Wood Mackenzie projects 774 TWh, or about 6% of China's grid, as AI capacity migrates west to renewable rich provinces.
By 2030, China's data centres are projected to draw 774 terawatt-hours of electricity, more than South Korea consumes today, according to a Wood Mackenzie forecast carried by Business Standard. The shift is already reshaping where AI capacity gets built: rather than clustering near coastal cities, new data centres are migrating to sparsely populated western provinces where land is cheap, grids are underused, and renewable generation is growing fastest.
A terawatt-hour is roughly the annual electricity use of about a million average US households, enough to power every home in a mid-sized American state for a year. At 774 TWh, Chinese data centres would draw about 6% of China's total projected electricity demand in 2030, roughly four times the current level, per the Wood Mackenzie forecast. The same 2030 figure would exceed South Korea's entire national electricity consumption today, a comparison the report uses to anchor the projection for non-specialist readers.
The siting pattern, which Wood Mackenzie calls "compute follows power," reverses the usual logic of putting data centres near users. Wanting Zhao, the firm's Asia Pacific power and renewables lead, said access to "reliable, cost-competitive and lower-carbon electricity will play a growing role in determining where and how new data-centre capacity is developed." Western provinces like Inner Mongolia, Gansu, and Ningxia already host most of China's new utility-scale solar and wind build, and grid planners have begun reserving transmission capacity for hyperscale tenants. The trade works in both directions: operators can lock in long-term power purchase agreements at lower cost, while the provinces monetize generation that would otherwise be curtailed.
The parallel to the United States is direct. American data-centre growth has become an election issue, with local opposition over water use, diesel backup generators, and rate increases feeding into Wall Street's reading of utility stocks. China has not seen the same zoning fights, but the same grid-strain question now shows up in provincial planning, where renewable curtailment, the practice of shutting off wind or solar because the grid cannot absorb the output, was already a problem in the west. Siting AI capacity there rather than in the load-heavy east lets operators draw on surplus generation, and lets planners keep the eastern seaboard's industrial demand served without expanding fossil capacity.
Wood Mackenzie also extends the line out to 2060. The firm projects data centres will reach about 17% of China's electricity consumption by 2060, a share that would put the sector on par with today's entire industrial base. That forecast depends on renewable buildout keeping pace, on transmission expansion to the western provinces, and on whether AI training continues to scale beyond current model architectures. The 2060 figure is softer than the 2030 anchor: it is a multi-decade projection from a single firm, not a target or policy commitment, and any rival forecaster could legitimately rebase it.
The 2030 number is what lands now: a single category of electricity demand, projected to outdraw a G20 economy's entire grid, on a five-year horizon.