NextFin News - China says it will generate its first electricity from nuclear fusion around 2030, and the deadline is no longer just a scientific ambition: it is being pulled forward by the power-hungry buildout of artificial intelligence. As data centers devour electricity at a pace that has stunned grid planners, Beijing is betting that the "artificial sun" reactors it has spent decades perfecting can move from laboratory curiosities to the backbone of an AI-first economy.
The tension at the heart of this story is simple: China's AI boom is arriving faster than its clean-power capacity can be built. Data centers already account for between 0.9% and 2.7% of the country's annual electricity, depending on the estimate, and government figures point to demand rising from roughly 150-200 terawatt-hours in 2025 to 400 TWh by 2030. Goldman Sachs Research has put the upside case even higher, at almost 600 TWh by the end of the decade. Fusion, once the domain of patient science, is now being framed as infrastructure — and the state is putting money behind the framing.
The 2030 Deadline: From Experiment to Electricity
The target comes from the Chinese Academy of Sciences. At the Fusion Energy Technology and Industry Conference 2026 in Hefei, officials said the BEST project — the Burning Plasma Experimental Superconducting Tokamak — is designed to demonstrate net fusion power gain and fusion-based electricity generation by around 2030. They described the milestone as "lighting humanity's first nuclear-fusion-powered lamp." A keynote at the opening ceremony said global fusion development is "approaching a historic turning point, from scientific exploration to energy demonstration."
A keynote at the opening ceremony said global fusion development is "approaching a historic turning point, from scientific exploration to energy demonstration."
BEST is the hinge between China's research machines and a real power plant. Construction began in 2023, with completion scheduled for 2027, and the facility sits in Hefei between two existing "artificial suns." The first is the EAST tokamak at the Chinese Academy of Sciences' Institute of Plasma Physics, which on January 20, 2025, set a world record by sustaining plasma above 100 million degrees Celsius for 1,066 seconds — more than 17 minutes — shattering its previous record of 403 seconds set in 2023. The second is the HL-3 tokamak in Chengdu, operated by the China National Nuclear Corporation's Southwest Institute of Physics, which has pushed atomic nuclei to 117 million degrees Celsius and is China's largest and most advanced tokamak. Chinese officials have said HL-3 is scheduled to attempt its first fusion ignition experiment around 2027.
The physics goal is specific: unlike earlier experimental devices that merely held hot plasma, BEST is designed to demonstrate actual "burning" of deuterium-tritium plasma — the point at which the fusion reaction becomes self-heating and the device moves from consuming energy to producing a net surplus. That distinction matters because it separates a laboratory demonstration from an engineering proof of concept.
Behind the timeline sits a state machine that western competitors cannot easily replicate. In July 2025, Beijing created the China Fusion Energy Company (CFEC), a wholly state-owned subsidiary under CNNC, with 15 billion yuan — roughly $2.1 billion — in registered capital. CFEC is not a startup chasing a Series C round; it is the central hub of a consortium of central state-owned enterprises and provincial investment vehicles, built to align long-term capital, component manufacturing, and national energy strategy under one roof. Its creation marked the beginning of what industry observers call a "national champion" era for Chinese fusion.
The private sector is smaller but growing. Chinese fusion companies — led by Energy Singularity, ENN, and Startorus Fusion — have raised about €1.5 billion in private capital, according to the European Fusion for Energy Observatory. Energy Singularity, based in Shanghai, achieved first plasma with its HH70 device in June 2024, making it the first tokamak built entirely with high-temperature superconducting magnets. The company has reported a record-breaking 21.7-tesla peak field in large-scale HTS magnet testing, and says it plans a demonstration tokamak that produces fusion output more than ten times its heating input by 2027.
Why AI Changes the Equation
For years, data centers were the efficiency success story of the digital age: workloads climbed while power use stayed flat. That era is over. The International Energy Agency reports that China accounted for 25% of global data-center electricity consumption in 2024, the second-largest consumer after the United States. And the growth is compounding. A single ChatGPT query requires 2.9 watt-hours of electricity, the IEA says, compared with 0.3 watt-hours for a Google search — nearly ten times as much.
Goldman Sachs Research estimates data-center power demand will grow 160% by 2030 compared with 2023 levels, with AI representing about 19% of that demand by 2028. Globally, the IEA projects data-center electricity consumption will more than double to around 945 TWh by 2030 and climb to roughly 1,200 TWh by 2035, growing about 15% a year — more than four times faster than all other sectors combined. In the United States, data centers are forecast to use 8% of national power by 2030, up from 3% in 2022.
The mechanism behind the surge is not hard to find. Efficiency gains in computing equipment and cloud services, which held data-center power flat for years, have slowed since 2020 — even as model sizes, training runs, and inference queries have exploded. This is the Jevons paradox in real time: cheaper, more efficient compute does not reduce consumption; it invites more of it. And the load profile of AI training is unforgiving. A hyperscale cluster cannot be throttled down at night and ramped up at noon the way a factory can; it needs baseload power that wind and solar alone struggle to guarantee without expensive storage.
China's challenge is geographic as much as quantitative. Its data centers cluster on the populous coast, close to users and capital, while its cleanest power — the wind and solar megabases — sits in the interior, thousands of kilometers away. The country has responded with the world's most ambitious transmission buildout. By the end of 2025, China had commissioned 45 ultra-high-voltage lines totaling 52,300 kilometers, accounting for more than 70% of its inter-regional and inter-provincial power-transmission capacity, according to Global Energy Monitor. State Grid said in March 2026 that it plans another 15 UHV lines between 2026 and 2030, which would raise cross-provincial transmission capacity by 35%.
