China's Answer to the Chip Ban Has a Price Tag

On September 7, 2026, China's technology ministry posted a document online. No press conference, no minister on camera. Just a planning paper for the "information and communications" sector, the kind of bureaucratic filing that rarely makes news anywhere.

Buried inside was a number: 9,800. That's how many exaflops (it's a measure of performance for a supercomputer that can calculate at least one quintillion floating point operations per second) of AI computing power China intends to have running by 2030, nearly six times what it had at the end of 2025. Attached to that number was another one, $532 billion, the price tag for getting there, and a rule that had barely existed three months earlier: none of it, as much as Beijing can help it, would run on an American chip.

The document is China's clearest answer yet to a question Washington has been testing since 2022: what happens when you cut a country off from the chips it needs to build AI? The answer, it turns out, isn't that the country stops trying. It builds its own supply chain instead, at a scale large enough to make the ban feel less like a wall and more like a toll it has decided to stop paying.

The Plan That Doubled in Three Months

The number wasn't always $532 billion. In June 2026, Chinese planners had floated a $295 billion, five-year proposal to link data centers into a national network, an ambitious figure on its own. By the time the Ministry of Industry and Information Technology published its own version in September, dated internally to mid-August, the commitment had grown by roughly 80%.

Beijing never explained the jump publicly. Analysts who track the plan read it less as a single new project and more as a widening of scope, from "connect the data centers that already exist" to "build the entire stack, chips and energy and storage included, without depending on anyone else for the pieces that matter most."

Did You Know?

China's intelligent computing capacity had already reached 2,185 exaflops by the end of June 2026, up 177% from a year earlier, according to the ministry. The 2030 target isn't a plan starting from zero. It's an acceleration of something already moving fast.

A Unit Most People Have Never Heard Of, Growing Six Times Over

An exaflop is a quintillion calculations a second, a number too large to make intuitive on its own. What matters more than the unit is the trajectory: from 1,590 exaflops at the end of 2025 to a targeted 9,800 by 2030, a roughly sixfold increase in under five years, paired with a storage target that more than triples, from 540 exabytes to 1,700.

That computing and storage buildout runs through the same power-hungry buildings that use so much water and electricity everywhere else in the world. China's version adds a second constraint on top of the usual one: it has to be built, largely, on chips the country makes itself.

The Silicon Has to Be Chinese, Even When That's Harder

The plan's chip mandate is specific: domestic suppliers, led by Huawei, are expected to cover at least 80% of the AI hardware behind the buildout. In May 2026, regulators had already cleared nine categories of domestically made AI chips, from Huawei, Alibaba, Biren Technology, and Moore Threads, for deployment across government and security-sensitive sectors.

Bloomberg's reporting put Huawei's 2026 target at roughly 600,000 Ascend 910C chips, manufactured through its partner SMIC, which is doubling capacity on its most advanced production lines to keep up. It's a direct bet against the company whose graphics-chip wager decades ago built the backbone of the global AI industry in the first place.

The ambition is running ahead of what China's supply chain can fully deliver today. Industry analysts tracking the supply chain estimate SMIC's wafer output could support over a million Ascend chips a year, but domestic memory production caps real shippable output at under 300,000 units without foreign chip stockpiles, a bottleneck the plan doesn't fully solve so much as budget around.

Nvidia's share of China's AI chip market, once above 40%, is reported by the South China Morning Post and industry analysts to be on track to fall to around 8% by the end of 2026, with Huawei's share rising toward half the market. In an April 2026 interview, Nvidia CEO Jensen Huang said the company's China market share had, in his words, dropped "to zero."

Why Scale Might Succeed Where Protest Hasn't

Washington's export controls work at the level of individual machines and chips: which lithography tool ships where, which processor crosses which border. China's plan is a bet that the fight doesn't have to be won at that level at all. If a country spends and builds fast enough, it doesn't need to close the technology gap chip for chip. It only needs to make the gap stop being decisive.

That bet is happening against a backdrop where bond markets are already nervous about how much money the global AI boom is burning through. China's $532 billion adds another enormous sum to that same question, but with a structural difference: the money is state-directed rather than raised from private credit markets, which changes who absorbs the loss if the bet underperforms. Researchers who study the plan note the targets are aggressive relative to current utilization, and that hitting 9,800 exaflops on schedule is far from guaranteed.

Knowlegic Perspective

Export controls are built on an assumption: that the country on the other end can't build a substitute, or at least can't build one fast enough to matter. That assumption holds when the target lacks capital, lacks state will, or lacks a domestic industry to subsidize. China, at this scale, lacks none of those three.

It's a pattern with precedent. Solar panels, high-speed rail, electric vehicles: in each case, Chinese industry started behind, absorbed years of losses underwritten by the state, and eventually became a dominant global supplier of technology it once had to import. A $532 billion plan applies the same playbook to computing power and chips, on a timeline of years rather than decades this time, because the pressure to close the gap is more immediate.

Whether it works depends on execution the plan itself can't guarantee: memory bottlenecks, utilization rates, and whether domestic chips can close the performance gap fast enough to matter before 2030 arrives. But the strategy itself is no longer in question. China isn't waiting for the chip ban to end. It's building around it, at a scale large enough to make the ban feel less like a wall and more like a toll it has decided to stop paying.

Sources & References

• China releases 5-year development plan for information, communications sector, State Council Information Office (2026)
• China's AI Infrastructure Plan: $295 Billion Project to Challenge US, Bloomberg (2026)
• China's AI computing plan targets sixfold growth and 100,000-card clusters, DigiTimes (2026)
• Forum: Technology in China's 15th Five-Year Plan, Stanford DigiChina (2026)
• Huawei and Cambricon lead China's AI server chip surge, piling pressure on Nvidia, South China Morning Post (2026)
• Huawei to double AI chip output in 2026, targeting 1.6 million dies, Invezz, citing Bloomberg (2025)