The U.S. Banned China From Buying Advanced Chips. China Spent $41 Billion on Chip Equipment Anyway.

August 29, 2023. A year and two weeks after the most aggressive technology export controls in a generation, Huawei dropped the Mate 60 Pro in China with zero advance marketing. Teardowns confirmed what U.S. officials feared: inside was HiSilicon’s Kirin 9000s processor, built on SMIC’s 7-nanometer process. The chip Washington said China couldn’t make.

Nobody at the Bureau of Industry and Security held a press conference about that.

The Controls: A Timeline

Start here, because the sequence matters.

October 7, 2022: The Biden administration drops the hammer. New rules ban exports of advanced chips (14nm and below for logic, 128-layer+ for NAND), chip-making equipment, and—crucially—the services of U.S. persons working at Chinese fabs. This last provision is the real weapon: American engineers at SMIC and YMTC are told to stop working or lose their citizenship protections. Dozens leave within weeks.

January 2023: Japan and the Netherlands sign on. ASML, the sole manufacturer of extreme ultraviolet lithography machines, begins restricting shipments of its most advanced systems. Tokyo Electron and other Japanese equipment makers follow. The controls become trilateral.

October 2023: Round two. The Commerce Department closes loopholes, restricting additional chip-making tools, expanding the Entity List, and tightening the threshold for what counts as “advanced.” Controls now cover some older DUV lithography systems, not just EUV.

December 2024: Round three. 140 additional Chinese entities added to the Entity List. New restrictions on high-bandwidth memory (HBM) chips—the components Nvidia GPUs need for AI training. Export controls on quantum computing equipment. The net keeps tightening.

That’s the offense. Now the scorecard.

What the Controls Achieved

Metric	Status	Assessment
EUV lithography blocked from China	✅ No EUV systems shipped since 2022	Clear win
Advanced AI chip access cut	⚠️ Nvidia designed H20/L20 compliance chips; China stocks up	Partial
Leading-edge node production stopped	❌ SMIC produced 7nm, possibly 5nm	Failed
Chinese equipment spending reduced	❌ China became #1 global buyer in 2024	Backfired
Multilateral alignment	✅ Japan, Netherlands cooperating	Win
Chinese self-sufficiency slowed	❌ Massively accelerated	Backfired

Two clear wins. One partial. Three backfires. That’s the honest math.

The $41 Billion Spending Spree

The semiconductor equipment market has a peculiar structure: a handful of companies in the U.S., Japan, and the Netherlands make the machines that make the chips. When you announce that a customer can’t buy your products starting next quarter, that customer buys everything it can this quarter.

China’s semiconductor equipment purchases in 2024 hit an estimated $41 billion, according to SEMI industry data—making it the world’s largest market for fab equipment, surpassing South Korea and Taiwan combined. That’s up from roughly $25 billion in 2022 and $30 billion in 2023. Every round of controls triggered a pre-compliance purchasing frenzy.

ASML itself told investors the story. In 2023, China accounted for 29% of ASML’s total bookings. By the first half of 2024, China was 49% of the company’s revenue—the largest single market. ASML shipped every DUV system it legally could before the tightened Dutch export rules took effect in January 2024. The machines China received—NXT:2050i and older DUV immersion systems capable of multi-patterning to reach 7nm—are now installed in Chinese fabs that are expanding capacity as fast as they can train operators.

The equipment makers’ stock prices tell you everything. Applied Materials, Lam Research, KLA, Tokyo Electron—all hit record highs during 2024. Their best customer was the country they were supposed to be cutting off.

SMIC: The 7nm Nobody Expected

SMIC was on the Entity List before the October 2022 controls. It had been there since December 2020. On paper, it was restricted to 14nm production. In practice, it figured out how to use DUV multi-patterning—running the same wafer through a lithography machine multiple times with different masks—to achieve effective 7nm feature sizes.

This is brutally inefficient. Where TSMC uses a single EUV exposure for critical layers, SMIC uses four or more DUV passes. Yields are low. Costs are high. Throughput is a fraction of what TSMC achieves at the same node. SMIC cannot mass-produce at 7nm the way TSMC mass-produces at 3nm.

But it can produce enough. The Kirin 9000s shipped in millions of Mate 60 Pro units. It’s not competitive with Apple’s A17 Pro, but it doesn’t need to be. It needs to be good enough for the domestic market. And it is.

Reports in December 2025, first from Reuters, indicated that SMIC or a related entity in Shenzhen had completed a prototype EUV lithography machine—the technology the entire sanctions architecture was designed to deny. Working production chips from a domestic EUV system are estimated at 2028–2030. That timeline would have seemed absurd in 2022. It doesn’t anymore.

