Is cheap energy the key to China gaining AI supremacy?

Leaders are hoping it will make up for its weakness in chips

Photograph: Getty Images

Listen to this story

Your browser does not support the <audio> element.

A rtificial intelligence is like a cake, says Jensen Huang, the boss of Nvidia, a chipmaker. ai applications, such as chatbots, are at the top. The next layer down is software, like the large language models (llm s) on which chatbots run. Then comes hardware, the semiconductors needed to train the models. This spring China’s ai firms are busy baking all of these layers. ByteDance, the company behind TikTok, has unveiled a slick new video-generation app. DeepSeek, a flashy startup, is due to release a powerful new llm. And Huawei, China’s tech champion, will unveil a new ai chip.

Though these firms keep China in the ai race with America, they are not pushing it into the lead. But there is another layer of Mr Huang’s cake that goes underneath all the others, and that is energy. Semiconductors require vast amounts of it to run the trillions of calculations behind the ai models. And China’s electrical grid has far more cheap power than the West. This disparity is known as the electron gap. Can China use it to achieve ai supremacy?

American companies seem spooked at the prospect. Sam Altman, the boss of OpenAI, has predicted the cost of ai will “eventually converge with the cost of energy”. In October his firm warned that China’s power advantage could “put us leadership [in ai] at risk”. The following month Mr Huang predicted that China “will win the ai race” for the same reason. In January Elon Musk, who owns xAI, another ai company, said that “based on current trends, China will far exceed the rest of the world in ai compute” because of its grid.

Chart: The Economist

ai companies are increasingly worried about access to energy. They are building ever bigger and more power-hungry data centres to support smarter models. Some are now at the gigawatt (gw) scale: equivalent to the power capacity of a nuclear-power station. Global demand to power such data centres could surge to 68 gw by 2027 and 327 gw by 2030, say researchers at rand, an American think-tank.

America’s ageing grid is already struggling to keep up. There is a huge backlog of data centres waiting to be connected. Firms are also wrestling with local opposition because data centres can push up power prices for residential users. Some are building off-grid generators. Others suggest ideas like building data centres in space rather than doing so in America. “Many ai projects are now constrained not by chip supply but by…whether enough reliable electricity can reach the building,” says one person at a semiconductor firm.

China has no such worries. Its power grid, the world’s largest, is still growing at a blistering pace thanks to massive state investment. It added over 500 gw of capacity just last year, to reach a total capacity of 3,800 gw, more than double that of America’s. Over the next five years China is set to add six times as much capacity as its rival. A bonanza of wind and solar projects is driving growth. And half of the world’s nuclear-power plants are also under construction in China, while the country is still building lots of coal-fired power. Chinese data centres can secure power for around three cents per kilowatt-hour, according to official figures, around half the rate many American ones pay. And because the government sets residential power prices separately, there is little risk of public opposition to power-hungry infrastructure.

Still, for all the panic about an electron gap, China is not yet exploiting it. A big reason is a shortage of chips. Since 2019 tightening American export restrictions have made it harder for Chinese firms to buy or build the advanced chips (those with feature sizes of seven nanometres [nm] or less) that power the latest models. Last year China’s tech firms were estimated to have spent $24bnonaiinfrastructure,suchasdatacentres;Americanonesspentover$350bn. Investments in data centres by China’s local governments have been mismanaged, leading to many getting built to low standards. Some reportedly have utilisation rates as low as 20%.

Pork and chips

As a result, China’s computing infrastructure is far weaker than its energy abundance could allow. Take Yanggao, a dusty spot in the northern province of Shanxi. Local officials claim it has become a “computing county”. A giant data centre has sprung up on the site of a former pig farm. It enjoys cheap power from wind farms, solar panels and a coal-fired power station; a cold climate to aid cooling; and a river to supply water. State-run media have paraded it as part of an “ ai wave” sweeping the province. But less than 0.1% of its chips are capable of the intense calculations needed to train ai s, according to a manager there.

There are signs that China will soon start leveraging its energy advantage. On March 5th Li Qiang, the prime minister, mentioned “hyperscale computing” (ie, giant data centres) for the first time in his annual state-of-the-nation address, promising to “launch new infrastructure projects co-ordinating computing capacity and electricity supply” this year. Chinese hyperscalers, meanwhile, are ramping up investment. Ken Liu, an analyst at ubs, a bank, expects China to build another 25 gw of ai data centres by 2029, having built just 5 gw over the past two years.

A build-out at that speed, notes Mr Liu, will depend on China manufacturing many more high-end chips domestically. Years of efforts to that end are bearing fruit. Huawei’s homegrown 7nm ai chips are still less powerful than American offerings, but they can close the performance gap when lots are stacked together. That consumes more energy, but it matters less when electricity is cheap. This year China’s leading foundry, Semiconductor Manufacturing International Corporation, which makes most of Huawei’s 7nm chips, plans to double its capacity for making them. In March, Reuters news agency reported that Hua Hong, another Chinese foundry, was also starting to make 7nm chips.

