Chinese AI Labs Run Short of Compute as Big Tech Locks Up Data-Center Power

Chinese AI Labs Run Short of Compute as Big Tech Locks Up Data-Center Power Bloomberg says Chinese AI developers are short of compute as hyperscalers hoard data-center contracts, turning the race for capacity into a contest for coal- and gas-fired power. Chinese AI labs are running short of computing power as the country's largest technology firms lock up data-center contracts ahead of them, Bloomberg reported. The shortage moves the contest for AI capacity from chips to the electricity and grid capacity needed to run them. Data centers already account for more than 1% of global electricity use, according to the IEA, and the contract that secures compute also claims the power behind it¹. The footprint is scaling fast. The IEA estimates the world's data centers consumed about 415 terawatt-hours of electricity in 2024 and will reach roughly 945 terawatt-hours by 2030, while its 2026 update reckons a single advanced server rack could draw peak power equivalent to 65 households by 2027⁵. The shortage Bloomberg describes points at a domestic bottleneck rather than an external one. If the largest firms have booked the contracts first, the labs left behind must either bid up scarce capacity or wait for it to be built. Chip supply tightens the squeeze: manufacturers of computer components in four EU countries cut their investment in China by an average of 46% between 2021 and 2023, the Economist reported⁴. The American build-out shows where this leads, and why it is a fuel-demand story. Data centers used about 4.6% of total US electricity in 2024, a share government estimates suggest could nearly triple by 2028, while some analysts see nationwide electricity use rising as much as 20% over the next decade². Natural gas supplied more than 40% of the electricity powering US data centers in 2024, and coal supplied 30% of data-center power globally, the IEA said². The clean-energy pledges are bending under the load. Google, which once aimed to run entirely on clean power by 2030, called that goal a moonshot as of 2026-05-19, with its emissions up nearly 50%; Amazon's rose 33%, Microsoft's more than 23% and Meta's more than 60%, Fortune reported². The mix is not flipping faster because of dispatchability. BloombergNEF expects solar to become the largest single source of power within a decade, yet still projects fossil fuels will supply 51% of the incremental generation built for data centers by 2050, because coal and gas can run around the clock³. Storage is the swing factor. Google has committed $1 billion of 100-hour batteries from Form Energy to a recent data-center project, and BloombergNEF expects battery prices to fall another 30% by 2035, undercutting coal and gas³. Whether long-duration storage scales fast enough to displace the thermal baseload that AI now calls on will decide how much coal and gas demand the build-out locks in³. For China the arithmetic is sharper, because its grid still leans heavily on coal, so marginal load from any hyperscaler build-out is met first by thermal generation rather than by new renewables². AI already consumes 10-20% of US data-center energy and that share is set to rise, a curve Chinese demand is tracking¹. The signal worth following is how much of the next wave of data-center load is served by coal and gas versus solar paired with the long-duration storage now being funded³.