The October Seventh Earthquake

On the second Friday morning of October 2022, sometime before sunrise on the East Coast, a small staff at the Bureau of Industry and Security finished a document they had been writing in increments since spring. It ran to roughly 140 pages of dense regulatory text, a length that would force the Federal Register’s typesetters into an unusually long publication queue the following week. It carried the bureaucratic title Implementation of Additional Export Controls: Certain Advanced Computing and Semiconductor Manufacturing Items; Supercomputer and Semiconductor End Use; Entity List Modification. Almost no one outside Washington’s small export-control bar would ever read it through. Almost everyone in the global semiconductor industry would, within a week, be working from someone else’s summary of it. The drafters had not given it a public name. The industry settled on one within a few news cycles. They called it October 7.

The rule was not, technically, a single rule. It was a stack. The drafters at BIS, working under Under Secretary Alan Estevez and Assistant Secretary Thea Kendler, had bundled nine distinct regulatory actions into one interim final document, with effective dates staggered across two weeks to give the supply chain time to absorb the structure. Some pieces took effect immediately on October 7. Some kicked in five days later, on October 12. The most extraterritorial provisions, the ones that reached through the foreign fabs and equipment makers of allies into territory no American export rule had previously claimed, would not bite until October 21. Inside that two-week corridor, every general counsel at every chip company in the world from San Jose to Hsinchu to Eindhoven to Suwon to Shanghai would spend long evenings reading the document with a thumb in the threshold tables, calling colleagues at three in the morning local time, and concluding that the world they had been operating in had changed.

The first test was logic chips. The rule defined a “production technology node” of 16 or 14 nanometers as the inflection point. Any logic integrated circuit fabricated using a non-planar transistor architecture, the rule’s clean way of saying FinFET or the gate-all-around designs that would succeed it, at that node or below, was now controlled when destined for a fab inside the People’s Republic of China. The choice of node was not arbitrary. It was the threshold at which leading-edge mobile and data-center chips began. It was where TSMC, Samsung, Intel, GlobalFoundries, and SMIC’s most advanced lines lived. It was, by no coincidence, the node SMIC had been struggling to push toward through the early 2020s and had only just achieved at modest yield. Setting the line at 14 nanometers froze China at the technology level it had reached, on paper, in late 2019. Anything below that line was now off-limits at the fab level.

Memory got its own thresholds, set against the same logic. DRAM was capped at 18 nanometers half-pitch and below, the node range Samsung and SK Hynix and Micron had been shipping since the late 2010s and the line that ChangXin Memory in Hefei had been racing toward. NAND flash was capped at 128 layers and above, an inflection that excluded older 64-layer and 96-layer products and explicitly targeted the modern 3D NAND stacks that Yangtze Memory in Wuhan had built its company around. The numbers were specific enough to draw a precise perimeter. Below the line, business as usual. Above the line, license required, with a presumption of denial.

Logic chips and memory chips had been controlled before, in narrower forms. The genuinely new instrument in the document was the section that defined “advanced computing” integrated circuits in performance terms, decoupled from process node entirely. The rule created a pair of new entries on the Commerce Control List, designated 3A090 and 4A090, that swept up any chip whose performance in tera-operations per second crossed a defined threshold and whose I/O bandwidth crossed a second one. The bandwidth threshold was 600 gigabytes per second of bidirectional interconnect. The TOPS threshold, calculated against an explicit bit-length multiplier that would later become known to engineers as Total Processing Performance, came out at 4,800. The two had to be met together. Below either, a chip fell outside the new category. Above both, a chip required a license to ship to China, and, depending on its end use, the license would be denied.

The thresholds had been chosen with a single product line in mind. Nvidia’s A100 data-center GPU, released in 2020 and the workhorse of every serious large-language-model training cluster outside Beijing, sat above both lines. So did the H100, the next-generation Hopper card Nvidia had launched in March 2022 and was preparing to ship in volume through the fall. So did AMD’s MI250 and the cards Intel had begun designing in its Habana acquisition. The rule did not name any of them. It did not need to. Within forty-eight hours of the announcement, Nvidia had quietly told its Chinese sales channel that A100 and H100 shipments to China were halted as of October 21, when the chip-level controls bit. The company’s lawyers had been expecting the move since August 26, when an earlier letter from BIS, drafted under similar legal authority but applied as a one-off to A100 and H100 only, had triggered a Form 8-K filing warning investors of up to four hundred million dollars in lost Chinese revenue. The October 7 rule was the same medicine, generalized.

