China to Pour $295 Billion into AI Computing Infrastructure, Challenging the US

China is launching a large-scale state program to develop AI infrastructure with a budget of $295 billion, the largest bet on computing power in the country's history as it competes with the US.

Scale and structure of the program

The investment plan covers several key areas. Funds will go toward building new data centers in strategically important regions, laying high-speed data transmission networks, creating state computing clusters for training large AI models, and subsidizing equipment purchases for Chinese technology companies.

• Construction of large data centers in several regions of the country
• Development of fiber-optic and satellite networks for fast data transmission
• State computing clusters for training foundation models
• Subsidies for purchases of GPU clusters by private AI companies
• Integration of AI infrastructure with industrial and government systems

The program is designed to run for several years and will be carried out through a combination of direct state investment and private capital backed by state guarantees. Notably, Beijing is moving AI infrastructure into the category of strategic national assets, comparable in status to energy and transport.

The race with American data centers

An equally large expansion is under way on the other side of the Pacific. Microsoft, Google, Amazon and Meta* together plan to spend more than $300 billion on expanding computing capacity in 2025–2026 alone. At the same time, the US administration is steadily tightening export restrictions on advanced chips, above all the NVIDIA H100 and its successors.

Here lies China's main structural challenge: without access to the latest American processors, the country has to build its infrastructure around alternatives, namely domestic Huawei Ascend designs and less powerful GPUs bought before sanctions were imposed. According to analysts, this creates a substantial gap in real computing power despite comparable financial investment.

Barriers: chips, energy, talent

Money solves a lot, but not everything. The program faces three serious structural constraints.

First, chips. The US is systematically cutting China off from advanced semiconductors. Domestic alternatives, including the Huawei Ascend 910C, still trail American counterparts in performance on large-model training tasks.

Second, energy. A modern data center for training foundation AI models consumes gigawatts of electricity. Concentrating the load in individual regions puts serious pressure on power grids, and generation expansion does not always keep up with demand.

Third, talent. Despite the country's huge engineering potential, leading AI researchers still gravitate toward American companies and universities, which offer higher pay and greater academic freedom.

What this means

Computing power is becoming the new oil in the global technology race. China's $295 billion investment signals that Beijing treats access to compute as a matter of national sovereignty, not just a business opportunity. For the global AI industry, this means that competition between American and Chinese systems over the next decade will be determined not only by the quality of algorithms, but by who has more computing power, and whether governments are ready to pay for it with public money.

*Meta has been designated an extremist organization and banned in Russia.