The significant energy requirements of AI accelerators are a key focus for companies and governments striving to stay at the forefront of technology. China is actively promoting the use of domestically produced GPUs to address this challenge.
Provinces like Gansu, Guzhou, and Inner Mongolia are providing significant 50% discounts on industrial energy for AI data centers utilizing domestic accelerators, along with financial incentives to attract companies to establish operations in the region, as reported by the Financial Times.
These areas provide energy that is 30% less expensive than China’s coastal regions, with companies benefiting from a discounted price of just 0.4 yuan per kWh.
China’s inland regions seem to be more equipped to handle the significant energy demands of AI industrial processing, despite the higher energy consumption associated with using Chinese GPUs instead of NVIDIA models.
China is not worried about the efficiency of foreign chips.
China has chosen to prohibit the purchase of NVIDIA H20 accelerator exports, leading to a significant rise in energy consumption for AI companies.
NVIDIA hardware is more energy efficient than the current Chinese accelerators available. The H20s were designed based on Grace Hopper, while the new Blackwells are even more efficient but are officially banned in the US.
Read further:
China appears to make up for its hardware limitations with a wide range of products and sufficient energy to support them. These limitations may not lag far behind those of its international competitors for much longer.
Via: Tom’s Hardware