What the August letter had been to Nvidia alone, the October 7 rule was to the entire advanced computing market, and to the equipment that produced it. The fourth and most legally inventive piece of the document was a new extension of the Foreign Direct Product Rule. The 2020 Huawei FDPR had reached through foreign fabs to capture chips made with American technology and headed to a single named company. The October 7 version, drafted by some of the same lawyers, generalized the logic: any 3A090 or 4A090 chip anywhere in the world, made with American semiconductor technology or software, and known or believed to be heading for a Chinese customer for an advanced-computing or supercomputer purpose, now required a license. The presumption was denial. The rule pulled the entire global GPU and advanced-AI accelerator market into the American export-control envelope without naming a single company.

It went further on the equipment side. The rule added a new tranche of semiconductor manufacturing items to the Commerce Control List, including specified deposition, etch, lithography, doping, and metrology tools, with license requirements for shipment to PRC-located fabs producing logic at 14 or 16 nanometers and below, DRAM at 18 nanometers half-pitch and below, and NAND at 128 layers and above. The thresholds matched the chip-level rules. The intent was to reach the supply chain on both sides. China would not be able to buy the most advanced chips. China would also not be able to buy the equipment to make them.

The fifth element in the document was the one no foreign capital had ever seen before. It was called, in the regulatory shorthand, the U.S. persons rule. Buried in the document under license-requirement language, it imposed a license requirement on any U.S. person, defined to include U.S. citizens, U.S. permanent residents, and U.S. corporations, who provided support of almost any kind to the development or production of integrated circuits at a Chinese fabrication facility producing logic at 14 nanometers and below, DRAM at 18 nanometers half-pitch, or NAND at 128 layers and above. Support was defined broadly. It included shipping items, transmitting items, transferring items, providing services, providing technical assistance. It applied wherever the U.S. person was located. A green-card holder working for a Chinese company in Shanghai now needed a Department of Commerce license to keep doing the job he had been doing the previous afternoon, and the license, the rule made clear, would generally not be issued.

The provision had been written with a specific population in mind. Yangtze Memory’s senior process team, a group whose breakthroughs on the Xtacking 3D NAND architecture had pushed the company past two hundred layers ahead of even Samsung’s road map, contained a substantial cluster of American passport holders and green-card residents, several of them returnees to mainland China after long careers at Intel, Lam Research, Applied Materials, and SanDisk. ChangXin Memory in Hefei had a similar core, smaller but consequential. Shanghai’s collection of mid-tier fabs and the various centers around the SMIC system carried more. By the most credible industry counts, somewhere on the order of several dozen senior U.S.-passport engineers, and a larger penumbra of American green-card holders and dual nationals, were embedded at the working levels of the Chinese leading-edge fabs on October 7, 2022. The U.S. persons rule, on its face, gave each of them a single decision. Stop supporting development at their employer or stop being employed.

The rule’s effective date for the U.S. persons provision was October 12, five days after announcement. The decisions, in practice, did not wait that long.

By the morning of October 8, internal Slack channels at the equipment vendors with embedded service teams in Chinese fabs lit up. Applied Materials’ Chinese support staff, the engineers who lived in Wuhan and Hefei and Shanghai supporting installed deposition tools, were summoned to local conference calls and told, with whatever delicacy management could muster on twelve hours of legal notice, to stop providing service that fell within the rule’s definitions. Lam Research’s etch service teams received parallel instructions. KLA’s metrology engineers were pulled. The withdrawal cascaded through the Chinese leading-edge fab system like a power cut. A senior engineer who had been calibrating a chemical vapor deposition tool on a 232-layer NAND line in Wuhan on Friday afternoon was, on Monday morning, sitting in his apartment waiting for legal guidance. By the end of the following week, a story had begun circulating in the Chinese semiconductor press that became one of the rule’s defining anecdotes, regardless of how many times it could be independently verified: a YMTC production line had gone idle because the American engineers on whose presence its calibration depended had walked out, and because the equipment makers’ service contracts had been suspended pending license review.