The Domestic Lithography Race

In September 2024, China’s Ministry of Industry and Information Technology promoted two new deep ultraviolet lithography machines developed domestically by Shanghai Micro Electronics Equipment (SMEE). One operates at 193nm wavelength with sub-65nm resolution and 8nm overlay accuracy. The second runs at 248nm with 110nm resolution.

Neither threatens ASML today. ASML’s NXT:2050i achieves 13nm half-pitch; SMEE’s best machine reaches roughly 65nm. That’s a generation behind in physics terms. But the gap was infinite in 2020—China had no production-grade lithography at all. Going from “nothing” to “one generation behind” in four years on the most complex manufacturing equipment humans have ever built is not failure. It’s the most expensive physics crash course in history, and it’s moving faster than the U.S. intelligence community’s consensus estimates.

Big Fund III: The $47.5 Billion Bet

In May 2024, China launched the third phase of its China Integrated Circuit Industry Investment Fund—commonly called Big Fund III—at 344 billion yuan ($47.5 billion). That’s larger than Big Fund I ($21B, 2014) and Big Fund II ($29B, 2019) combined.

The money is pointed at equipment and materials. Not chip design. Not fab capacity. Equipment. China’s leadership drew the obvious lesson from the sanctions: the chokepoint isn’t the chip. It’s the machine that makes the chip. Big Fund III is an attempt to build domestic versions of every tool in the semiconductor supply chain—lithography, etching, deposition, metrology, inspection—so that future sanctions have nothing left to sanction.

Whether $47.5 billion is enough is debatable. ASML alone spent $4 billion on R&D in 2023, building on three decades of accumulated engineering. You can’t buy institutional knowledge. But you can hire engineers, and China is producing 8 million STEM graduates per year. The talent pipeline is not a bottleneck.

The AI Chip Problem

The sharpest bite of the controls is in AI training hardware. Nvidia’s H100 and its successors remain the dominant platform for large language model training. Each chip contains billions of transistors at TSMC’s 4nm node, paired with SK Hynix’s HBM3 high-bandwidth memory. China cannot manufacture either domestically.

Nvidia responded to the October 2022 controls by designing compliance chips—the A800, then the H20 and L20—deliberately hobbled to fall under the export thresholds. China bought them by the hundreds of thousands. When the December 2024 rules tightened HBM restrictions, even these compliance designs faced scrutiny.

Meanwhile, Huawei’s Ascend 910B AI accelerator—manufactured at SMIC on 7nm DUV—shipped to major Chinese tech companies including Baidu, Alibaba, and ByteDance. Benchmarks suggest it reaches roughly 60–70% of the H100’s performance. Again: not competitive at the frontier. But training models with 30–40% more chips and 50% more time still produces functional AI systems. China’s AI labs are not standing still. DeepSeek’s R1 model, released in January 2025, demonstrated frontier-competitive performance reportedly trained on Ascend hardware.

The Scorecard Nobody Wants to Publish

Goal	2022 Expectation	2026 Reality
Deny EUV access	Permanent chokepoint	Working prototype by 2025; production 2028–30
Cap at 14nm	Hard ceiling	7nm in production, 5nm rumored
Slow equipment buildup	Constrain capacity	$41B spent in 2024 alone; record pace
Maintain AI compute gap	5+ year advantage	~18 months and closing
Rally allies	Multilateral regime	Japan/Netherlands on board; compliance uneven
Deter self-sufficiency drive	Too expensive to replicate	$47.5B Big Fund III; 8M STEM grads/yr

The controls delayed China. That is real. SMIC’s 7nm yields are poor. Its throughput is a fraction of TSMC’s. Domestic lithography is a generation behind. The AI compute gap exists. Without the sanctions, Huawei would be buying ASML EUV machines openly and SMIC would be at 5nm by now.

But delay is not denial. And the controls did something the architects may not have intended: they gave China’s entire semiconductor establishment a unified, existential purpose. Before October 2022, domestic equipment makers were marginal players struggling against ASML, Applied Materials, and Lam Research in open competition. After October 2022, they became national security priorities backed by the largest state semiconductor fund in history, guaranteed domestic customers, and an engineering workforce that treats catching up as a matter of civilizational survival.

The Sputnik analogy gets overused. But when you tell a country with the world’s second-largest economy, 8 million STEM graduates per year, and $47.5 billion in fresh capital that it cannot have a technology—you are placing a very specific bet. You are betting that the technology is too hard to replicate within the window that the sanctions buy you. And you are betting that you will use that window to do something decisive with your advantage.

Three years in, the sanctions bought a window. Nobody in Washington can tell you what it was used for.