China's Sputnik Moment?

U.S. Commerce Secretary Gina Raimondo was sitting through a scripted state lunch in Beijing on August 29, 2023, when a small phone went on quiet sale on the website of a company she had spent the previous month telling Congress she had largely cut off. Huawei did not announce a launch event. There were no posters on the Wangfujing flagship store. The Mate 60 Pro simply appeared in the cart, listed at 6,999 yuan, with a paragraph of marketing copy claiming a “breakthrough” Kirin 9000s chip and what readers in technical forums quickly noticed was a 5G-class downlink speed under a 4G label. The phones sold out in hours. By the time Raimondo’s motorcade left the Diaoyutai State Guesthouse, Chinese social media was running side-by-side photographs of her Beijing schedule and the Mate 60 product page, captioned with variations on the same joke. State media called it a gift. The Global Times suggested, in a Saturday editorial, that perhaps the Secretary deserved a commission.

A week later, the Canadian reverse-engineering firm TechInsights opened one of the phones at its lab in Ottawa and confirmed what no Western intelligence agency had wanted to confirm. The Kirin 9000s system on chip had been fabricated inside the People’s Republic of China, by Semiconductor Manufacturing International Corporation, on a 7-nanometer-class process node SMIC labeled internally as N+2. The die measured roughly 107 square millimeters. It carried thirteen metal layers. Its transistor pitches and metal stack characteristics, on the imaging slides the firm circulated to its subscribers, sat noticeably close to the corresponding measurements on a TSMC N7 die from five years earlier. Neither machine had used an extreme ultraviolet scanner. The Dutch monopoly Washington had spent four years bricking off was not, on the evidence of the chip in front of TechInsights’ microscopes, the only way to reach 7 nanometers. SMIC had reached it by other means, under sanction, in volumes that suggested mass production rather than a one-off demonstration.

The reaction in Washington was incoherent. The Commerce Department asked for an investigation. The White House declined to draw a conclusion. Raimondo, in a House committee hearing six weeks later, called the chip “incredibly disturbing” and conceded her department had no evidence of “scale” production, a hedge that survived in the press for about as long as it took Bloomberg to estimate that SMIC’s 7-nanometer line was running somewhere between twenty-five and fifty thousand wafers a month. A senior Republican asked whether the export controls had failed. Raimondo answered that they had not, and that the United States needed to enforce them harder. The audio of that answer would replay in Beijing think tanks for the rest of the year, often with a second clip appended: a line from a People’s Daily editorial the previous October, in which a writer had recycled a phrase that had been circulating in Chinese policy circles since the spring of 2018. The Americans had handed the country its Sputnik moment. The country had decided, slowly at first and then with a sustained intensity that surprised even its own planners, to behave accordingly.

The phrase was not new. Chinese commentators had been reaching for the Soviet rocket since at least April 2018, when the Trump administration’s first ZTE penalty had stopped the Shenzhen telecom maker’s production lines for ninety days and exposed, on the front pages of every business paper in the country, how thoroughly Chinese national champions still ran on American silicon and American design tools. Xi Jinping had toured a Wuhan memory fab the same week the ZTE order landed, accompanied by Zhao Weiguo of Tsinghua Unigroup, the conglomerate then driving the country’s most aggressive bets in memory chips. Zhao, a heavyset man with the manner of a property developer, was photographed one step behind the General Secretary in a pose that would soon look very different. Within a month, Xi was delivering speeches in which “core technologies” were classified as instruments of state, and the phrase that would echo through the next six years of party documents, “self-reliance” or zili gengsheng, began doing the work it had done in the 1960s when Mao had used it to describe surviving without the Soviets. The historical analogy was deliberate. After ZTE, after Huawei in 2019, after October 7, 2022, after each escalation of what Washington called “small yard, high fence” and what Beijing called something less polite, the same line repeated. Sputnik. The country that had launched the satellite had won the panic that followed. The country that had been panicked into responding had won the decade after.

