The Assault on Huawei

On the morning of December 1, 2018, a forty-six-year-old Chinese executive in a dark hoodie and white sneakers stepped off Cathay Pacific Flight 838 at Vancouver International Airport, expecting nothing more eventful than a layover on her way from Hong Kong to Mexico City. Meng Wanzhou was Huawei’s chief financial officer, a deputy chair of the board, and the eldest daughter of Ren Zhengfei, the secretive founder who had built her company from a Shenzhen telephone-switch reseller into the largest telecommunications-equipment maker on earth. She carried seven electronic devices, including an Apple MacBook and several iPhones. Officers of the Canada Border Services Agency were waiting on the jetway. They escorted her to a secondary inspection room and held her there for three hours. Royal Canadian Mounted Police officers then arrested her on a provisional U.S. extradition request. The Department of Justice in the Eastern District of New York wanted to charge her with bank and wire fraud, specifically with helping Huawei use a Hong Kong shell company called Skycom to evade American sanctions on Iran.

The arrest was the public surface of a long subsurface investigation, and it announced something Washington had been deciding privately for three years and was about to decide publicly. Huawei was not going to be allowed to continue rising. The president, the Pentagon, the intelligence services, and a small caucus of China hawks who had spent the Trump administration’s first two years ascending the National Security Council all agreed on this. They did not yet agree on the instrument. Meng’s arrest used one set of tools: the criminal code, the extradition treaty, the bank-fraud framework the United States had built to police Iran sanctions. Over the next eighteen months, a different set of tools would be assembled and tested on Huawei. The new tools would weaponize the geography of semiconductor manufacturing itself.

Beijing recognized the stakes immediately. Within nine days of the Vancouver arrest, Chinese state security detained two Canadian citizens, the diplomat Michael Kovrig and the entrepreneur Michael Spavor, on espionage charges no observer outside the Chinese system took seriously. The “two Michaels” would spend over a thousand days in prison, released within hours of Meng’s eventual return to Shenzhen in September 2021. The hostage diplomacy made the Huawei case unmissable in every capital that mattered. By the spring of 2019, when Donald Trump signed a national-emergency declaration about telecom equipment, the question was no longer whether the United States would move against Huawei. It was how thoroughly.

The first move came on May 15, 2019. Trump invoked the International Emergency Economic Powers Act and signed Executive Order 13873, which declared a national emergency in information and communications technology and directed the Commerce Secretary to block transactions deemed an “unacceptable risk” to national security. The order did not name Huawei in its text, but every reporter in the briefing room knew who it was about. The next morning, BIS published the operative document. Huawei Technologies Co., Ltd., along with sixty-eight non-U.S. affiliates spread across twenty-six countries, was being added to the Entity List with a presumption of denial. Any item subject to the Export Administration Regulations, including chips, design software, manufacturing equipment, and technical assistance, would now require a license to ship to Huawei, and the license was presumed refused.

Putting one of the largest companies in the world on the Entity List, named alongside North Korean front companies and Tehran’s nuclear procurement networks, was a categorical step. Inside Huawei’s Shenzhen campus, the news was received as confirmation of the worst-case scenario the company had been quietly planning for. Two days later, just past midnight on May 17, He Tingbo, president of Huawei’s chip-design subsidiary HiSilicon, sent an internal letter to her engineers that became, within hours, the most-shared corporate memo in the modern history of the Chinese internet. She wrote that for years HiSilicon had been preparing “spare tires,” backup chip designs based on alternate intellectual property and alternate suppliers, against the day all American technology became unavailable. That day had now arrived. Tonight, she said, every spare tire was being moved to the main wheels.

The letter scanned in Mandarin as a fight song. On Weibo, the hashtag pulled in over two hundred million views before noon. Inside HiSilicon, engineers who had spent a decade as the unglamorous supporting players of a company famous for its handsets and 5G base stations suddenly found themselves the moral center of a national struggle. Outside Shenzhen, in Hong Kong and Tokyo and Hsinchu, a quieter audience read the letter as a more delicate confession. The spare tires HiSilicon had been hoarding were mostly design IP. The fabrication of those designs into actual silicon still happened, almost without exception, in a single set of buildings on the western coast of Taiwan.

