While the US Blocks Data Centers, China Races Ahead

The United States still owns the world’s largest stockpile of data centers, chips, cloud platforms and AI talent. But its advantage is increasingly constrained at home by a familiar American problem: local opposition, slow permitting, expensive power and a grid that cannot always keep pace with demand. China, by contrast, is treating AI infrastructure as a national build-out, using state coordination and vast land-and-energy planning to expand computing capacity quickly.

The result is an asymmetric race. America is trying to preserve a commanding lead while managing a political backlash; China is trying to close a large gap by building faster, cheaper and at greater scale. Neither side can win on data centers alone, but the facilities are becoming a decisive measure of whether AI leadership can be sustained.

America’s infrastructure advantage

The scale of the US lead remains striking. Stanford University’s 2026 AI Index counted 5,427 data centers in the United States in 2025—more than ten times any other country, and found that the US consumes more energy for data centers than any other nation. Germany had 529, the United Kingdom 523 and China 449 in the same dataset.

The US also dominates the high-end segment that matters most for frontier AI. It has 84 dedicated AI data centers, more than the next eight countries combined, while Epoch AI estimates that the US accounts for nearly three-quarters of global AI computing power, against just over 14 percent for China.

That advantage rests on more than buildings. American firms control much of the world’s advanced cloud infrastructure, AI models, investment capital and chip ecosystem. Nvidia alone accounts for more than 60 percent of global AI computing capacity among major chip designers, according to Stanford’s index. In other words, Washington’s lead is not merely a matter of square footage; it is an integrated system of chips, software, capital, talent and electricity.

The backlash at home

Yet the US is now fighting over where, whether and how quickly those facilities can be built. Data centers are not abstract “cloud” infrastructure: they are enormous industrial loads that consume electricity, require cooling water, occupy land and can raise transmission and utility costs.

Public resistance has hardened. A Los Angeles Times opinion column cited polling finding that more than seven in 10 Americans oppose building AI data centers near where they live, with 48 percent strongly opposed. Opposition is increasingly organized around electricity bills, water use, noise, land values and the fear that communities will bear the local costs of infrastructure whose benefits accrue largely to national technology firms.

The policy response has become tangible. New York imposed a one-year moratorium on data centers using 50 megawatts or more, becoming the first US state to enact a statewide pause on such projects. Texas Governor Greg Abbott ordered a pause on approvals for new data-center projects through the state grid-interconnection process, citing reliability concerns amid a surge in electricity demand. Monterey Park, California, became the first US city to permanently ban data centers through a public vote.

The backlash is not confined to a few high-profile cases. S&P Global reported that local governments in 40 states have active bans or moratoriums, while more than 50 projects were canceled in 2026 amid community opposition. The Atlantic Council reported that 75 major projects worth more than $130 billion were delayed or canceled in the first three months of 2026 because of organized local resistance.

These fights reflect a real tension. AI companies need gigawatt-scale campuses, but communities are being asked to accept the burdens before they see clear local benefits. The result is a slower, more contested path to expansion, even in the country with the deepest pool of capital and technology.

China’s speed advantage

China’s approach is different. Beijing has made computing infrastructure a strategic priority, coordinating land, power, investment and industrial policy through national programs rather than leaving siting decisions primarily to fragmented local politics.

Its flagship strategy, “East Data, West Computing,” launched in 2022, designated eight national computing hubs and ten initial data-center clusters. The idea is to place much of the computing load in western regions with abundant land and energy, then transmit or route the results to demand centers along the coast.

The physical evidence is visible in places such as Ulanqab, Inner Mongolia, where CNN reported that nearly 90 data centers are slated for construction. The city has become emblematic of China’s effort to convert available land and power into AI and big-data capacity at remarkable speed.

China still trails the US by a wide margin in high-end facilities and computing power. But its ability to build quickly matters because AI infrastructure is cumulative: electricity capacity, fiber networks, chip deployment, cooling systems and skilled operations all take years to assemble. A country that can approve, power and construct facilities faster can compound its advantage over time.

China also faces fewer of the local constraints that slow US projects. It has permissive eminent-domain laws, centralized energy planning and far less visible grassroots resistance to data-center siting. That does not mean Chinese projects are costless or environmentally painless; it means the political system is better able to override or absorb local objections.

Energy is the real battlefield

The central constraint for both countries is electricity. AI training and inference require dense clusters of specialized chips, and those chips require reliable power, cooling and grid connections.

The International Energy Agency estimates that US power demand from data centers will more than double from 2024 to 2030, reaching 426 terawatt-hours, or roughly 9 percent of total US electricity demand. Goldman Sachs has projected that US data-center power demand could more than double from 2025 levels to 66 gigawatts by 2027.

Global AI data-center power capacity had already reached about 29.6 gigawatts by the fourth quarter of 2025, roughly the amount needed to power New York State at peak demand.

For the US, this creates a bottleneck: the technology exists, the money exists, but the grid, transformers, turbines, transmission lines and skilled labor may not arrive quickly enough. Almost 40 percent of US data-center projects have reportedly fallen behind schedule because of energy, equipment and labor shortages.

For China, the challenge is different. It can build physical infrastructure rapidly, but it must overcome restrictions on access to the most advanced foreign chips and continue improving domestic alternatives. Its strategy therefore combines massive construction with efficiency gains, open-source model diffusion and deeper integration of AI into manufacturing, logistics and public services.

Why data centers decide more than AI

Data centers are often described as the “factories” of the AI era. That metaphor is useful, but incomplete. They are also instruments of economic statecraft.

Whoever controls large-scale compute can train larger models, deploy AI services more cheaply, attract developers and set technical standards. Countries and companies that lack access to compute may become dependent on foreign cloud providers, foreign chips or foreign platforms. That is why the US restricts China’s access to advanced AI hardware, while China invests heavily in domestic chips and infrastructure.

The contest is therefore not simply “US versus China.” It is also a contest between two models of industrial power:

The coming test

The US does not need to abandon data centers to retain its lead; it needs to build them more intelligently. That means pairing new campuses with dedicated generation and transmission, improving permitting without weakening environmental safeguards, requiring developers to pay a fair share of grid costs, and giving communities tangible benefits rather than simply warning them about national competition with China.

China’s task is equally demanding. Rapid construction can narrow the infrastructure gap, but AI leadership also depends on chip performance, software ecosystems, research talent and commercial adoption. Beijing’s ability to build fast will matter most if it can convert that capacity into usable, efficient and globally competitive AI systems.

For now, the US retains the stronger hand: more facilities, more computing power, more advanced chips and deeper private-sector capacity. But China is building with unusual speed, while America is increasingly debating whether it wants the infrastructure at all. In an AI race measured in gigawatts, that difference may matter as much as any algorithm.