Observed Signal · Nov 7, 2026 · Site Visit · Source: SemiAnalysis · Impact: 3/5 · Sentiment: Positive
China's Galaxy Datacenter: Fast, Cheap, Vertically Integrated
SemiAnalysis visited Galaxy, an 800MW AI datacenter campus in Inner Mongolia built by Envision Group. The first 120MW phase was delivered within a year at roughly a fifth of US costs. The campus features a 5MW pilot of solid-state transformers (SSTs) converting 10kV AC directly to 800V DC, eliminating intermediate transformers and UPS stages. Cooling is handled without central chillers, using a shared water loop for liquid- and air-cooled servers. Envision's vertical integration spans wind farms, battery storage, power electronics, and datacenter design. The project supplies over 60% of its power from wind via a source-grid-load-storage model. The article contrasts this with US Stargate Abilene and highlights trends in power architecture and energy access reshaping datacenter development.
Highlights China's advanced, cost-efficient datacenter construction methods, including SST-based 800V DC power and vertical integration, which could influence global AI infrastructure strategies.
Track OpenAI Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your Polaris7 workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Galaxy is an 800MW AI datacenter campus in Ulanqab, Inner Mongolia, owned by Envision Group.
- The first 120MW phase was delivered within a year at a fifth of US costs.
- A 5MW pilot uses solid-state transformers to convert 10kV AC to 800V DC, eliminating traditional UPS stages.
- The cooling design uses a shared water loop for liquid- and air-cooled servers without central chillers.
- Envision aims to supply over 60% of the first two phases' power from its own wind farms.
Connected Companies & Entities
“Stargate Abilene in Texas is a gigawatt-scale campus built for OpenAI.”
“A small pilot datacenter built with Tencent's modular T-Block units runs on Envision's off-grid system in Chifeng.”
“Crusoe began building the Abilene campus in mid-2024 and leased it to Oracle for OpenAI.”
“Crusoe leased the Abilene campus to Oracle, which runs it for OpenAI.”
“CATL is one of DeepSeek's largest outside investors.”
“Zhongheng Electric works with Alibaba on datacenter power systems.”
“Zhongheng Electric works with ByteDance on datacenter power systems.”
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
China's Ulanqab Data Centers: Who Pays for the Grid Wait
In Ulanqab, a Chinese AI data center hub, grid infrastructure is still under construction, causing delays between completed data centers and their power connection. Operators like GDS invest before service begins and collect no fees until billing starts. The Financial Times reports about 15 GW of committed capacity, while Goldman Sachs counted 12.5 GW. Data centers used 3.27 billion kWh in H1 2026, up 89.5% year-over-year. Power costs in Ulanqab are about Rmb 0.358 per kWh, much cheaper than Beijing's Rmb 0.80. However, substation construction can take three to five years, far longer than the 12 months to build a data center hall. The article examines the financial risk operators carry during these waits.
Behind-the-Meter Power Challenges for Datacenters
SemiAnalysis reports that behind-the-meter (BTM) power generation for AI datacenters has become mainstream, with 75GW of firm orders tracked, 20GW ordered in Q2 2026 alone. Major deals include Microsoft's 5GW with Chevron and Crusoe, Google's 930MW aeroderivative turbines and 900MW Bloom fuel cells, and OpenAI's 1.4GW campus with Jenbacher engines. The report details six key challenges: contracts & bankability, permitting, fuel supply, equipment procurement, workforce, and electrical physics. It highlights permitting delays (e.g., Oracle's Project Jupiter) and the rise of Energy-as-a-Service vendors like VoltaGrid. The analysis emphasizes the shift towards reciprocating engines and fuel cells, and the growing importance of balance-of-plant equipment.
DIGITIMES: AI Data Centers Shift to 800VDC Power
DIGITIMES Intelligence has released a report examining the shift toward 800VDC (volt direct current) power architecture in AI data centers. As AI computing scales, power delivery is becoming a critical system-level design issue, prompting data center operators, server makers, and semiconductor suppliers to rethink how electricity is delivered. The report, titled 'AI Data Center 800VDC Power Architecture Takes Shape; Future Hurdles Require Balancing Multiple Key Factors,' analyzes how 800VDC architecture could reshape power delivery and create strategic questions for the semiconductor and infrastructure ecosystem. Key areas of focus include conversion efficiency, power density, thermal management, and the role of power semiconductors like GaN and SiC. The transition is expected to intensify competition beyond device performance, involving collaboration with server OEMs and ODMs. This shift represents a broader redesign of power movement through AI infrastructure, potentially redistributing value across the supply chain.
Receive all future signals for OpenAI & competitors in your briefing
Never miss material M&A deals, leadership changes, account wins, or product rollouts. Polaris7 monitors 20,000+ primary sources 24/7, delivering verified telemetry directly into your executive briefing.