Officials are encouraging data centres in the western provinces that have plenty of wind, solar and hydropower (and cooler average temperatures). By 2028, China hopes to connect all these data centres into a single pool that can provide cheap computing resources nationwide. Such efforts should allow China’s power advantage to more than make up for its weakness in chips by the late 2020s, reckons Lin Boqiang, of the China Institute for Energy Policy Studies at Xiamen University. “All we have to do is keep building,” he says.

At the moment, China’s leaders are mainly focused on ai deployment: trying to push ai tools into the broader economy to make it more productive. Officials are especially excited about applying ai to the physical world through such things as self-driving vehicles, robots and smart factories. Abundant energy, and hence cheaper ai models, should help as companies will be more likely to actually use them.

For American tech bosses like Mr Altman the electron gap is more worrying in relation to the idea of artificial general intelligence (agi), an ai that can surpass the cognitive abilities of humans. An agi might suck up far more power than even today’s cutting-edge ai s. Might China be the one to eventually develop it? Until recently China’s leaders have seemed wary of the idea, seeing it as more of a risk than an opportunity. But in October Alibaba became the first big Chinese firm to announce it was pursuing agi. And in March China released its new five-year plan, for the period to 2030. It included a call to “explore development paths for agi ”. ■

---

논증 분석

유형: causal

핵심 주장

중국의 압도적인 저비용 전력 인프라(electron gap)는 칩 부족이라는 약점을 상쇄하여 2020년대 후반 AI 패권 경쟁에서 중국에 결정적 우위를 제공할 수 있다.

논리구조

1. 전제: Jensen Huang(Nvidia CEO)의 ‘AI 케이크’ 비유에 따르면, AI의 가장 근본적인 층은 애플리케이션·소프트웨어·하드웨어(반도체) 아래에 있는 에너지이며, 반도체는 AI 모델 구동에 막대한 전력을 필요로 한다.
2. 진단: 중국의 전력망은 세계 최대로 총 3,800GW 용량을 보유하고 있으며, 데이터센터 전력 단가가 약 3센트/kWh로 미국의 절반 수준에 불과해 구조적인 ‘electron gap(전자 격차)‘이 존재한다.
3. 논거: [Sam Altman], [Jensen Huang], [Elon Musk] 등 미국 빅테크 리더들이 중국의 전력 우위가 AI 리더십을 위협한다고 공개적으로 경고하며 미국 측이 이미 위기감을 느끼고 있음을 보여준다.
4. 진단: 미국은 노후화된 전력망, 데이터센터 연결 대기 적체, 주거용 전기요금 인상에 대한 시민 반발 등으로 AI 인프라 확장이 칩 공급이 아닌 전력 확보에 의해 제약받고 있다.
5. 반론: 현재 중국은 전력 우위를 충분히 활용하지 못하고 있다. 2019년 이후 강화된 미국의 수출 통제로 7nm 이하 첨단 칩 조달이 어려워졌고, 작년 AI 인프라 투자액이 240억 달러로 미국(3,500억 달러 이상)의 7% 수준에 그쳤으며, 일부 데이터센터 가동률은 20%에 불과하다.
6. 논거: 중국이 전력 우위 활용을 본격화할 징후가 나타나고 있다. Li Qiang 총리가 연례 국정연설에서 처음으로 ‘하이퍼스케일 컴퓨팅’을 언급했고, UBS 애널리스트 Ken Liu는 2029년까지 중국이 25GW의 AI 데이터센터를 추가 구축할 것으로 전망했다.
7. 논거: Huawei의 자체 7nm AI 칩은 미국 제품보다 성능이 낮지만, 다수를 병렬로 적층하면 성능 격차를 좁힐 수 있다. 전력이 저렴하므로 추가 전력 소모는 큰 문제가 되지 않으며, SMIC와 Hua Hong이 7nm 칩 생산 능력을 확대하고 있다.
8. 논거: 중국은 서부 내륙의 풍력·태양광·수력 자원과 데이터센터를 연계하고, 2028년까지 전국 단일 컴퓨팅 풀로 통합하는 인프라 전략을 추진 중이며, 이는 전력 우위의 전국적 활용을 가능하게 한다.
9. 논거: Alibaba가 중국 대기업 중 처음으로 AGI(인공일반지능) 개발을 선언했고, 중국의 신규 5개년 계획(2030년까지)에 ‘AGI 개발 경로 탐색’이 명시되어 에너지 집약적인 AGI 개발 경쟁에서도 전력 우위가 전략적 자산이 될 수 있음을 시사한다.

결론

Xiamen University China Institute for Energy Policy Studies의 Lin Boqiang에 따르면, 중국의 전력 우위는 2020년대 후반에 칩 약점을 충분히 상쇄할 것이며, AI 배포 및 AGI 경쟁 모두에서 결정적 변수가 될 전망이다.

Subscribers can sign up to Drum Tower, our new weekly newsletter, to understand what the world makes of China—and what China makes of the world.

Explore more

→See the latest from topics you follow