China sets a 2030 target for intelligent computing capacity

A five-year infrastructure target

China is targeting 9,800 EFLOPS of intelligent computing capacity by 2030, according to TechNode's reporting on the information and communications industry's 15th Five-Year Plan. The report also identifies RMB 3.8 trillion of information-infrastructure investment over 2026–2030. It describes clusters with at least 10,000 accelerator cards, including some with 100,000 or more.

TechNode puts capacity at 2,185 EFLOPS at the end of June and reports 52 facilities with more than 10,000 accelerator cards each. The figures describe the reported baseline and planned expansion; they are not a forecast of revenues earned by AI services.

Using the two reported capacity figures, the 2030 target is roughly 4.5 times the end-June baseline. That calculation describes the scale of the ambition; it is not an estimate of growth in commercial AI demand. The RMB 3.8 trillion figure covers information infrastructure across the plan, so describing all of it as spending on AI chips would overstate the allocation.

The official explanation is broader than accelerator counts

MIIT's explanatory document describes 13 headline indicators across overall development, innovation, green development, infrastructure and adoption. Computing capacity sits alongside storage, networks and energy-efficiency measures. The ministry also identifies network and data security, standards and industrial applications as areas of work.

That broader framework matters because a cluster is a system. Buying accelerators does not by itself provide a useful service if data cannot reach them efficiently or if the software cannot distribute a workload across the installed hardware. Capacity expansion and productive use are separate milestones.

The plan also addresses networks, energy and governance

MIIT's September 7 explanation groups the work into six areas and 26 tasks. Alongside computing facilities, it calls for network upgrades, space networks and infrastructure that combines intelligent operation with energy efficiency. Separate tasks cover core technologies, supply-chain resilience and standards. The ministry also identifies research on 6G and networks connecting AI agents.

The 13 headline indicators include 5G and 5G-Advanced base-station density, 50G-PON access ports, storage capacity, broadband adoption and the number of connected terminals. On environmental performance, the plan tracks carbon emissions per unit of telecom business and electricity-use efficiency at new large computing facilities. These make it a communications-sector program, not simply a national GPU purchasing target.

Implementation is to involve coordination across ministries, provinces and businesses, support for land, electricity, data and spectrum resources, and ongoing monitoring with midterm and final evaluations. The explanation encourages financing support but does not turn the overall investment figure into an awarded contract for a named supplier.

How to interpret the headline number

EFLOPS expresses an enormous number of floating-point operations per second, but a national total needs a consistent calculation method to support comparisons. Arithmetic precision, the hardware included and whether the figure is theoretical or measured all affect its meaning. The linked planning explanation does not resolve those questions for a specific international comparison.

Readers should therefore avoid converting the target directly into an equivalent number of a particular foreign GPU. The more useful follow-up is how much capacity becomes available to customers, at what utilization and price, and with what performance on training or inference. Those operational measures would show whether the infrastructure program is translating into usable computing.