The first instrument was money, on a scale that shook even the analysts who had been tracking it for years. China had begun pouring state capital into its semiconductor industry in 2014, when the State Council approved the National Integrated Circuit Industry Development Outline and stood up the China Integrated Circuit Industry Investment Fund, the vehicle everyone, including its own staff, called simply the Big Fund. The first phase, registered on September 26, 2014, raised approximately 138.7 billion yuan, around twenty-one billion U.S. dollars at the prevailing rate. It funded SMIC, YMTC, Tsinghua Unigroup’s various memory ventures, and a long list of smaller equipment and design companies. The second phase, established on October 22, 2019, raised 204 billion yuan, about twenty-nine billion dollars, with the Ministry of Finance as the largest single shareholder. The third phase, announced on May 24, 2024, raised 344 billion yuan, roughly forty-seven and a half billion dollars, with a fifteen-year duration extending to 2039. State-owned banks, including ICBC and China Construction Bank, were directed to commit equity rather than the debt they had pushed into the earlier funds. The two largest existing Big Fund vehicles together would, when fully deployed, exceed a hundred billion dollars in equity capital aimed at a single industry.

The Big Fund was, if anything, the smaller part. CSIS researchers, working from Chinese government and industry filings, eventually estimated that aggregate central, provincial, and municipal subsidies and equity injections into the Chinese semiconductor industry had passed a hundred and fifty billion dollars by the early 2020s, with roughly two-thirds of that flow originating below the central level. Hefei, a second-tier city in Anhui Province that few Western policymakers could locate on a map, had become the largest city-government chip funder in the world, having bankrolled ChangXin Memory Technologies after its first effort to take a stake in Micron through a domestic vehicle was blocked by U.S. regulators. Wuhan funded YMTC. Quanzhou funded Jinhua before its dramatic implosion under American indictment. Shanghai funded SMIC and Hua Hong. Shenzhen funded HiSilicon, by way of Huawei, and a thicket of smaller fabless designers. The provinces competed on tax holidays and free land. By 2022, on industry counts that varied with definition, the country had registered tens of thousands of “semiconductor” companies, the great majority of them shell companies harvesting subsidies. The Caixin journalist Yu Hairong, covering the Big Fund’s anti-corruption purge that summer, put it plainly. The country had decided to flood the field, on the theory that water finds its own level and that the survivors would be enough.

The flood produced everything floods produce, including a dramatic share of failure and outright fraud. Wuhan Hongxin Semiconductor Manufacturing, known as HSMC, was the most spectacular of the early implosions. It had been founded in late 2017 by a trio of men with no semiconductor experience, one of whom appears to have used a pseudonym throughout the venture. The Wuhan municipal government had committed land and infrastructure. By June 2019 the trio had recruited Chiang Shang-yi, the seventy-three-year-old former chief technology officer of TSMC, as chief executive, and used Chiang’s reputation to extract several billion dollars in promised investment, hire several hundred former TSMC engineers at multiples of their Hsinchu salaries, and order an ASML deep ultraviolet immersion scanner that arrived in Wuhan in December 2019 amid televised fanfare. The single delivered tool was promptly mortgaged to a state-owned bank. By the summer of 2020, with no fab construction visible on the site and salaries unpaid, Chiang had resigned and returned to Taiwan. Local officials took the company over. The total reported loss ran between $18 billion and $20 billion, depending on which subsidies were counted. A Caixin Global investigation in March 2021 listed five additional projects of similar provenance that had collapsed the same way over the previous eighteen months. The Big Fund’s own management arm, Sino-IC Capital, came under formal anti-corruption investigation in July 2022, with longtime chief executive Ding Wenwu and his deputy Lu Jun both detained. The political message was clear. The flood would continue, but the rakes were on.

