Showing posts with label this is the one advantage China has -- China's Electricity Advantage. Show all posts
Showing posts with label this is the one advantage China has -- China's Electricity Advantage. Show all posts

Saturday, November 21, 2026

AI: In the AI race with the US, this is the one advantage China has -- China's Electricity Advantage

https://str.sg/ojRDP

*In the AI race with the US, this is the one advantage China has*

The Straits Times 
2026-11-21

Yew Lun Tian
Assistant Foreign Editor

Philip Wen
Regional Correspondent


BEIJING/BANGKOK – Rising from a floating platform 70km off China’s southern coast, two giant wind turbines perched on a V-shaped frame stand as tall as a 73-storey building. Whirring at full speed, they can generate enough power a year to supply 30,000 households.

Tethered to the ocean floor by three heavy cables, Ming Yang Smart Energy’s OceanX, which took The Straits Times a bumpy five-hour round trip to reach by speedboat, is designed to tap the stronger, steadier winds in deep offshore waters.

OceanX already generates the most power of any floating wind platform – according to the China Classification Society, a state-owned enterprise that sets technical rules for ships and offshore installations – and at what the company says to be a fraction of the cost of its American and European competitors.

Having successfully passed technical reviews including weathering three Category 5 super typhoons, Ming Yang plans to break its own record and commission an even larger floating twin turbine platform with a total capacity of 50MW – triple that of OceanX – as early as 2027. 

“What we can do is something other countries, including the US, cannot copy or surpass,” chairman Zhang Chuanwei said. 

The engineering feat is but one example of the scale and ambition of China’s renewable energy industry, and its role in driving the massive expansion of the country’s electricity-generating capacity overall. 

It is this vast power buildout that could help China bridge the gap in its race for artificial intelligence supremacy with the United States – supplying the electricity needed to run increasingly power-hungry computing systems and data centres, even as Washington seeks to restrict Beijing’s access to its most advanced American semiconductors.

China generated 10.6 trillion kilowatt-hours of electricity in 2025 – roughly a third of global output and 2.4 times as much as the US.

In 2025 alone, China added electricity-generating capacity equivalent to more than three times France’s entire installed capacity. Under its 15th Five-Year Plan, an economic and social blueprint announced in June covering 2026 to 2030, China aims to add a further 1,500 gigawatts of electricity-generating capacity – more than the installed capacity of the entire European Union. Two-thirds of that growth is being targeted via renewable sources. 

For the US, the constraint is how quickly its electricity infrastructure can keep pace with the AI boom. New data centres need grid connections, transmission lines and other infrastructure that can take years to deliver. The International Energy Agency estimates that building transmission lines in advanced economies can take four to eight years, while waiting times for critical grid components have doubled over the past three years. 

“The US doesn’t even make enough of the transformers needed for power plants and the grid. It is heavily dependent on imports from other countries, including China,” said Wen Yonggang, a professor and President’s Chair of Data Science at Nanyang Technological University (NTU) in Singapore.

U.S.-China ‘electron gap’

The energy disparity has alarmed some of America’s most prominent technology leaders. OpenAI warns that an “electron gap” with China threatens US leadership in artificial intelligence, urging Washington to accelerate the construction of new power capacity.  

“The US cannot maintain its AI advantage if the ‘electron gap’ with the PRC persists, and our ability to build new energy capacity is trapped in a regulatory chokehold,” OpenAI said, referring to China by its formal name, the People’s Republic of China.

Training and running AI models require vast numbers of specialised chips working together in data centres, consuming enormous amounts of electricity both to perform calculations and to keep the equipment cool. How quickly that power can be supplied helps determine how quickly companies can expand their AI capabilities.  

Elon Musk, whose xAI is building power-hungry supercomputers in the US, has called electricity the key constraint on AI growth and predicted that China would eventually “far exceed the rest of the world” in computing power partly because it generates more electricity than the US and is adding capacity much faster.

Former Google chief executive Eric Schmidt told the US Congress in 2025 that America needed energy “in all forms” to win the AI race against China. 

“We need energy in all forms, renewable, non-renewable, whatever. It needs to be there, and it needs to be done quickly,” Schmidt said.

The “electron gap” creates a consequential trade-off in the US-China technology race: America’s lead in advanced chips produced by the likes of Nvidia may be constrained by how quickly it can supply power, while China’s electricity buildout could give it more scope to deploy larger numbers of less efficient domestic processors.  

Huawei is pursuing that approach, linking hundreds of its Ascend processors into systems designed to compete with Nvidia’s offerings. The strategy consumes considerably more electricity, making the cost and availability of power an important part of China’s efforts to work around US restrictions on advanced chips.

