When the contest for artificial intelligence moved from algorithms to electricity, the race changed shape. Elon Musk has warned that compute without reliable power is an empty prize; Chinese firms appear to be acting on that logic, quietly buying and building the infrastructure the AI era will demand.
One recent filing shows a new company, Zhonghe (Xiangshan) Nuclear Energy Co., registered with roughly RMB 250 million in capital, in which an Alibaba-affiliated firm is a shareholder. Public disclosures also link Alibaba Cloud to a smaller holding company that is part of the project’s ownership chain. The plant is described as a very large deployment — roughly 7.2 gigawatts of installed capacity and an annual output on the order of 55 billion kilowatt‑hours — that, once complete, would supply substantial baseload power to compute hubs.
The logic is straightforward: modern AI — from large-model training to the sprawling server farms that serve those models — is voracious for electricity. Industry estimates and warnings from the International Energy Agency project dramatic growth in data‑centre consumption this decade. That has pushed cloud providers and big tech into energy markets: buying renewables, underwriting transmission, siting on‑site generation and, increasingly, owning generation outright.
Alibaba’s move is not an isolated bet. Chinese internet groups have been assembling an energy stack for some time. Ant Group and cloud rivals have backed fusion research, backed microgrids, or contracted directly for dedicated substations. Overseas, Microsoft, Google and Amazon have pursued long‑term power purchase agreements or solar acquisitions; the difference is that Chinese firms are coupling generation ownership with a domestic industrial base that can supply transformers, turbines, UPS systems and liquid‑cooling equipment at scale.
That industrial base is where China’s immediate comparative advantage lies. A large share of global transformer capacity, a strong position in high‑voltage transmission technology and a cohort of specialist suppliers that produce critical data‑centre components fast and cheaply have become strategic assets. Investment banks such as Goldman Sachs have recently spotlighted several Chinese power‑equipment makers that stand to benefit from the global scramble for reliable compute power.
The consequence is geopolitical and commercial. If AI success depends as much on secure, low‑carbon baseload power as on chips and models, then control over generation and associated hardware becomes a form of industrial sovereignty. Countries or companies that cannot secure that energy may find themselves sidelined even if they lead in processors or software.
There are caveats. Nuclear projects are costly, politically sensitive and long‑lived. Ownership by internet conglomerates raises governance and regulatory questions: how will grid operators, national regulators and local communities respond when commercial cloud demand shapes generation decisions? Export controls, tariffs and security screeners could complicate the international flow of equipment and technology, even if Western utilities eventually accept Chinese hardware out of necessity.
For global tech firms the message is blunt: securing compute at scale increasingly means securing electrons, not just chips. For investors and policy makers it is a reminder that energy strategy and industrial policy are now central to competitiveness in the digital economy. The interplay of private capital, public utility planning and national industrial capacity will determine who runs the AI infrastructure of the near future.