Level 1 — Absolute Beginner

Artificial intelligence needs very strong computers. These computers do millions of small jobs every second.

China wants many more of these computers. The government made a new plan on Monday and Tuesday this week.

The plan is for five years. It goes from 2026 to 2030. China wants to make its computer power more than four times bigger.

The plan will cost a lot of money. Big computer buildings will open in many parts of the country. Other countries are watching this plan closely.

computer

A machine that stores information and does work with it.

plan

An idea about what to do in the future.

government

The group of people who run a country.

power

Strength, or how much work a machine can do.

year

A period of twelve months.

money

What people use to buy things.

country

A land with its own government, like China or Brazil.

build

To make something, like a house or a building.

Level 2 — Elementary

China's Ministry of Industry and Information Technology published a new five year plan for the information and communications sector on September 8. Its headline goal is 9,800 exaflops of intelligent computing capacity by 2030.

An exaflop is a measure of speed. One exaflop means a quintillion calculations every second, which is a one followed by eighteen zeros. In June the country had 2,185 exaflops, already 177 percent more than a year earlier.

Reaching the target would need more than four times today's capacity. The plan sets aside about 3.8 trillion yuan, or roughly 532 billion US dollars, of total spending on information technology infrastructure across the five years.

The document also calls for the orderly deployment of computing clusters containing more than ten thousand accelerator cards each. China already runs 52 of these large facilities under a national programme called East Data West Computing, which sends work from the crowded eastern cities to cheaper sites in the west.

ministry

A government department responsible for one area of policy.

capacity

The largest amount that something can hold or do.

calculation

A sum or mathematical operation.

infrastructure

The basic systems and structures a country needs, such as roads or networks.

cluster

A group of computers working together as one system.

accelerator

A special chip that speeds up certain kinds of computing work.

deployment

The act of putting something into active use.

target

A result that someone is trying to reach.

Level 3 — Intermediate

Beijing rarely announces an ambition without also publishing the arithmetic behind it. On September 8 the Ministry of Industry and Information Technology released its five year development plan for the information and communications sector, and the number that will be quoted everywhere is 9,800 exaflops of intelligent computing capacity by 2030. The comparison that gives it weight is the June figure of 2,185 exaflops, itself up 177 percent in a year.

An exaflop denotes a quintillion floating point operations per second, and intelligent computing capacity is not quite the same measure as classical supercomputing performance, since it counts the lower precision arithmetic that neural networks actually use. That caveat matters when comparing national totals, but the direction of travel is unambiguous: the plan implies more than a quadrupling in four and a half years.

Behind the target sits roughly 3.8 trillion yuan, about 532 billion US dollars, of cumulative information technology infrastructure spending. The document calls for the orderly deployment of clusters exceeding ten thousand accelerator cards, language that signals both encouragement and restraint, since provincial governments across China have been racing to announce data center projects whose demand is far from assured.

The programme extends the existing East Data West Computing framework, which routes processing from the dense and expensive coastal provinces toward sites in the west with cheaper land and power. Fifty two facilities of ten thousand cards or more are already running. The strategic subtext is not hidden. With American export controls restricting access to the most advanced foreign accelerators, China's answer is to compensate through scale, domestic silicon and the sheer aggregation of capacity.

arithmetic

Calculation with numbers; the numerical basis of an argument.

denote

To mean or represent something.

precision

The level of exactness used to represent numbers in computing.

caveat

A warning that something should be understood with limits.

quadrupling

An increase to four times the original amount.

restraint

The act of holding back or limiting something.

aggregation

The gathering of many separate parts into one total.

subtext

A meaning that is implied rather than stated directly.

Level 4 — Advanced

Industrial policy in China is legible chiefly through its numbers, and the five year plan for the information and communications sector released by the Ministry of Industry and Information Technology on September 8 supplies an unusually specific one: 9,800 exaflops of intelligent computing capacity by 2030. Set against the 2,185 exaflops recorded in June, itself a 177 percent annual increase, the target requires the country to more than quadruple its aggregate accelerator throughput inside four and a half years, and it does so at a moment when the most capable foreign silicon remains restricted by American export controls.

The unit deserves scrutiny before the ambition does. Intelligent computing capacity aggregates lower precision arithmetic of the sort neural networks consume rather than the double precision benchmarks by which conventional supercomputers are ranked, so the figure is not directly commensurable with a TOP500 table. That is a methodological caution rather than a dismissal. Inference and training workloads genuinely run in reduced precision formats, and a state that can marshal a large volume of such capacity has purchased real optionality regardless of how the accounting is presented.

The financial commitment, roughly 3.8 trillion yuan or about 532 billion US dollars of cumulative information technology infrastructure investment, is accompanied by a phrase worth reading carefully. The plan urges the orderly deployment of clusters of more than ten thousand accelerator cards, and orderly is doing considerable work. Provincial and municipal authorities have spent two years competing to announce computing projects, many of them speculative, and Beijing is simultaneously encouraging aggregate scale and attempting to discipline a build out that risks stranding capital in facilities without workloads.

Architecturally the programme deepens the East Data West Computing framework, under which compute intensive tasks are migrated from the land constrained and power expensive eastern seaboard to interior provinces where both are cheaper, with fifty two facilities of ten thousand cards or more already operational. The wager is straightforward and consequential. Denied the frontier accelerators, China proposes to substitute quantity, domestic fabrication and geographic arbitrage for per chip superiority, betting that at national scale the aggregate is what determines who can train and serve the largest models.

legible

Able to be read or understood clearly.

aggregate

The total formed by combining many separate parts.

commensurable

Able to be measured or compared by the same standard.

methodological

Relating to the methods used to measure or study something.

marshal

To gather and organise resources for a purpose.

optionality

The value of having future choices available.