That set of buildings was TSMC. By 2019 Huawei was TSMC’s second-largest customer after Apple, accounting in some quarterly slices for as much as 14 percent of foundry revenue. HiSilicon’s flagship Kirin processors, the chips that gave the Mate and P-series phones their global competitiveness, were taped out in Shenzhen, masked at TSMC’s Hsinchu facilities, and fabricated in Tainan and Taichung. Huawei’s 5G base-station ASICs, the silicon that let the company underbid Ericsson and Nokia for telecom contracts in eighty countries, were TSMC products too. The first Entity List action did not, on its face, touch any of this. The Entity List restricted U.S.-origin items going to Huawei. A chip designed in Shenzhen and fabricated in Taiwan was a Taiwanese item. TSMC’s lawyers, after long conversations with American counsel, took the position that the May 2019 rule did not require them to stop shipping to their second-largest customer. They kept shipping.

This was the loophole the next year of policymaking would close. The China hawks who had pushed the Entity List action, including Pottinger at the NSC, Cordell Hull at Commerce, Keith Krach at State, and a small bench of trade lawyers who understood the export-control statutes the way most Washington appointees did not, recognized within weeks that the rule as written was leaky. Huawei’s American-component dependence was real but bounded. Its design dependence on American electronic-design-automation software, the toolchains made by Cadence, Synopsys, and Siemens-owned Mentor Graphics, was deeper. Its manufacturing dependence on American capital equipment, the deposition tools from Applied Materials, the etchers from Lam Research, the inspection systems from KLA, was deeper still. The hawks wanted a regulatory mechanism that reached through the Taiwanese fab and grabbed the American technology embedded inside it.

What they reached for was old, narrow, and obscure. The Foreign Direct Product Rule had been written into the Export Administration Regulations in 1959, in the early years of the strategic-technology controls the United States used against the Warsaw Pact. Its original purpose was to capture finished products made abroad with American technology, say a Belgian-fabricated valve made on an American lathe, and subject them in defined circumstances to American export licenses if they were headed to a Soviet-bloc destination. For sixty years the rule had been a footnote in the export-control toolkit, used sparingly against a handful of adversary destinations. It had never been used to wall off a single foreign company from the global supply chain. A working group inside Commerce, in close consultation with the NSC, spent the autumn of 2019 and the winter of 2020 quietly redrafting it.

The first version was published on May 15, 2020. Its language was technical and its effect surgical. From that date forward, any chip that was the direct product of certain American technology or software, and was designed by Huawei or HiSilicon, or was being produced for Huawei using semiconductor manufacturing equipment that was itself the direct product of American technology, required a license from BIS before it could be shipped. The presumption was denial. A grace period gave shipments already in the production pipeline 120 days to clear out. The drafters did not name TSMC. They did not name any specific fab. They wrote a rule that traced American technology forward through the supply chain wherever it had embedded itself, and relied on the fact, true of every leading-edge fab on earth, that the deposition, etch, and metrology equipment inside the cleanroom had at least one American manufacturer of record. TSMC’s Tainan fabs ran on American tools. So did SMIC’s. So did Samsung’s. So did every fab anywhere in the world that produced silicon below 28 nanometers.

Within a week, every general counsel in the global semiconductor industry had read the rule and reached the same conclusion. TSMC could not legally accept new orders from Huawei after May 15. Existing orders could continue to flow until mid-September. Anything beyond September would require a license, which would be denied. Mark Liu, TSMC’s chairman, addressed shareholders in early June with the controlled understatement that was his trademark. Any disruption from the Huawei loss would be short-term, he said. Other customers would absorb the freed capacity. The 5G build-out and Apple’s appetite would more than fill the hole. The remarks were intended to reassure investors. They were also a quiet acknowledgment that one of the most consequential business relationships in the modern semiconductor industry was over.

Inside Huawei, the response was a logistics scramble at industrial scale. Through the summer, the company turned its accumulated cash on hand toward the single goal of pushing as much silicon as possible across the 120-day grace window. HiSilicon engineers in Shenzhen finalized the masks for the Kirin 9000, a five-nanometer system-on-chip designed for the Mate 40 Pro flagship launching that fall. TSMC’s most advanced fab, in Tainan, dedicated extraordinary capacity to Huawei orders during June, July, August, and the first half of September. Trade-press reporting at the time described Huawei essentially renting TSMC’s 5-nanometer node for the duration of the window. By the time the deadline passed on September 14, TSMC had delivered, by various estimates, on the order of fifteen million Kirin 9000 dies, a stockpile meant to last as long as the Mate 40 Pro could be sold and Huawei’s premium phone business kept on life support.