Even that is not enough to fully close the gap. Fusion's promise is not that it will plug in next year; it is that it offers dense, carbon-free baseload that can sit wherever the compute is, without the intermittency tax that renewables impose. For a country that imports much of its fossil fuel, that is an energy-security argument as much as a climate one.
Cyclical Shortage or Structural Shift?
The first question any investor should ask is whether this is a cyclical power squeeze that will ease, or a structural regime change. The evidence points firmly to structural, on three grounds.
First, the efficiency cushion is gone. The historical pattern — flat power demand despite rising workloads — has reversed, and there is no sign it will return on its own. Model complexity is growing faster than chip efficiency, and inference demand scales with the number of users, not the size of the model. Second, the load profile has changed. AI data centers run hot and flat, 24 hours a day, which favors always-on generation over intermittent sources. Third, the timeline has compressed. China is not waiting for fusion to be proven at commercial scale before treating it as strategic; it is building the industrial base now, on the bet that proof will arrive before the grid breaks.
This is a bet Beijing is not making alone. The United States, Japan, and the United Kingdom are all accelerating fusion demonstration projects, with many targeting electricity generation before 2040. But China's approach is distinctive in both scale and structure. Fusion is the third phase of its "thermal reactor–fast reactor–fusion reactor" nuclear roadmap, a strategy first articulated in the early 1980s to guide the long-term evolution of its civilian nuclear system. It was named a "strategic frontier" technology in the 14th Five-Year Plan (2021-2025) and elevated to a "Future Industry" in the 15th (2026-2030), signaling that the state intends to treat fusion as an industrial sector, not just a research program.
The second-order implication is where the story gets interesting. If fusion remains decades away — the base case for most sober analysts — the immediate beneficiaries are not fusion companies but the enablers of the power crunch: ultra-high-voltage transmission builders, grid-equipment suppliers, transformer manufacturers, and the nuclear-fission and renewable developers who will carry the load through the 2030s. China's data-center demand will not wait for fusion; it will be met by whatever can be built fastest. Fusion, in this reading, is a long-dated call option on the AI energy problem, and the state is the only player with a balance sheet long enough to hold it.
The Counter-Thesis: Fusion Is Still Decades Away
The strongest argument against this narrative is the simplest: no fusion device has yet produced net electricity commercially, anywhere. ITER, the international megaproject in France to which China is an equal partner alongside the United States, the European Union, and four others, has been delayed repeatedly and is not expected to begin full deuterium-tritium operations until the 2030s. A longer-range roadmap described by CNNC's chief scientist envisions a staged path: a first fusion ignition experiment around 2027, full design capability for an engineering test reactor around 2030, a demonstration reactor by 2045, and fusion electricity feeding the national grid only by mid-century.
That longer timeline is a direct challenge to the 2030 electricity claim. If the 2030 target slips — as fusion targets have slipped for half a century — then today's buildout is a hedge, not a plan, and the capital poured into CFEC and the private players is buying optionality rather than capacity. The history of fusion is littered with confident deadlines that moved right.
The rebuttal is that China is running both timelines at once, and they are not contradictory. The 2030 target is for a demonstration facility — a proof that fusion electricity is physically possible at pilot scale — while the mid-century date is for grid-scale deployment. What matters for investors is not whether fusion powers Shanghai in 2030 (it will not); it is whether the engineering milestones along the way — first plasma, net gain, sustained burning — de-risk the technology fast enough to attract the trillions of dollars needed for commercial rollout. A successful pilot changes the cost of capital for the whole sector.
The falsifying signal is concrete and near-term. If BEST fails to achieve net fusion power gain by 2030, or if HL-3's ignition experiments do not demonstrate sustained burning plasma by 2027-2028, the structural thesis weakens materially. At that point, fusion reverts to a long-dated research program, and China's AI power problem gets solved by fission, natural gas, and coal instead — with consequences for the country's carbon targets.
What Comes Next
In the short term, the winners are the enablers of the power crunch. Watch the transmission and grid-equipment builders, the ultra-high-voltage supply chain, and the fission and renewable developers who will carry the load through the 2030s. China's data-center demand will not wait for fusion; it will be met by whatever can be built fastest, and that is a near-certain revenue stream for the next decade.
In the medium term, watch the milestone cadence. HL-3's first ignition experiment around 2027 and BEST's completion in 2027 are the two dates that separate rhetoric from engineering. A successful net-gain demonstration by 2030 would trigger a global re-rating of fusion as an investable asset class rather than a science project. Private capital — currently about €1.5 billion in China versus roughly $3 billion raised by Commonwealth Fusion Systems alone in the West — would follow, and the supply chains for superconducting magnets, tritium breeding blankets, and high-heat-flux components would become investable themes in their own right.
In the long term, the structural question is whether fusion becomes a regime change in energy economics: carbon-free baseload at a cost that undercuts the intermittency tax on renewables. If it does, the countries that built the supply chains first will own the next energy order. If it does not, the AI boom will still have forced a historic buildout of fission, transmission, and renewables, and fusion will remain the promise that never quite arrived.
China's fusion bet is not a prediction that reactors will light up cities in 2030. It is a statement that the AI age has made the cost of waiting higher than the cost of trying — and that a state with a long enough balance sheet can afford to find out.
Explore more exclusive insights at nextfin.ai.