In July 2025, President Donald Trump signed an executive order to accelerate federal permitting for large data centres and their supporting energy infrastructure, including transmission lines and equipment for gas, nuclear and geothermal power.

Both American and Chinese technology companies are also expanding their computing infrastructure beyond their own borders, including in South-east Asia. Those investments serve growing regional demand while also giving the companies access to additional electricity supplies. 

The global spending boom comes amid the growing spectre of an investment bubble due to mounting concerns that AI revenues will struggle to justify the vast investment in chips and data centres, as well as vocal safety fears that increasing autonomous systems could enable cyberattacks or act beyond human control. 

Yet the rivalry and deep suspicion between the US and China add pressure to not let up. Washington fears surrendering its technological lead; Beijing fears being locked into dependence on American technology. For both, slowing development carries the risk of giving a strategic rival an advantage.

The grand stakes underpinning the AI race extend beyond the commercial realm. Such is the known and unknown potential for cutting-edge AI that it is taken as a given that leadership in the technology would likely deliver a competitive advantage in military and intelligence capabilities, and accelerate scientific discoveries. That prospect makes the chips, data centres and electricity underpinning AI matters of national security.

Cheap and abundant power

China’s advantage lies in the potential of its electricity system as a whole: how much power it can generate, how swiftly it can move that power to where it is needed, and how much users pay for it. These factors will help determine how efficiently China can expand the data centres powering its AI ambitions.

China’s electricity expansion serves ambitions that extend beyond AI. Beijing is building for an economy that will consume much more electricity, with more cars, factories and buildings running on electrical power, while seeking to curb emissions and strengthen energy security.

China began laying the groundwork for this vast electricity system long before the current AI boom, reflecting its ability to plan years ahead and marshal resources towards strategic goals. The same long-term approach helped it build its dominance in rare earths, another critical resource in its AI rivalry with the US. 

That planning has also been sustained across successive policy cycles, with Beijing now seeking to harness its electricity advantage for AI. Its 15th Five-Year Plan calls for the “coordinated deployment of green electricity and computing power”, alongside the development of a nationwide computing network and large-scale AI computing clusters. 

In contrast, such policy continuity is harder to sustain in the US, where political power changes hands and infrastructure projects face resistance from local communities. The Democrats’ sweeping victory in the November midterm elections has raised concerns that Trump’s push to accelerate AI and data centre development could stall. Democrats, who will take control of Congress in January, are preparing tougher regulations on AI and data centres, particularly to protect consumers from rising electricity costs. 

Meanwhile, the scandal that forced Jensen Huang to resign as Nvidia’s chief executive has exposed divisions within the Trump administration over how best to preserve America’s lead in AI. And the departure of one of the most central figures responsible for the AI boom has also increased investors’ doubts about whether demand for computing power can justify the vast sums of money being poured into data centres.

On Nov 16, The US Commerce Department’s Bureau of Industry and Security (BIS) issued an interim final rule outlining tightened controls on advanced chip shipments to Singapore, Malaysia, Thailand and Indonesia, alongside Brazil. It also introduced a licensing requirement for overseas data centres that use such chips to supply large amounts of computing power remotely to Chinese companies, seeking to close down a loophole in China’s access to cutting-edge semiconductors.

“The new remote access rules and other BIS measures will make it more difficult for China to access advanced GPUs (graphics processing units),” said Kyle Chan, a fellow in the John L. Thornton China Center at the Brookings Institution, an American think-tank. “At the same time, Beijing is already all-in on its semiconductor self-sufficiency drive, so I don’t think this changes their basic policy direction. These are ultimately bumps in the road for China’s longer-term push for resilience in AI supply chains.”

Driving China’s electricity output is a vast and still-expanding fleet of power plants, wind turbines and solar panels. By the end of 2025, China’s installed generating capacity – the maximum output its power plants can produce – had reached 3.9 billion kilowatts, up 16.1% from a year earlier. The China Electricity Council, a national industry association, expects that figure to reach around 4.3 billion kilowatts by end 2026.

Renewables are driving much of this expansion, accounting for more than four-fifths of the generating capacity added in 2025, bringing their share of total installed capacity to about 60%.  

Installed capacity is not the same as electricity generated because the wind does not always blow and the sun does not always shine. Wind turbines and solar panels do not produce their maximum output continuously. Renewables, including hydropower, generated about 38% of China’s electricity in 2025.

China’s electricity is not just plentiful. It is also cheap. The advantage is starkest in western regions like Yunnan, Xinjiang, Gansu, Ningxia, and the western parts of Inner Mongolia, where industrial power averages just 0.39 yuan (seven Singapore cents) per kilowatt-hour – about two-thirds of the US average of 8.89 US cents (11.5 Singapore cents). 