Yangtze Memory itself moved within days. Internal communications instructed U.S. citizens and green-card holders in core technology positions to stop work, with the company terminating contracts on October 24 for those whose roles fell within the rule’s perimeter. The quiet question, which Chinese commentators in business publications hesitated to put plainly, was how many of these people there had been to begin with. The number was politically uncomfortable in both directions. A senior YMTC engineer told Caixin that a portion of the company’s NAND breakthroughs had been led by American team members, and that those team members were now leaving the building. Apple, which had been close to designating YMTC as a NAND supplier for a tranche of iPhones, had already drawn the same conclusion through its own legal-risk review, dropping the design-in by the end of the same month under what it called political pressure but was, more accurately, a sober reading of the same regulatory text. ChangXin Memory in Hefei went through a parallel scrub of its U.S.-passport personnel. SMIC moved more slowly because its U.S.-citizen exposure at the leading edge was smaller, but its Shanghai system reorganized within weeks to route around the gaps.

The departures rolled out beyond the fabs themselves. American directors sitting on the boards of Shanghai-listed Chinese chip-equipment companies resigned within the month. American executives running Chinese subsidiaries of multinationals filed restructuring paperwork to put non-U.S. nationals in positions that touched the regulated activities. Recruitment networks that had spent half a decade pulling Taiwanese, Korean, and overseas Chinese engineers into the mainland chip system suddenly faced an inverse problem. The most senior people they had recruited from Silicon Valley and Hudson Valley, including process veterans from Intel and Micron whose value to the mainland fabs had been precisely their decades of leading-edge experience at American-headquartered companies, were now liabilities. Several of them were back on planes to California within days. A handful of others stayed in China, gave up their U.S. passports, and accepted the consequences. Most simply left.

The rule’s drafters had anticipated this. Inside the National Security Council, the man who had quarterbacked the inter-agency policy process across most of 2022 was Tarun Chhabra, the deputy assistant to the president and coordinator for technology and national security. Chhabra had spent the spring and summer holding meetings between Commerce, Defense, State, Energy, and the intelligence community, hammering out a structure that could survive the federal interagency’s tendency to dilute. The political ceiling above him was the National Security Advisor, Jake Sullivan, who had been signaling the underlying doctrine in public since at least midsummer. On September 16, three weeks before the rule went out, Sullivan had given a speech at the Special Competitive Studies Project’s Global Emerging Technology Summit that said, in a single line, what the rule that was about to be published was meant to do. The United States, he said, had previously maintained a sliding-scale approach in which it sought to stay only a couple of generations ahead of competitors in critical technology. That, he continued, was no longer the strategic environment. Given the foundational nature of certain technologies, advanced logic and memory chips among them, the United States would have to maintain as large a lead as possible.

The phrase mattered. For thirty years, American technology policy toward China had rested on a polite fiction: that the United States was content to remain ahead but did not actively seek to deny the trailing competitor its own progress. That fiction had let the chip industry globalize, allowed TSMC to become the world’s foundry, and let Beijing plan its industrial strategy through the late 2010s on the assumption that with enough money and patience the gap would close. Sullivan’s September speech had retroactively announced the doctrine the new rule would express. The United States was no longer trying to lead. It was trying to widen the lead by holding the competitor in place. In Washington the framing was small yard, high fence. In Beijing the framing for the same instrument was less polite.

Inside the Department of Commerce, the public face of the rule belonged to Estevez, the Under Secretary, and Kendler, the Assistant Secretary for Export Administration. Estevez was a career Defense Department official, a logistics specialist who had spent decades inside the Pentagon’s acquisition machinery before crossing the river to BIS. He carried himself with an institutional steadiness that did not look like an architect of geopolitical earthquakes. He spoke, in the press release that accompanied the rule, of his “north star” at BIS being to ensure the agency was doing everything in its power to prevent sensitive technologies with military applications from being acquired by Chinese military, intelligence, and security services. Kendler, a former Justice Department prosecutor who had built export-control cases against Iranian and Chinese procurement networks during the Trump administration before returning to a senior policy seat under Biden, took the lead on technical communication. She told reporters and industry, repeatedly across the following weeks, that the People’s Republic had poured resources into supercomputing and was using those capabilities to monitor and surveil its own citizens and to fuel its military modernization, and that the rule was a response calibrated to that diagnosis. Her phrasing was deliberate. She did not call China an adversary in the Cold War sense. She defined a specific set of technological capabilities and a specific set of end users and drew the perimeter precisely around the intersection. The body of the rule did the same.

