China is sweetening the deal for homegrown artificial intelligence by cutting electricity costs for data centers that choose domestic chips from companies like Huawei and Cambricon. With access to foreign accelerators restricted and energy efficiency lagging, these incentives are designed to keep cloud providers committed to China’s own AI ecosystem.
A recent report indicates that many Chinese AI processors are 30% to 50% less power efficient than Nvidia’s H20, pushing power consumption—and utility bills—significantly higher. To offset that gap, local governments in data center hubs such as Gansu, Guizhou, and Inner Mongolia are offering subsidies that can slash electricity bills by up to 50%, but only when facilities run domestic chips. Data centers using foreign hardware do not qualify for these discounts.
This policy effectively realigns total cost of ownership for AI infrastructure. Even if local chips draw more power, deep energy discounts can make the economics work, encouraging large-scale deployments of domestic accelerators. Vendors are also leaning on scale to close the performance gap: Huawei and others are pursuing massive clusters that string together thousands of chips or raise power envelopes to rival top-tier systems, though such configurations can exceed the energy draw of platforms like Blackwell-class servers.
China’s grid capacity and the ongoing build-out of inland data center regions provide a foundation for this strategy. The mix of cheaper land, abundant power, and targeted subsidies makes these provinces attractive for AI training and inference at scale, helping keep workloads and investment onshore.
Long term, the race is about more than electricity. Catching up on performance-per-watt will require progress across the supply chain, from advanced semiconductor nodes and high-bandwidth memory to cutting-edge packaging. Analysts expect it will take years for domestic AI chips to consistently match global leaders on efficiency and raw throughput.
In the meantime, expect a rapid acceleration in deployments built around Chinese accelerators as operators recalibrate around energy subsidies, cluster utilization, and software optimization. The near-term trade-off is clear: higher power draw offset by aggressive electricity incentives, with the broader goal of building a self-sustaining, globally competitive AI hardware ecosystem.
