A fundamental transition is underway in China's artificial intelligence landscape. Rather than competing primarily on the scale and sophistication of foundational models, the country's technology sector is now concentrating on building and deploying AI agents at scale. According to analysis from the China Telecom Research Institute—the research division of the state-owned telecommunications operator—this shift is projected to drive nearly tenfold expansion in computing demand across China over the coming two to three years.

The research institute's findings, shared by state broadcaster CCTV on Saturday, contain a striking projection about how computing resources will be allocated. By 2029, inference operations are forecast to represent 80% of China's total computing-power market, surpassing the share devoted to model training. This distinction matters significantly: training constructs a model once, whereas inference represents the computational cost each time a user interacts with that model—making it an ongoing operational expense rather than a one-time capital investment.

Chinese technology firms are anticipated to invest approximately 600 billion yuan—roughly $89 billion—in AI infrastructure this year alone. The report characterizes this spending as exceeding one-tenth of all investment flowing through the Chinese economy.

Europe's infrastructure race against the clock

Europe is pursuing a parallel infrastructure strategy but operating on a compressed timeline. The European Commission initiated a bidding process in July for as many as seven gigafactories, structured as a EUR 30 billion initiative combining approximately EUR 10 billion in public funding with EUR 20 billion anticipated from private sector partners. However, only around EUR 1 billion has been firmly committed so far—an amount Chinese technology companies are projected to spend on AI roughly every five days during the current year.

The programme's schedule calls for application deadlines on 12 November, with award announcements anticipated in early 2027. Construction is slated to commence that same year, with operational capacity targeted for mid-2028. On the surface, this timeline places European facilities online approximately one year ahead of China's projected computing inflection point, though this assumes no further delays occur.

Delays and uncertainty mounting

The programme has already experienced multiple setbacks. The initial bidding window was pushed from May to July, evaluation standards faced repeated postponements, and the pool of prospective bidders has contracted substantially—from roughly 70 companies expressing interest to approximately ten expected applicants.

Financial commitments present an even more pressing concern. Much of the public funding component depends on budget allocations for the 2028 to 2035 period that European member states have not yet finalized.

The strategic mismatch

The underlying strategic challenge extends beyond scheduling. Inference has become the domain where profitability concentrates, and companies capturing this market segment are increasingly focused on refining existing models rather than developing new ones from scratch. China's research institute is essentially forecasting demand for running deployed models at scale. Meanwhile, Europe remains in the earlier phase of determining where to construct the physical infrastructure itself.

Source: The Next Web