Beijing Scales AI Computing Up to 130,000 Petaflops in Tech Race With the US

On 21 July 2026, Beijing authorities announced plans for a sharp increase in the city's computing capacity: add another 50,000 petaflops of intelligent computing in the second half of the year and exceed 130,000 petaflops by year-end. The goal is to accelerate AI adoption in key industries amid the technology race with the US.

What Beijing announced

Beijing plans to add 50,000 petaflops of intelligent computing capacity in the second half of 2026 and bring the city's total to more than 130,000 petaflops by the end of the year. The Beijing Municipal Bureau of Economy and Information Technology announced this at a press briefing on 21 July 2026. One petaflop is 10 to the 15th power floating-point operations per second, the basic unit for measuring the computing infrastructure used to train and run AI models.

Alongside the capacity expansion, the city announced new measures to support the so-called "token economy", a bet on AI models generating tokens at scale in industrial and service scenarios.

• Plus 50,000 petaflops in the second half of 2026
• Total capacity of more than 130,000 petaflops by the end of 2026
• Announcement made on 21 July 2026 at a press briefing
• Source: Beijing Municipal Bureau of Economy and Information Technology
• Goal: accelerate AI adoption in key industries

What is the "token economy"?

The "token economy" is an approach that measures the success of AI infrastructure not by the number of models deployed, but by the volume of tokens those models generate for real industry tasks. A token is the smallest unit of text a language model processes and outputs; the more tokens a system can produce, the broader its use in industry, services and government services.

For Beijing, this means a change of metric: the city is incentivizing not just model development, but mass use of models inside key industries. The additional 50,000 petaflops in the second half of 2026 directly increase this "token output": the computing base determines how many requests the infrastructure can serve at the same time.

Why it matters for the race with the US

Expanding computing capacity is China's direct response to technological competition with the US, where access to compute has become the key resource for AI development. Beijing is betting on infrastructure: the more petaflops available to local companies and the public sector, the faster models are trained and the faster they serve user requests inside the country.

"The city will add another 50,000 petaflops of intelligent computing capacity in the second half of 2026," the

Beijing Municipal Bureau of Economy and Information Technology said in a statement at the press briefing.

The 130,000-petaflop mark by the end of 2026 is a bid to reduce China's dependence on external compute suppliers. With restrictions on deliveries of advanced chips in place, a domestic computing base is becoming a strategic resource on par with data and algorithms.

What this means

Beijing is shifting the AI race from the level of individual models to the level of infrastructure: computing capacity is becoming a strategic resource. The target of 130,000 petaflops by the end of 2026 shows that China is betting on the scale and availability of compute as a way to speed up AI adoption across the economy and keep pace with the US.

Frequently asked questions

How many petaflops does Beijing plan to add?

Beijing intends to add 50,000 petaflops of intelligent computing capacity in the second half of 2026 and exceed a total of 130,000 petaflops by the end of the year.

What is a petaflop?

One petaflop equals 10 to the 15th power floating-point operations per second. It is the standard unit for measuring the capacity of the computing infrastructure on which AI models are trained and run.