For all the visible waste, the survivors learned to swim. SMIC, the Shanghai foundry Richard Chang had founded in 2000 with a mix of Taiwanese, mainland, and overseas Chinese capital, had been pushed by the cumulative weight of U.S. controls into building a parallel, sanctions-resistant version of the modern chip industry. Its central engineering bet was made in October 2017, when it hired Liang Mong-song as co-chief executive. Liang was sixty-five, a Taiwanese process engineer of legendary stamina who had been one of TSMC’s “six knights” of R&D in the 1990s, had stormed out after being passed over for a senior promotion in 2009, had spent the Samsung years that followed catapulting Korean foundry yields from behind to ahead of his old employer at the 14-nanometer node, and had been the subject of a long civil suit by TSMC over trade secrets. He arrived at SMIC with a caravan of disciples. He raised the company’s struggling 14-nanometer yield from a few percent to over ninety inside ten months. Then he refused, on grounds of process feasibility under sanctions, to use any technology that required EUV. He pushed the company to develop what he called the N+1 node and then the N+2 node using deep ultraviolet immersion lithography pushed to its limits with quadruple patterning, a punishing technique in which a single chip layer was decomposed into four overlapping exposures and reassembled by alignment. The yields were poor. The wafer cost was reportedly forty to fifty percent above TSMC’s at the comparable node. The throughput per scanner was a fraction of TSMC’s. The cycle times were long and the rework rates were high. None of that mattered, in the only sense that mattered, which was that the line was running.

In Wuhan, Yangtze Memory Technologies, founded in July 2016 with backing from Tsinghua Unigroup, the Hubei provincial fund, and the Big Fund, had taken the same approach in 3D NAND flash. YMTC’s first chief executive was Simon Yang, a Chinese-American engineer who had spent more than a decade at Intel and had been a founding employee of SMIC in 2001 before returning to mainland projects in the 2010s. The company’s signature innovation, a die-bonding approach called Xtacking, took the array transistors and the peripheral CMOS logic of a NAND chip, fabricated them on separate wafers, and bonded them at the wafer level, allowing aggressive interface speeds without redesigning the underlying memory cell stack. By August 2022 YMTC had unveiled, at the Flash Memory Summit in Santa Clara, the X3-9070 NAND chip with 232 active layers and Xtacking 3.0. Western analysts who had assumed the company was at least two generations behind looked at the data and revised their notes. Four months later, on December 15, BIS added YMTC to the Entity List, citing its connection to the Huawei supply chain. Equipment vendors froze. American engineers stationed in Wuhan flew home. A senior YMTC executive told Caixin the company would “build with what we have,” and YMTC set about replacing the cut-off equipment with domestic alternatives from Naura, AMEC, ACM Research, Piotech, and a thicket of smaller suppliers, several of which had multiplied their valuations on the news. By 2024 some of those companies were in the global top twenty by sales of certain categories of chip equipment, on Japanese trade group counts. None could replace ASML, but at the lower-end nodes that constituted most of the world’s wafer demand, that was beginning to look beside the point.

ChangXin Memory Technologies, the Hefei DRAM start-up funded by the city government after the Micron acquisition path was blocked, took the third leg of the indigenization strategy. It was founded in May 2016 with GigaDevice founder Zhu Yiming as operating chief and a pool of returnee engineers, several from Inotera, the Taiwanese DRAM joint venture Micron had absorbed. By 2020, CXMT was in mass production of DDR4 and LPDDR4 on a 19-nanometer node. By 2024, it had moved to 17 nanometers and was sampling a node it called G4, with an LPDDR5 product line aimed at Chinese smartphone makers eager to escape Samsung and SK Hynix dependencies. It was not technologically competitive with the Korean leaders. It did not need to be. The cost-driven mature DRAM market in mainland China was large enough on its own to absorb a producer with imperfect performance and tolerable yield, and the company moved by mid-decade to add a third Hefei fab.

The talent layer running underneath the money was, by 2024, the part of the strategy Western analysts had been slowest to acknowledge. The Thousand Talents Plan, the program initiated in 2008 to recruit overseas Chinese scientists and engineers back to mainland labs and faculties, had been formally retired in 2018 after the FBI began investigating American researchers who had taken its grants. It was replaced, by 2019, with the less-publicized Qiming program operated by the Ministry of Industry and Information Technology, with packages including home-purchase subsidies, several million yuan in signing bonuses, and explicit semiconductor focus. Below the central programs sat a cascade of provincial and municipal counterparts, often deeper-pocketed and faster-moving. Recruiters for the various funds worked Taiwan especially hard. The Hsinchu engineering community, in which the founding generations of TSMC, UMC, and MediaTek had grown up together and which still constituted a large fraction of the world’s actively employed leading-edge process engineering talent, became the principal poaching ground. A Caixin Global investigation in August 2020 documented over a hundred TSMC engineers who had moved to Quanxin Integrated Circuit Manufacturing, HSMC, and other mainland ventures over the previous two years, often through Korean or Singaporean staffing intermediaries that let recruits nominally remain employed offshore while they relocated their families to Wuhan, Hefei, or Shanghai. By 2022, Taiwan’s Investigation Bureau had begun publicizing prosecutions, and the Mainland Affairs Council had announced a tightening of cross-strait labor rules. Recruiters adapted: routing through third countries, formally hiring Taiwanese engineers as Singaporean or Hong Kong employees, accelerating the substitution of returnee mainland engineers from American and European graduate programs into senior process roles. By the most cited Taiwanese estimates, more than three thousand Taiwanese chip engineers had relocated to mainland projects in the years after 2018.

