China’s largest technology companies are preparing to spend much more on AI infrastructure, but a new Moody’s analysis reported by the South China Morning Post argues that the country will still trail the United States by a wide margin in absolute capital expenditure and data-center power. The contrast is becoming one of the defining features of the global AI race: China is scaling rapidly, yet it is doing so from a smaller base and under tighter hardware constraints.
Moody’s estimates that leading Chinese technology companies could lift annual capital expenditure from roughly US$65 billion in 2025 to between US$140 billion and US$165 billion in both 2026 and 2027. The ratings agency projects that U.S. hyperscalers will spend more than US$785 billion in 2026. The comparison is not a verdict on model quality or business outcomes, but it highlights the financial scale behind the largest American cloud and AI platforms.
China’s response is not simply to copy the U.S. spending model. It is to build where costs are lower, deploy across several layers of the stack, and make more use of domestic hardware and software. That logic runs through the country’s recent efforts to expand national compute capacity, including the 2.45 million petaFLOPS milestone reported in August. The question now is whether less expensive infrastructure can offset a widening gap in headline spending.
Capital expenditure is rising sharply
The spending already appears in company disclosures. Moody’s cited Alibaba’s reported 67.7 billion yuan in capital expenditure for the quarter ending in June 2026, Tencent’s 52.8 billion yuan, and Baidu’s 11.4 billion yuan. Each figure reflects a different business mix, but together they show that AI infrastructure has become a core corporate priority rather than an experimental line item.
Alibaba, Tencent, and Baidu must fund more than foundation-model training. They are investing in cloud capacity, enterprise AI services, inference systems, data-center networks, and the operational tooling needed to sell AI products at scale. Baidu’s 11.4 billion yuan figure, for example, comes as it attempts to turn its model and cloud operations into a more durable commercial engine. Tencent’s increase sits alongside a broad push into enterprise software and agent-oriented services.
The money is also being raised through markets. Alibaba’s announced plan for an HK$80 billion share placement, which it said would support AI initiatives, illustrates how funding requirements are shaping corporate finance. EastFrontier’s coverage of ByteDance’s planned Ulanqab compute buildout similarly showed the size and complexity of the facilities Chinese companies are considering as they seek more training and inference capacity.
Still, capex totals are not interchangeable with AI output. A yuan spent on a new data center does not necessarily translate into the same usable compute as a dollar spent on an advanced U.S. facility. Processor availability, networking, power quality, cooling, software maturity, and workload mix all affect what the installed base can actually do.
Lower costs do not erase the capacity gap
Moody’s said China benefits from lower land, construction, and green-power costs, which can reduce the price of building data centers on a per-megawatt basis. That cost advantage is meaningful. It helps Chinese companies stretch investment budgets and build facilities in regions where power and land are more available.
Yet the ratings agency’s own capacity figures show why efficiency should not be confused with parity. It estimated China had 28GW of data-center power capacity at the end of 2025, compared with 52GW in the United States. By 2030, it projected 67GW for China and 100GW for the United States. Gigawatts are a power measure, not a direct count of AI chips or an estimate of model capability, but they remain a useful indicator of the physical base supporting compute deployment.
China also confronts a hardware mix that differs from the U.S. market. Access to the most advanced Nvidia processors is limited by U.S. restrictions, pushing companies toward domestic accelerators, lower-tier imports where allowed, and software optimization. EastFrontier examined this constraint in China’s new training-compute test for domestic AI chip makers, where the challenge was not just producing chips but making them ready for large-scale model work.
That is why lower construction costs matter but cannot settle the race. A cheaper building or power contract improves economics. It does not automatically supply high-end accelerators, mature development tools, or the dense networking required for the largest training clusters.
The financing model is becoming a strategic variable
Moody’s analysis also raises a financial question. Chinese companies often have smaller revenue bases and lower market valuations than their U.S. counterparts. A major expansion in capex may therefore require more debt, asset sales, or equity issuance. That can make infrastructure plans more sensitive to investor confidence and to the speed at which AI services begin generating revenue.
The tension is already clear in the cloud sector. Companies need capacity before they can reliably sell AI services, but prolonged spending can depress free cash flow if demand takes longer to mature. This is not unique to China. U.S. hyperscalers face the same basic investment cycle, albeit with much larger balance sheets and more established cloud revenue.
China’s advantage may lie in selecting where to spend rather than matching every U.S. project. Regional clusters, state-supported facilities, lower-cost power, and faster adoption of domestic hardware could allow companies to deploy more economically in some applications. That strategic focus on regional capacity reflects how energy, location, and infrastructure policy have become intertwined in China’s AI buildout.
But the market should not reduce the story to a simple claim that China is catching up. Moody’s projections suggest two things can be true at once: Chinese AI companies are raising spending aggressively, and the absolute gap with U.S. infrastructure can still widen. The outcome will depend on whether Chinese firms can turn lower-cost capacity and better utilization into commercially sustainable AI services.