For power-hungry data centres, a seemingly small 0.1-US cent difference in electricity prices can add up quickly: at a steady 10 MW IT load, it would shift annual energy costs by about US$131,000, before accounting for load growth.

A 2025 Schneider Electric report, citing data from the Energy Research Institute of China’s National Development and Reform Commission, said electricity accounts for about 57% of data-centre operating costs.

But national abundance does not guarantee that a data centre can obtain reliable electricity where and when it needs it. About 9% of China’s potential wind and solar generation went unused in the first half of 2026, according to the National Energy Administration. The industry term is curtailment: output that turbines and panels could have produced but was held back because the electricity system could not accommodate it.

Turning China’s energy resources into a computing advantage depends on the connections between its power plants, grids and server farms.

To deliver generated electricity to where it is needed, China is building a vast network of ultra-high-voltage transmission lines that carry power over long distances with relatively low losses. These lines connect energy-rich western regions with demand centres in the east, allowing coastal cities and their data centres to tap electricity generated thousands of kilometres away. The network already spans more than 50,000km – longer than the circumference of the earth.

China is poised to convert its comparative advantage in affordable, abundant electricity into a new stream of AI service exports. Overseas clients, chiefly across South-east Asia – Indonesia, Malaysia, Singapore, Thailand and Vietnam – send requests via application programming interfaces (APIs) to data centres in China, where the computing is done before the results are sent back to them. 

Channelling energy from West to East

Just under an hour’s drive from central Guiyang city, under-construction apartment blocks lined the broad roads of Guian New Area alongside low-rise industrial buildings. Here in the mountainous Guizhou province, Huawei, Tencent, Apple and China’s three main telecoms operators have established data centres to serve customers far beyond the province.

At Huawei’s tightly guarded compound, elderly visitors and families with young children posed for photographs against pastel-coloured buildings, a clock tower and arched bridges. A waterfall roared at one end of the complex – part of a landscape that, according to Huawei, also helps dissipate heat from its computing equipment. 

Nearby, Tencent’s Seven-Star Data Center takes advantage of the naturally cool conditions inside a hillside.

The cluster embodies Beijing’s “East Data, West Computing” strategy: moving some of the processing and storage of data from the prosperous eastern seaboard to inland regions with more plentiful land and energy. Guizhou’s cool climate helps reduce the electricity needed to keep servers from overheating. The approach brings computing closer to power supplies, while creating a new industry for regions far from China’s coastal technology hubs.

The experiment has also drawn visitors from overseas. At a permanent exhibition in Guizhou devoted to the strategy, schoolchildren and organised groups were among the few visitors. 

A Tanzanian who works at a data centre in his home country was among those who visited in August. The engineer, who wanted to be known only as Abdul, was on a trip sponsored by China’s Ministry of Commerce. He said he wanted to see how China’s east-west approach worked and take useful ideas home.

For traditionally poorer and more remote provinces like Guizhou, the promise is that the data flowing west will bring opportunity with it – better infrastructure, new businesses and jobs that give young people a reason to stay rather than seek their fortune elsewhere.

Zhao Zehe, a 72-year-old retired insurance agent, said that while previously there was limited economic opportunity in the local area, now there was growing recognition that came with the data centre industry, which employs his nephew. 

“It’s good if it gives young people a reason to stay in Guiyang instead of moving away to bigger cities for work,” he told ST.

Launched in 2022, “East Data, West Computing” is organised around eight national computing hubs encompassing 10 data centre clusters across eastern and western China. By March 2026, these hubs accounted for more than 80% of China’s installed AI computing capacity, according to the National Data Administration.

China is also experimenting with ways to supply renewable electricity directly to data centres, cutting transmission losses and operating costs.

In Qingyang, in north-western China’s Gansu province, a pilot project pools electricity from wind and solar farms to supply data centres directly. According to local authorities, the arrangement has raised the park’s clean-energy share to 88% and lowered the electricity price users pay to about 0.40 yuan per kilowatt-hour.

Off Shanghai’s Lingang, an underwater data centre draws over 95% of its power directly from a nearby offshore wind farm. As an added benefit, it uses seawater for cooling, cutting energy consumption by more than a fifth compared with land-based counterparts.

Using brute compute to compete

Yet China’s electricity advantage comes up against a substantial gap in advanced processors. Nvidia’s cutting-edge processors are prized for their ability to handle the vast calculations needed to train and run sophisticated models. 

Nvidia’s appeal also rests on the software and networking built around its processors. Its CUDA platform provides tools that developers use to make their models run efficiently, while high-speed connections allow large numbers of chips to work together. Replacing Nvidia, therefore, means adapting software and rebuilding systems, as well as finding an alternative processor.