The cumulative effect was visible by mid-decade in places that had nothing to do with leading-edge logic. China’s domestic share of mature-node wafer capacity, in the segment that produced the analog, microcontroller, and power management chips that ran cars, factories, washing machines, and increasingly artillery and missile guidance, climbed from under fifteen percent in 2018 to roughly thirty percent by 2024 on Boston Consulting Group counts, with credible projections from BCG, ITIF, and Japanese trade groups that the share would pass forty percent before the end of the decade. Etch and deposition tool localization rates, the share of equipment in Chinese fabs sourced from Chinese vendors, climbed past forty percent by 2024. Localization was still poor in lithography, metrology, ion implantation, and EUV photoresists, all of which remained chokepoints owned outside China. But the trajectory at the trailing nodes was unmistakable. The world’s largest chip-buying market was building a parallel, indigenously sourced version of the bottom three-quarters of its own demand stack.

What none of the catalogued spending and recruiting and localization could yet do, by the moment Raimondo flew home in 2023, was reach the leading edge under sanction. SMIC’s 7-nanometer line was running, but it was running expensive, slow, and at modest volume. Domestic alternatives to ASML’s leading-edge DUV scanners existed only in research-line prototypes, with Shanghai Micro Electronics Equipment, the Chinese national lithography champion, still struggling to qualify a production-grade 28-nanometer immersion tool the rest of the world had been shipping since 2010. EUV remained the harder choice. Without an EUV machine, no Chinese fab could move below the 5-nanometer node on the kind of yield that would let it compete in volume against TSMC’s mature 3-nanometer process. The wager Liang and his colleagues at SMIC and YMTC and CXMT were running, with the Big Fund’s money behind them and the Communist Party’s patience above them, was that the stack of multi-patterning improvements, equipment substitution, design optimization, and architecture innovations would compound fast enough to keep the Chinese frontier within one or at most two generations of the world’s, until either a domestic EUV scanner shipped or the geopolitical environment changed. Outside observers parsed the wager differently. Dan Wang, writing his end-of-year letter from Yunnan in early 2023, argued that the U.S. controls had handed Beijing the political license to do something the Chinese state had been incapable of doing on its own, namely to force its state-owned banks, provincial governments, and undisciplined venture funds to behave like a single industrial planner over a decade-long horizon. Gregory Allen at CSIS argued the opposite, that the controls had bought the United States a five-to-ten-year window in which Chinese leading-edge supply remained capped, and that the question that mattered was whether Washington could use that window. Both could be right. Most of the Chinese system was indifferent to which.

What was certain, by the time the Mate 60 Pro’s launch had completed its slow propagation through Western analyst notes and Western political reactions, was that the country that had been the source of the Sputnik panic in 1957 had now produced one of its own. The phrase had begun in Chinese policy circles as a self-applied warning. By the end of 2023 it was being used in Washington, with the same nervous edge the original had carried in Eisenhower’s White House, to describe a country that had decided, while the United States was still arguing about whether sanctions worked, to do the long, expensive, partially corrupt, partially cynical, partially heroic work of building most of the chip industry over again on its own terms. The work was nowhere near finished. On the evidence of the wafers running through Shanghai and Wuhan and Hefei, the work was no longer in the realm of speculation. The next chapters of the chip war, in Beijing’s reading, were going to be about whether the country that had been told it could not have the most advanced silicon would simply have it anyway, given enough years and enough yuan and enough engineers willing to fly to where the engineering was being done.