Zhongwei already hosts ten large-scale data center parks with approximately 3.31 million standard server racks, and six of China’s top ten computing service providers have established operations there
Decision Focus
On 26 June 2026, China Telecom Ningxia Branch commenced operations at what is described as China’s first AI data center connected directly to renewable generation, located in Zhongwei, Ningxia Hui Autonomous Region. The facility is designed to run on 100% green electricity through a direct power supply arrangement, pairs that supply with liquid cooling to achieve a reported PUE of 1.15, and sits inside China’s state-directed Eastern Data, Western Computing initiative. The operational signal for global heads of data center energy: a government-backed operator has demonstrated a direct generation-to-facility model at AI-compute scale—one that structurally sidesteps the interconnection queue, basis risk, and certificate accounting that dominate Western renewable procurement. How that model compares to—and pressures—current PPA and co-location strategies is the question worth examining now.
90-Second Brief
In recent days, china Telecom Ningxia Branch has opened an AI data center in Zhongwei designed to operate entirely on directly connected green power. The facility, built to a wind-powered liquid-cooling standard, reports a PUE of 1.15. Zhongwei already hosts ten large-scale data center parks with approximately 3.31 million standard server racks, and six of China’s top ten computing service providers have established operations there. The project is part of China’s national Eastern Data, Western Computing initiative, which routes compute workloads to renewable-rich western regions, an approach that resolves, in one structure, the supply-matching problem Western operators are still navigating through layered PPA portfolios and multi-year interconnection timelines.
What Is Really Happening?
The Zhongwei facility is not an isolated efficiency project. It is an execution proof point for a state-coordinated model of co-locating AI compute directly with renewable generation. China’s Eastern Data, Western Computing initiative routes AI workloads to regions where wind and solar are abundant and land costs are low, designed to eliminate the basis risk and curtailment exposure that troubles Western virtual PPAs. Rather than procuring certificates after the fact or managing a portfolio of offtake agreements across grid zones, this model places generation infrastructure and computing infrastructure in a single value chain.
The direct connection bypasses typical grid intermediation. There is no ISO queue to manage, no locational marginal pricing exposure, and no RECs purchased separately to match load. Whether that simplicity is replicable outside a state-directed economy is an open question, but the efficiency outcome is documented: a reported PUE of 1.15 at AI-compute density, combined with a self-reported 100% green supply, collapses the traditional tension between reliability, sustainability, and cost optimization into a single operating model.
Zhongwei’s infrastructure density is also worth reading at scale. Ten large-scale data center parks concentrated in a single city—attracting six of China’s top ten computing service providers—was achieved by orienting city-level economic strategy around computing and clean energy together rather than retrofitting clean energy onto an existing load center. That sequencing is the part Western markets have not yet successfully replicated.
Why It Matters for Global Heads of Data Center Energy
The pressure point is not competitive threat—it is procurement model comparison. Western hyperscalers and colo operators are managing interconnection queues measured in years, basis risk on regional PPAs, and a renewable supply market tightening as AI buildout competes directly with industrial electrification. The Zhongwei model represents a structural answer to each of those constraints, even if that answer was enabled by state land allocation, state utility planning, and centrally mandated workload routing.
For operators currently evaluating generation co-location or direct-wire renewable arrangements with IPPs or developers, this facility establishes a visible efficiency reference. If your portfolio operates above PUE 1.3 on partial renewable coverage, the combined efficiency and emissions gap relative to this design is no longer a theoretical aspiration—there is a functioning reference installation at AI-compute scale.
The direct supply model also carries implications for how 24/7 carbon-free energy commitments get structured. Temporal matching and certificate attribution remain the dominant Western approaches, but the Zhongwei design sidesteps that accounting debate by making grid intermediation optional. That is architecturally significant, even if regulatory and grid structure differences mean the approach cannot be directly imported into a PJM or ERCOT context today.
Procurement teams tracking China’s renewable buildout as a downstream supply chain signal should also note the hardware dimension. The project is designed to increase demand for AI chips, liquid-cooling systems, and server infrastructure in a market already concentrating compute density. Equipment categories where Chinese and Western operators compete for the same supply—particularly liquid-cooling hardware—may face procurement lead time effects as this model scales.
Forward View
If the Eastern Data, Western Computing program extends the direct-wire model across additional facilities, it creates a running benchmark for what AI compute efficiency looks like when generation and load are designed together from the outset. Western operators pursuing co-location strategies with IPPs or SMR developers will face a credibility test against that benchmark—not immediately, but within the planning horizon of a decision made in 2026.
Western interconnection frameworks and FERC oversight mean direct replication is structurally unlikely in the near term. But the model may accelerate commercial pressure on ISOs and utilities to develop direct-wire tariff structures or streamlined co-location approvals. ERCOT’s competitive framework and PJM’s ongoing interconnection reform process are both arenas where that pressure could surface operationally within the next 12 to 24 months—worth flagging now for teams engaged in those regulatory proceedings.
The liquid-cooling and renewable integration pairing is also a forward supply chain signal. If wind-powered liquid-cooling becomes a recognized AI data center design standard in a major compute market, it will drive hardware supply chain prioritization at scale—potentially compressing lead times and pricing availability for the same equipment Western operators are specifying in new builds.
What Is Still Uncertain
The source is CGTN, a Chinese state-owned media outlet, and the reported figures—PUE of 1.15 and 100% green supply—are self-reported by the operating entity with no independent verification cited. The precise technical structure of the direct-wire arrangement, the carbon accounting methodology used, and whether the 100% green claim reflects real-time generation coincidence or a certificate offset approach are not specified. Those distinctions matter considerably for any operator benchmarking their own 24/7 CFE methodology against this facility.
The operational mechanics also leave gaps. A direct connection between variable wind generation and an always-on AI compute facility requires either storage buffering, grid backup, or workload curtailment when generation undersupplies load. None of those mechanisms are described in the source, which leaves the reliability architecture partially opaque. Without that detail, the 1.15 PUE claim cannot be fully evaluated against the operating conditions it assumes.
Finally, the Eastern Data, Western Computing initiative’s relevance to non-state operators depends on whether commercial and regulatory structures elsewhere move toward enabling similar co-location arrangements. The model’s durability as a genuine market precedent—rather than a state-directed deployment—remains unconfirmed.
One Question for Your Team
Your current renewable sourcing strategy—whether structured as a VPPA, a physical PPA with certificate matching, or a temporal 24/7 CFE program—was built around the grid architectures and regulatory frameworks you operate within today. The Zhongwei facility operationalizes an alternative that removes grid intermediation and certificate accounting from the equation entirely. The question worth bringing to your team is not whether to replicate it directly. It is: which of your planned or in-development facilities, in which jurisdictions, are geographically close enough to a wind or solar-rich site that a direct co-location structure would be technically feasible—and what specific regulatory or commercial barrier is currently the first obstacle to pursuing it?
Sources
- Cgtn — China launches first AI data center powered by 100% green electricity (Link)
