Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Unlock Editor’s Digest for free
Roula Khalaf, editor of the FT, picks her favorite stories in this weekly newsletter.
China will win the artificial intelligence race, says Jensen Huang. At first glance, it’s easy to assume that Nvidia’s billionaire founder is just talking his book. Nvidia stands to benefit most from any narrative that encourages the U.S. to increase its investment in AI or ease regulatory restrictions on its development, thereby boosting demand for Nvidia chips. But is he right?
Not long ago, about a fifth NvidiaData center revenue comes from China. Its success depends on a steady stream of orders for its chips from governments, cloud providers and AI research labs around the world. The fear of China’s advance in AI is increasing this demand.
Despite it, Huang’s warning could contain some truth. AI development has begun to shift from being primarily constrained by the availability of high-end chips to being constrained by power supply.
Accordingly, a GPT-4 model can consume up to 463,269 megawatt hours of electricity per year Research by scientists from the University of Rhode Island, the University of Tunis and Providence College. That’s more than the annual energy consumption of more than 35,000 U.S. households. This demand reflects the growing share of AI workloads in data center power consumption. According to Rystad Energy, global electricity consumption by data centers will more than double by 2030 and reach about 1,800 terawatt-hours by 2040, enough to power 150 million U.S. homes for a year.
As a result, the price and availability of electricity will increasingly determine the pace of AI progress. China has a lead here. A record amount of renewable energy capacity was added last year, mostly through new solar and wind installations. Solar power alone grew by about 277 gigawatts, while wind power contributed about 80 GW, bringing total new renewable capacity to more than 356 GW, far exceeding the total capacity in the United States.
This increase in renewable energy is part of a larger plan. Beijing has linked industrial policy to its efforts to strengthen the national power grid, develop large solar projects in Inner Mongolia, expand hydroelectric power in Sichuan and build high-voltage transmission lines to carry cheaper electricity from the interior to demand centers on the coast.
Municipalities also grant preferential electricity tariffs to companies such as Alibaba, Tencent and ByteDance Promoting local AI computing. These subsidies help offset the lower efficiency of Huawei’s domestic chips, allowing China to train AI models at a lower overall cost.
Meanwhile, in the US, wholesale electricity costs have risen, with prices now at just 10% 267 percent higher than five years ago in areas near data centers. But investment in many types of renewable projects, including large wind and solar projects, fell in the U.S. in the first half of the year, reflecting policy changes and regulatory uncertainty. The White House also detailed an executive order ending subsidies for wind and solar energy.
Some argue that China’s energy advantage cannot fully offset its backlog in chips and models. In fact, Nvidia’s H100 and Blackwell GPUs remain ahead of Chinese alternatives like Huawei’s Ascend 910B in terms of memory bandwidth and performance.
This imbalance would have been crucial in the hardware-dominated period of technological competition, when access to advanced chips for computers and smartphones determined who led entire industries. The USA, for example, slowed the rise of Huawei by restricting the supply of high-end chips from 2019.
But the difference today is that energy is now growing faster than transistors: Chip performance gains have slowed to single digits, while China’s renewable electricity generation continues to grow by double digits each year. Falling electricity costs are increasing the amount of computing power that can be purchased for the same budget, and expanding network capacity is making it possible to train models more frequently and for longer periods of time.
The race to master AI is new, but part of a centuries-old history. Throughout history, every technological superpower has emerged based on cheap energy. Cheap, abundant coal fueled Britain’s Industrial Revolution. In the United States, oil and hydropower increased its dominance in manufacturing and military technology in the 20th century.
The battle to control AI is often portrayed as a competition over chips and the controls that govern them. But the power will belong to those who can keep the AI models running.