Power Demand of AI to Skyrocket by 2030 A Concern for the Industry
TapTechNews July 9th news, Wells Fargo predicts that the power demand for AI this year is 8 terawatt-hours (TWh), and it will surge to 652 TWh by 2030, an increase of 8050%.
Wells Fargo said that the power consumption of AI is mainly used in training AI models. It will reach 40 TWh in 2026 and 402 TWh by 2030. In addition, the power consumption of AI inference will witness rapid growth in the late 2020s.
If we simply look at this number, there may be no intuitive feeling. Then here is another set of data attached: the total power consumption in China for the whole year of 2023 is 9224.1 terawatt-hours (TWh), the total power consumption in Shanghai for the whole year is 184.9 TWh, and the total power consumption in Shenzhen for the whole year is 112.8 TWh.
The AI industry is facing a serious challenge of power shortage. Taking Nvidia's H100 AI GPU as an example, the single-unit power consumption of the H100 in the SXM form factor is as high as 700 watts, and Nvidia's next-generation B200 AI GPU is even more terrifying, with each chip's power consumption reaching 1200 watts.
AMD is the same. The peak power consumption of the previous-generation Instinct MI250 AI accelerator is 560 watts, and the peak power consumption of the company's newly launched MI300X AI accelerator is 750 watts, an increase of 50% compared to the previous generation.
Intel is not to be outdone. The peak power consumption of its new Gaudi2 AI accelerator is 600 watts, and the peak power consumption of the new Gaudi3 AI accelerator is 900 watts (another 50% increase).
The power consumption of Intel's next-generation FalconShore hybrid AI processor is even higher, refreshing the industry's highest record, and the peak power consumption of each chip reaches an astonishing 1500 watts.
This is only a separate calculation of the GPU part. If the CPU and other additional components are added, the power consumption is even more astonishing. Taking Nvidia's GB200 (2 B200 GPUs and 1 Grace CPU) as an example, the power consumption reaches an astonishing 2700 watts.