# China Sets a 2030 Target of 9,800 Exaflops of AI Computing Backed by 3.8 Trillion Yuan

The plan requires more than a fourfold increase from the 2,185 exaflops recorded in June and explicitly ties new clusters to domestically made chips.

- Published: 2026-09-10T06:24:02.019Z
- Canonical: https://polylog.news/ai/2026-09-10/china-sets-a-2030-target-of-9-800-exaflops-of-ai-computing-b
- Publisher: Polylog (AI desk)
- Section: geopolitics
- Sources: [Polylog editors](https://polylog.news)

China's Ministry of Industry and Information Technology (MIIT) published its information and communications development plan for 2026 to 2030 on September 7. It sets a target of [9,800 exaflops of intelligent computing capacity](https://www.nbd.com.cn/articles/2026-09-07/4574675.html) by 2030 and cumulative information infrastructure investment of 3.8 trillion yuan, roughly 530 billion dollars at current exchange rates.

The distance between the current level and the target is large. National intelligent computing capacity stood at 2,185 exaflops at the end of June, which Chinese official figures put at 177 percent higher than a year earlier. Reaching the target therefore needs more than four times the June level within four years. The plan calls for orderly deployment of clusters at the ten-thousand-accelerator and hundred-thousand-accelerator scale, and for strengthening compatibility between that infrastructure and domestic computing chips.

The unit requires careful interpretation. Chinese intelligent computing totals aggregate low-precision throughput used for training and inference, not the double-precision measure used in scientific supercomputing rankings, so the headline number is not comparable to a Top500 figure (the standard ranking of the world's fastest supercomputers) and mixes accelerator generations of very different efficiency. It is a procurement and buildout target, not a capability claim.

The compatibility requirement is where the geopolitics show up most clearly. Washington's export restrictions on advanced accelerators have made the supply of foreign chips unreliable, and MIIT is now writing domestic silicon into a five-year infrastructure indicator. That converts a workaround into a planning requirement for provincial operators and state-linked data center builders, whose purchasing decisions follow these documents closely.

## What this means

A state target expressed in installed compute pulls demand toward Chinese accelerator vendors, optical interconnect suppliers and grid capacity, because provincial buildouts are scored against a national indicator rather than against return on capital. American accelerator makers lose an addressable market they were already losing to controls, while power generation and transmission become the binding constraint inside China. If domestic chips cannot supply the volume, the target slips or is met with lower-efficiency parts, and the second outcome would show up as rising energy use per unit of usable training throughput.

## What to watch

- Whether provincial governments publish matching capital plans, which would show the national target is being funded rather than announced.
- Reported shipment volumes for domestic accelerators, the practical test of whether the compatibility requirement can be met.
- Electricity supply agreements for new clusters, since power availability, not silicon, is the constraint that usually decides delivery dates.