The 24 MW facility combines seawater cooling with direct offshore wind supply, sidestepping the grid revenue problem entirely at demonstration scale
The Breaking Point
The gap is stark. China’s 2026 government work report set a target of 80% renewable supply to data centers by 2030, against a baseline of 11% in 2023. By 2025, coal’s share of Chinese data center electricity stood at approximately 70%, with renewables at roughly 20% and nuclear at 10%. In three years, the policy target moved aggressively upward while the physical mix barely shifted—and that is the break grid operators are now refusing to quietly absorb.
The scale of what is being asked to change makes the resistance comprehensible. China’s data center power capacity stood at approximately 32 GW at the end of 2025 and is projected to reach roughly 60 GW by 2030. The IEA projects that electricity demand from Chinese data centers will increase by approximately 175 TWh by 2030—a rise of around 170% from 2024 levels—placing China alongside the United States as one of the two primary drivers of global data center power growth. Asking that growth surge to run on a fundamentally different fuel mix within four years, without a mechanism to reconcile grid economics, is where the policy hit its limit.
Where the Shift Accelerated
The friction surfaced publicly at a recent Beijing industry conference, where Pei Shanpeng of State Power Investment Corporation articulated the core problem: GPU economics eliminate the demand flexibility that grid planners need. Once expensive compute hardware is purchased, operators are economically incentivized to run it at maximum intensity—creating a load profile closer to a base-metal smelter than an industrial user that can be curtailed when renewable generation dips.
Grid operators have a second concern that is financial rather than technical. Direct “green power” connections between renewable generators and data centers route electricity outside the main grid, reducing the revenue base that funds transmission and distribution infrastructure. Wang Zelin of State Grid Jibei Electric Power Research Institute has noted that even a 15% adjustable load from the data center sector could materially ease capacity expansion pressure over the next three to five years—but achieving even that modest threshold requires data center operators to accept curtailment events their GPU economics actively resist.
China is also testing alternatives at the edge of what is technically feasible. In 2026, HiCloud Technology and state-owned China Communications Construction launched what they describe as the world’s first offshore wind-powered underwater data center demonstration project in Shanghai Lingang. The 24 MW facility combines seawater cooling with direct offshore wind supply, sidestepping the grid revenue problem entirely at demonstration scale. Whether it can be replicated at the GW scale the national target requires is an entirely different question.
Where This Hits Global Heads of Data Center Energy
The China dynamic carries operational signal for any senior energy leader managing a multi-region portfolio, even one with no China footprint.
The core mechanism is the same everywhere: AI workloads impose near-constant, high-density power draws that are structurally hostile to the intermittency of wind and solar. A PPA structure that works well for a mixed-use enterprise campus—where some loads flex with generation availability—fails when the load is a GPU cluster running inference around the clock. That is not a Chinese problem; it is an AI infrastructure problem. Operators who have committed to high renewable percentages in their sustainability targets and are now deploying AI-optimized capacity at scale face the same arithmetic their Chinese counterparts are confronting, with the added exposure that their commitments are often public and contractually binding.
There is also a procurement precedent risk. China’s approach of mandating direct co-location between data centers and renewables in western provinces—prioritizing construction in resource-rich regions—mirrors strategies being advocated in other jurisdictions. If those policies produce the same grid operator resistance seen in China, the timeline assumptions embedded in current interconnection queue strategies and PPA terms may prove optimistic. Grid operators across multiple markets are watching AI-driven load growth with increasing concern about reliability and infrastructure cost recovery. The resistance in China is a leading indicator, not an isolated event.
What Could Still Change the Read
Several variables could alter the trajectory in either direction. First, the 80% target itself may be revised or reinterpreted. China’s work report language emphasizes integration between computing and power networks without specifying enforcement mechanisms, leaving room for provincial-level flexibility that could dilute the national headline figure.
Second, the demand scale is genuinely uncertain. More aggressive scenarios project Chinese data center power demand reaching 400 to 600 TWh by 2030—a range that nearly triples the IEA’s central projection. If demand lands at the upper end, the renewable supply problem becomes geometrically harder; if it lands lower due to efficiency gains or AI workload consolidation, the gap may be narrower than the current mix implies.
Third, IEA analysis referenced in the source suggests that between 2024 and 2030, coal is expected to remain the largest source of additional electricity for Chinese data centers, contributing approximately 90 TWh—roughly matching the projected renewable addition over the same period. The clean inflection point for Chinese data center supply is placed after 2030, when nuclear and renewables are projected to push coal’s share toward decline. That post-2030 timetable is not confirmed; it depends on grid investment, interconnection build-out, and whether demand growth is matched by clean supply additions at the required pace.
The Question This Leaves Your Team
If your AI-optimized capacity is contractually committed to high renewable matching percentages, how confident are you that your current PPA and REC structure can actually deliver those percentages against the load profile your GPU clusters will generate—and what is the grid operator’s position on flexibility in the markets where you are most exposed?
Sources
- Energynewsbeat — Chinese Grid Operators Resist Plans To Boost Renewables To Power AI (Link)
