China’s AI compute demand is set to be driven by inference, not training, by 2029

By NewSunSEO | September 17, 2026

Readers tracking the AI build-out get a clearer map of where the money will run: a state-backed forecast expects inference workloads to overtake training and reach 80% of China’s computing-power market by 2029, reshaping how cloud capacity, chips, and data centres get sized over the next three years.

The forecast comes from the China Telecom Research Institute, the research arm of the state-owned carrier, and was carried by state broadcaster CCTV on Saturday. It frames China’s AI build-out as a pivot from competing on large models and raw computing power toward deploying and selling AI agents, which it expects to drive close to a tenfold rise in annual computing demand over the next two to three years.

Why inference is overtaking training

Training builds a model once. Inference is the cost paid every time someone uses one, which turns AI from a capital expense into an operating expense. The report puts the practical effect in one number: inference will account for 80% of China’s computing-power market by 2029, overtaking demand from training.

The near-term spend picture is already large. Chinese technology companies are expected to spend close to 600 billion yuan, about $89 billion, on AI this year. The report puts that at more than a tenth of all investment in the country.

Europe’s gigafactory timetable

Europe is preparing for the same demand on a different clock. The European Commission opened bidding in July for up to seven AI gigafactories, a EUR 30 billion programme with about EUR 10 billion of public money and EUR 20 billion hoped for from private investors. Roughly EUR 1 billion of it is actually committed.

Applications close on 12 November. Awards are expected in early 2027, construction starts that year, and the machines are due to run by mid-2028. On paper that lands a year before China’s training-to-inference crossover. The plan also assumes nothing else slips.

What has already slipped

Things have slipped. Bidding moved from May to July, the evaluation criteria were delayed more than once, and interest fell from about 70 interested companies to roughly ten expected bidders.

Most of the public half of the EUR 30 billion envelope depends on a budget for 2028 to 2035 that member states have not agreed.

The five-days benchmark

The pace gap is large enough to benchmark in days. Chinese technology companies are on course to spend EUR 1 billion on AI about every five days this year. The committed public funding for Europe’s seven gigafactories sits at roughly the same figure, spread across an entire programme.

Where the margin now sits

The economics have moved alongside the forecast. Inference is where the margin sits, and the companies capturing it are optimising existing models rather than building new ones from scratch. China’s report is a forecast about running models at scale. Europe is still deciding where to put the buildings.

FAQ

What share of China’s compute market will inference take by 2029?

The China Telecom Research Institute forecast, carried by CCTV, expects inference to reach 80% of China’s computing-power market by 2029, overtaking training demand.

How much are Chinese tech companies spending on AI this year?

The report says Chinese technology companies are expected to spend close to 600 billion yuan, around $89 billion, on AI in 2025, which it puts at more than a tenth of all investment in the country.

What is the budget for Europe’s AI gigafactories?

The Commission opened bidding in July for up to seven gigafactories on a EUR 30 billion programme, with about EUR 10 billion of public money, EUR 20 billion hoped for from private investors, and roughly EUR 1 billion committed so far.