The race to build artificial intelligence-driven economies is transforming how the global economy and societies are organized.
Breakthroughs in generative AI and machine learning in recent years have accelerated a new industrial revolution—one in which nations that can develop and deploy the most powerful computing infrastructure will be best positioned to reap the most economic, military and geopolitical advantages.
A central part of this competition is AI data centers, also called “AI factories.” Unlike traditional data centers, these sprawling facilities house a high density of specialized AI chips known as graphics processing units, or GPUs.
They require customized architectures optimized for AI workloads and consume vast amounts of energy. Building them demands massive capital, sustained government and private-sector investment, and institutional flexibility to adapt to the disruptive innovations of an emerging AI-driven economy.
While the United States and China are both racing to harness AI for economic growth and strategic leverage, their approaches to AI factories are diverging–reflecting the distinct economic conditions and technological constraints each faces.
Washington is doubling down on capital-intensive buildouts, betting that private-sector investment and global partnerships will solidify American leadership in AI.
Beijing, meanwhile, is recalibrating—scaling back its once-frenzied data center investment boom amid a slowing economy, industrial overcapacity and a fast-changing tech landscape.
America’s data center bet
The US has embraced an all-in strategy, as shown in President Donald Trump’s recently unveiled AI Action Plan, which focuses on mobilizing private and foreign capital to scale infrastructure as quickly as possible. Tech giants like Microsoft, Google, Amazon and Meta are investing unprecedented sums to build next-generation data centers at home and abroad.
In 2025 alone, the “Magnificent Seven” tech firms are on track to invest a record US$364 billion in data center construction and upgrades. This spending surge is now contributing more to US GDP growth than all consumer spending—a historic first in America’s consumption-driven economy.
Mega-projects like the $500 billion Stargate supercomputing campus – a partnership between OpenAI, Oracle, and SoftBank – illustrate the ambition. AI capital spending is now estimated to account for nearly 2% of US GDP, approaching the scale of the 19th-century railroad boom.
Abroad, Washington is leveraging its technological edge to deepen geopolitical ties, with Trump brokering multibillion-dollar data center deals with the United Kingdom and the United Arab Emirates. US companies are also expanding partnerships with Norway and Japan to build a global AI infrastructure network anchored by an American tech stack.
However, this investment spree has raised concerns about a possible bubble. Analysts warn that the sector’s speculative momentum and unclear business models echo the dot-com and telecom bubbles of the past.
The dominance of hyperscalers—cloud giants like Amazon Web Services, Google Cloud and Microsoft Azure—has raised antitrust and security concerns, with policymakers warning that their control of AI infrastructure creates systemic vulnerabilities.
The 2024 CrowdStrike update glitch that paralyzed global operations by disrupting Microsoft’s cloud services exposed the risks of such single points of failure. Additionally, these mega data centers are exacerbating water insecurity issues in many local communities.
Despite concerns over market concentration, security risks and environmental costs, America’s AI infrastructure development shows no sign of slowing. Unlike traditional infrastructure, AI data centers depreciate rapidly and require constant reinvestment to keep pace with advances in hardware and model design.
The result is an ecosystem heavily reliant on political backing, private capital and economies of scale—one that reinforces hyperscalers’ dominance and embodies America’s techno-capitalist push.
China’s cooling AI data center boom
In the wake of ChatGPT’s 2022 debut, Beijing made data centers a priority, prompting a wave of investment in “national computing hubs” across the country.
By 2024, more than 500 projects had been launched, from Inner Mongolia to Guangdong. Many were part of the “Eastern Data, Western Computing” initiative, which aimed to relocate data storage and processing from coastal cities to inland regions with cheaper land and more renewable energy.
For a time, AI infrastructure was seen as the next great economic driver, capable of offsetting a property market slump and slowing economic growth. But the release of DeepSeek shifted industry focus from training large models to inference–the process of deploying those trained models to generate results. China’s new “AI+” initiative reflects this pivot to inference, aiming to accelerate AI commercial applications.
The earlier building spree has left Beijing with a hangover. By 2025, as much as 80% of China’s new computing capacity, built through the government’s push for data centers in 2022, reportedly sits idle.
President Xi Jinping has issued unusually blunt warnings against over-investment in AI, criticizing “neijuan” (内卷) or involution—a term describing wasteful, self-reinforcing competition and industrial overcapacity.
His statement appears to mark a significant change in China’s stance. During a visit to France last year, Xi insisted that China faced no overcapacity problem.
The pullback on AI data centers reflects a disconnect between top-down industrial policy and actual market demand, compounded by technical setbacks caused by US export controls on cutting-edge chips. Many of China’s data centers in its western regions were built for model training just as the demand shifted toward inference.
Conducting inference typically requires different hardware configurations – faster, more responsive chips that prioritize low latency and efficiency over sheer computational power. As a result, much of China’s existing infrastructure is ill-suited to the application-oriented needs of its major eastern cities.
To address this gap, Beijing has announced plans to build an inference-focused data center in Wuhu, a southeastern prefecture, to serve major urban markets like Shanghai, Hangzhou, and Nanjing.
Hardware shortages remain one of Beijing’s biggest obstacles. The US currently controls over 70% of global computing capacity while using export controls to restrict China’s access to advanced chips such as Nvidia’s H100 and critical packaging technologies.
Domestic alternatives like Huawei’s Ascend chips still lag behind Western hardware in performance and energy efficiency. Moreover, because advanced AI clusters require not only chips but highly engineered interconnections across tens of thousands of processors, US firms continue to lead in systems-level design.
Nevertheless, the Trump administration has reversed some restrictions on US semiconductor exports to China, making Washington’s debate on export controls a critical variable in the AI competition.
Techno versus state capitalism
China’s recalibration underscores the structural limits of its state-led development model. After decades of using industrial policy to scale sectors like solar panels and electric vehicles, Beijing now faces familiar challenges of overcapacity and misaligned incentives.
The US, by contrast, embodies the strengths and weaknesses of a techno-capitalist system. Deep capital markets and a vibrant private sector enable rapid scaling and experimentation but also fuel speculative bubbles and corporate concentration.
The AI competition will hinge not only on who builds more data centers, but, more importantly, on how effectively one’s economic structure can evolve to align investment with technological shifts and market demand in an increasingly AI-driven world.
Lam Tran is an analyst on US foreign and tech policy. She has worked previously at the Center for Strategic and International Studies think tank, the Reischauer Center at Johns Hopkins SAIS and Congressional Research Services.