In response, Huawei says it has found a brute-force workaround to match Nvidia’s semiconductor dominance: glueing and folding its chips together. The Chinese national technology champion’s main approach is to build ever-larger computing systems. Instead of trying to match Nvidia chip for chip, Huawei connects hundreds of processors so they can work together as one giant computer.

Its flagship CloudMatrix 384 system, for example, links 384 Ascend 910C chips, compared with 72 GPUs in Nvidia’s GB200 NVL72 system. The sheer number gives Huawei an edge on some measures: CloudMatrix offers about 1.7 times as much raw computing power and 3.6 times the memory capacity. 

Getting hundreds of chips to work together efficiently is difficult because they need to exchange huge amounts of data without creating bottlenecks. Huawei tackles this with its own high-speed interconnect system, UnifiedBus, which lets data move between chips with very low delays and at speeds of up to 392 gigabytes per second.

Huawei is also experimenting with stacking chip components vertically, rather than placing them side by side. Its LogicFolding architecture puts active circuit layers face-to-face and connects them through tiny vertical links. This lets Huawei pack more computing power into the same space, even when it cannot make transistors much smaller. Early data suggests the technique can increase transistor density by 55% and improve energy efficiency by 41%.

Research group Epoch AI estimated in a September analysis that Nvidia’s B300 offered roughly seven times the theoretical computing performance of Huawei’s Ascend 950 series.

According to a transcript of an investor meeting circulating online, DeepSeek founder Liang Wenfeng acknowledged that Huawei still trails Nvidia at the individual-processor level: he said about four Huawei 950 cards are needed to match one of Nvidia’s latest high-end AI processors, a gap he put at roughly two years. But he said Huawei could make up for this by combining large numbers of chips into supernodes that can perform the same tasks as Nvidia’s latest systems.

However, this strength-in-numbers strategy has a major drawback: it uses far more electricity. Huawei’s CloudMatrix system draws about 560kW, nearly four times as much as Nvidia’s comparable system. 

Brookings Institution’s Chan said China remained constrained by overall production capacity for advanced semiconductors. 

“China is trying to overcome its chip disadvantage by building larger clusters of chips. However, China is still constrained by overall production capacity for advanced semiconductors. In addition, scaling up chip clusters may run into problems with speed and reliability,” he said.

How AI is like a ‘five-layer cake’

So, can China’s strength in electrons compensate for its weakness in silicon?

The jury is still out, said NTU’s Wen. In the near term, however, China may have an advantage, he told ST.

The US is struggling to add power quickly enough for its rapidly expanding AI data centres, with grid connections, transformers and other equipment becoming bottlenecks. Building new grid infrastructure can take five to 15 years.

“Energy is a key bottleneck for the United States in its data centre buildout. China, in contrast, is able to build new power generation rapidly, particularly renewable energy away from coastal regions,” Chan said.

“However, it would be a mistake to equate China’s ability to rapidly build energy capacity with a decisive advantage in compute buildout. China is still heavily bottlenecked on chips. Meanwhile, American AI accelerators are getting increasingly power efficient,” he added.

DeepSeek’s breakthrough in early 2025 and Moonshot AI’s Kimi K3 in July 2026 showed that Chinese companies can produce models that compete with some of the best from the US despite restrictions on their access to advanced chips. Kimi K3, for instance, was released as a 2.8 trillion-parameter, open-weight model whose performance approached leading US systems.

But strong models do not necessarily mean Chinese chips have caught up. DeepSeek, Moonshot and other Chinese AI firms have also been accused by US authorities and American AI companies of using distillation – training their models partly on the outputs of more advanced foreign models. China has rejected such accusations as unfounded and described distillation as a widely used industry technique.

For Wen, that makes China’s hardware gap difficult to pin down. “It’s hard to say how far Chinese chips lag behind,” he said. “Some semiconductor experts say two to three generations; some say just one to two.”

More fundamentally, he cautioned against thinking of electricity and chips as substitutes. Having more of one cannot simply make up for having less of the other.

He pointed to the recently-departed Nvidia chief Jensen Huang’s description of AI as a “five-layer cake”: energy at the bottom, followed by chips, infrastructure, models and applications. Each layer depends on the ones beneath it.

“Energy is the foundation,” Wen said. “Without energy, even if you have the chips, they are useless.”

luntian@sph.com.sg philipwen@sph.com.sg

Additional reporting by Michelle Ng

From Silicon Valley to Shenzhen, the battle for technological supremacy is being fought on a new frontier. The Straits Times’ correspondents go behind the algorithms to explore how geopolitics are reshaping the global AI race. Read more about China’s open-source models at  https://str.sg/tzbDw

Its vast power buildout could help bridge the “electron gap”. The Straits Times examines the imbalance and how it is reshaping the AI rivalry.