The U.S. government understood the stockpile would happen. It also noticed something else. Through the summer, American intelligence and trade officials identified a pattern of secondary buyers, distributors in Hong Kong, holding companies in Singapore, third-party design houses in Taiwan and Korea, acquiring American semiconductors and reselling or repackaging them for Huawei. Whether these were arm’s-length transactions or organized circumvention varied case by case, but the pattern was unmistakable. The first FDPR was a wall around the front door. Money and parts were leaving by the back.

On August 17, 2020, BIS came back with the second rule. This one closed the back door. The amended FDPR no longer required that the chip be designed by Huawei. Any chip, designed anywhere, made with American technology, and known or suspected to be destined for Huawei or its affiliates, now required a license. BIS simultaneously added thirty-eight more Huawei subsidiaries to the Entity List, bringing the total above 150 entities across more than forty countries. The Temporary General License that had carved out narrow exceptions for legacy network maintenance was allowed to expire. From September 15 forward, the only chips that could legally reach Huawei were those produced and packaged before the deadline, those falling below the rule’s technical thresholds, or those whose American content was attenuated enough that the FDPR did not bite.

The September 15 cutoff was watched in Shenzhen with the precision of a hurricane forecast. Through the final weeks, Kirin 9000 wafers moved out of Tainan in batches and into Huawei’s bonded warehouses. By the time the deadline expired, the last shipments had cleared Taiwan. Yu Chengdong, the head of Huawei’s consumer business, had already conceded the inevitable in early August at the China Info 100 industry summit. “This year may be the last generation of Huawei Kirin high-end chips,” he told the audience. He called it a big loss. The Mate 40 Pro would launch in October as planned, but it would also, for the foreseeable future, be the last Huawei phone with a top-tier HiSilicon chip inside it.

The Mate 40 Pro launched on October 22 in Shanghai to a livestream audience that watched in something close to mourning. Reviewers in Europe and Asia treated it as the technical match of any flagship from Apple or Samsung. It had a curved display, a Leica-tuned camera array, satellite navigation chips comparable to the iPhone 12, and the full thermal headroom of TSMC’s 5-nanometer node. It was a product without a successor. By the time it reached store shelves in mid-November, Huawei had announced the sale of Honor, its mid-range sub-brand, to a state-backed consortium led by the Shenzhen municipal government and a domestic distributor called Digital China. The reported price was on the order of fifteen billion U.S. dollars in cash. The structure was a partial reverse-engineering of the sanctions framework killing Huawei: by ejecting Honor entirely from the corporate envelope, the new entity could in principle source American chips again. The speed and the political optics made clear it was a triage decision, not a strategic one. Huawei was choosing what to save.

The numbers fell faster than even pessimistic forecasters had projected. Huawei had entered 2020 holding roughly 17 percent of the global smartphone market and, in the second quarter, briefly overtook Samsung as the number-one shipper, the only time a non-Korean and non-American company had reached the top of the industry. By the fourth quarter, Huawei was sixth, shipping 33 million units, a 41 percent year-on-year decline. Through 2021, its consumer-business revenue dropped by roughly half. Total company revenue, which had grown 19 percent in 2019, contracted by nearly 30 percent in 2021. By 2022 Huawei’s global smartphone share had collapsed below 3 percent. The 5G handset business, which the company had targeted as the platform on which it would consolidate global leadership, ceased shipping new 5G phones in international markets entirely. Where Huawei had once been Samsung’s only credible global rival in premium Android, the field was now Apple, Samsung, and a fragmented Chinese pack of Xiaomi, OPPO, and Vivo, all of whom could still source American silicon because they were not on the Entity List.

What the cutoff did to a single company is one chapter of the story. What it did to the export-control framework is the larger one. The rule set written for Huawei in May and August 2020 became, almost immediately, a template. It demonstrated that the United States could draw a perimeter around a foreign firm and use the indirect leverage of American-origin tooling to seal that perimeter without cooperation from the firm’s home government. It demonstrated that the leverage worked even when the proximate manufacturer, TSMC in this case, was politically aligned with the United States and commercially indispensable to it. And it demonstrated that the leverage worked at a speed measured in weeks, not years. A Huawei chip line that had taken a decade to build was severed in two regulatory moves separated by ninety-four days.

