China Subsidizes Electricity to Offset Poor Power Efficiency of AI Chips, Reducing Data Center Costs by Up to 50%

To address the power demands of its domestic AI chip sector, China is rolling out substantial electricity incentives aimed primarily at companies such as Huawei and Cambricon, known for their ‘power-hungry’ AI chips.

China’s Strategy: Electricity Incentives for Local AI Chip Manufacturers

Following a decisive rejection of NVIDIA’s AI chips—voted “100% no” —China’s AI industry is now heavily focused on local alternatives. Reports indicate that these domestic chips are criticized for their poor performance-to-power efficiency ratios. According to a Financial Times article, Chinese chips display a significant power efficiency shortfall of “30% to 50%”lower than NVIDIA’s H20 AI chip, leading to skyrocketing electricity costs for data centers.

Local governments in data-centre concentrated regions like Gansu, Guizhou, and Inner Mongolia are responding by providing subsidies that can reduce electricity bills by up to 50% for data centres utilizing domestic chips.

However, data centres powered by foreign chipsets such as those from NVIDIA do not qualify for such incentives.

In its efforts to compete with NVIDIA, companies like Huawei are reportedly focusing on boosting power consumption figures or consolidating numerous AI chips into large clusters, which can result in energy demands that significantly exceed the capabilities of existing systems like Blackwell. Nevertheless, China’s extensive grid infrastructure appears capable of supporting this domestic technology ecosystem. Given the enticing incentives from local administrations, interest in Huawei and other native chip makers is likely to grow.

Huawei AI chip timeline presentation on stage, listing models: Ascend 910C, 950PR, 950DT, 960, and 970 with release dates. — Huawei AI chip roadmap | Image Credits: Huawei

China seems committed to limiting the infiltration of American technology, which poses challenges for its local industry striving for high-performance computing capabilities. It may take several years for the local AI sector to reach a competitive level against NVIDIA’s offerings. Besides architecture development, China must also establish robust production lines for semiconductors, high bandwidth memory (HBM), and advanced packaging techniques.

Source & Images