lets operators ship lightweight data packets over fiber instead. Three engineering numbers stand out from our cross-check of the CIIE disclosures: 1. 800V HVDC + Solid-State Transformers: Co-developed with Trina Solar (SHSE: 688599), the prefabricated power bay hits 98.5% conversion efficiency and targets PUE below 1.15 while stripping out bulky 48V copper busbars. At Qinghai's Sanjiangyuan microgrid pilot, co-located solar-storage compute cut power bills by ~50% and total OpEx by 15%. 2. 8+ Hours Off-Grid Buffering: Zhonghai Energy Storage pairs the compute container with data-center-grade iron-chromium flow batteries, bridging wind and solar intermittency for 24/7 inference continuity already proven off-grid in Lankao. 3. 12 Racks Per Container, Scalable to 100 Pods: Each factory-built unit (8-12 weeks customization plus as fast as 1 week on-site setup) draws 100 kW to several MW and runs 1T+ parameter open-source models locally, or links up to 100 pods into a city-scale token factory. We think one business-model detail matters most for equity holders: Wang confirmed Intel has no plans to sell Intel-branded containers to end users. After spending three quarters standardizing liquid-cooling and power interfaces across dozens of Chinese suppliers, Intel is using turnkey power-compute reference designs to anchor Xeon 6+ and 18A silicon inside China's inference buildout, where CAICT data shows inference overtook training demand in 2026. As grid queues stretch past two years in the US and Europe, how long before modular off-grid token pods migrate from western China's curtailed renewables to global power bottlenecks?
もっと見る