That demonstration was studied closely inside the U.S. government, inside the Pentagon, inside the offices of the trade lawyers who advised TSMC and Samsung and ASML, and, with particular intensity, inside the strategy teams at Nvidia, AMD, and Intel. The Huawei perimeter had been drawn around handsets and 5G base stations. The technology that was clearly going to dominate the next decade, generative artificial intelligence trained on enormous clusters of accelerator chips, was beginning by 2021 to look like a much larger strategic vector than smartphones had ever been. The American chip designers who had spent the late 2010s shipping A100- and H100-class GPUs into Chinese data centers, including data centers operated by Huawei’s own cloud business, began a quiet conversation with their general counsel about how an FDPR rewritten around their products would behave.

It did not take long to find out. On October 7, 2022, BIS published a rule that extended the FDPR template, in modified form, beyond Huawei to a class of chips defined by performance thresholds. Any chip exceeding a defined combination of computational throughput and interconnect bandwidth, made anywhere in the world with American technology, would now require a license to ship to a Chinese customer for “advanced computing” or “supercomputer” end use. The presumption, again, was denial. The rule named no companies. It did not need to. Within forty-eight hours, Nvidia had announced that the A100 and H100 GPUs at the heart of the data-center business it had spent a decade building were no longer shippable to Chinese customers without a license. Within a month, the company had begun designing the down-binned A800 and H800 variants intended to slip beneath the rule’s thresholds. The thresholds would be tightened the following year. The pattern was, line for line, the Huawei pattern. The instrument was the same. The drafters were, in many cases, the same people. Only the target was different.

The Huawei case had answered a question the Japan trade war of the 1980s, the Sematech recovery of the 1990s, and the slow drift of leading-edge fabrication to Taiwan had all left ambiguous. The question was whether American semiconductor leverage was real or notional. In 1986, when Washington had told Tokyo it could not undersell American memory makers, the leverage had been quasi-political, a function of trade-agreement diplomacy and tariff threat, and the Japanese had absorbed it without losing the leading edge. In 2020, when Washington told a Chinese champion that the global fab base was no longer available to it, the leverage was material and immediate. A company that had projected itself, reasonably, as the future of mobile computing and 5G infrastructure was reduced inside eighteen months to a regional player struggling to ship phones at all.

The lesson was not lost on Beijing. By the autumn of 2020, Chinese industrial policy had quietly tripled its commitments to domestic semiconductor manufacturing capacity, design tooling, and equipment substitution. The Big Fund, the state-directed semiconductor investment vehicle seeded in 2014, accelerated its second tranche of disbursements. SMIC, the Shanghai-based foundry that had spent two decades trailing TSMC by roughly two process nodes, was instructed to find a way to a 7-nanometer process without the extreme-ultraviolet lithography that the Netherlands, under American pressure, would shortly stop shipping to it. Huawei, no longer permitted to buy advanced chips from anyone, began pouring resources into a clandestine domestic supply chain of partner foundries, equipment companies, photoresist suppliers, and EDA toolmakers that could in principle route around the American perimeter. The work was opaque, and for several years the results were poor.

In late August 2023 the work surfaced. A new Huawei flagship called the Mate 60 Pro went on quiet sale on the company’s website, with a Chinese-fabricated processor inside it that no Western analyst had thought possible on the timeline observed. Its reverberations would dominate the months that followed. The point here is narrower. Whatever else the cutoff had done, it had not stopped Beijing from spending the intervening three years building a domestic alternative, and the alternative had begun, against expectation, to ship.

What the Huawei episode established, beyond the fate of any single phone or company, was a precedent in the deepest legal-political sense. Through the summer and autumn of 2020, the United States had operationalized a doctrine: the global semiconductor supply chain, despite its surface appearance of distributed multinational autonomy, ran through American chokepoints dense and load-bearing enough that Washington could, with a single rule and ninety days’ notice, sever any foreign customer from the leading edge of computation. The doctrine was tested first on a Chinese telecommunications giant whose smartphone business was the most visible target available. It would be applied next, more carefully and against a vastly larger surface area, to the AI accelerator market the rest of the decade would be organized around. The instrument was the same. The drafters were the same. The signal Huawei sent through the system in those eighteen months, in the silent flow of Kirin 9000 wafers across the September 14 deadline, in Yu Chengdong’s apology at the Mate 40 launch, in the sale of Honor at a price set by panic, was the signal that defined what came after. The world the export controls would now operate on had been built. Its first occupant had been crushed inside it. The next ones, larger and more strategically central, were already being moved into position.