Above all of them sat Gina Raimondo, the Secretary of Commerce, who had spent the previous eighteen months running the political machine that produced the CHIPS and Science Act. Raimondo was a former venture-capital partner and the former governor of Rhode Island. She was a politician, not a regulatory technician, and the October 7 rule was, in the public framing she chose, a national-security action she had endorsed and her Bureau had executed. In remarks delivered the following month at a Massachusetts Institute of Technology Technology Review event, she described the rule’s logic in plain terms. The previous American export-control posture, she said, had been reactive. The new posture was strategic. The intent was to deny China the ability to manufacture or acquire the chips at the heart of large-scale AI training and advanced military modernization, full stop, without pretending the goal was anything else.

Outside Washington, the document landed with roughly the consequences its drafters had wanted, plus several they had not. The morning Hong Kong markets opened after the announcement, Chinese chip-related equities dropped sharply, with the Hang Seng tech sub-index closing the week down on the order of several percentage points. The People’s Daily and the Global Times denounced the move within forty-eight hours, with editorial language that ranged from procedural objections about the unilateral character of the rule to broader claims that the United States was abusing the concept of national security to suppress Chinese technological development. Foreign Ministry spokesman Wang Wenbin, in scheduled press conferences over the following two weeks, used variations of the same line: that the United States was politicizing and weaponizing economic and scientific issues. The denunciations were calibrated. There were no immediate retaliatory tariffs, no parallel Chinese export-control announcements, no industrial nationalizations. The Chinese policy elite was watching, taking notes, and deciding what counter-instruments to develop on what timeline.

The reaction in capitals that mattered for the rule’s enforcement was less performative and more anxious. The Foreign Direct Product extension reached through fabs in Taiwan and Korea and through equipment makers in the Netherlands and Japan. None of those allied governments had been formally consulted in the rule’s drafting, although senior American officials had briefed foreign capitals across August and September on the doctrine. The Dutch government, which had to decide whether ASML’s deep-ultraviolet lithography sales to SMIC fell under its own emerging export-control regime, found itself in immediate negotiation with Washington over what a coordinated allied posture would look like. Tokyo, which oversaw Tokyo Electron and Screen Holdings and Nikon and Canon, faced parallel decisions. Both governments, after months of negotiation that culminated in a tripartite understanding announced in late January 2023, would adopt national export-control regimes broadly aligned with the U.S. semiconductor manufacturing equipment provisions, although without formally adopting the U.S. persons rule or the Foreign Direct Product extension. The diplomatic price of the unilateral October 7 publication was several months of bilateral and trilateral repair work in The Hague and Tokyo. The American officials who had pushed the rule out without waiting for allied alignment had concluded the price was worth paying.

Inside the American semiconductor industry the rule was its own kind of seismic event. Applied Materials, the largest American semiconductor equipment maker, derived roughly thirty percent of its revenue from China; Lam Research’s exposure ran in the mid-thirties; KLA’s was lower in absolute terms but concentrated on advanced metrology systems whose buyers were precisely the leading-edge Chinese fabs the rule was now blocking. Applied Materials warned investors that the new controls would reduce its fourth-quarter revenue by roughly four hundred million dollars; Lam, the same season, told investors to expect annual revenue impacts in the two-to-two-and-a-half-billion-dollar range; KLA reported smaller absolute losses but suspended advanced-tool support to a list of Chinese customers within days. None of the three publicly opposed the rule. Privately, their Washington staffs spent the months that followed trying to shape the implementing details of every subsequent revision, recognizing that a unilateral American posture without allied alignment would simply transfer business to Tokyo Electron, ASML, Lasertec, and the rising Chinese domestic equipment sector. The argument did not stop the rule. It would shape every rule that followed.

The policy and analytic community processed the rule fastest at the Center for Strategic and International Studies, where Gregory Allen, the senior fellow who ran the AI and emerging-technology desk, published a paper titled “Choking off China’s Access to the Future of AI” on October 11, four days after the rule landed. Allen’s central framing was that the rule was structured around four interlocking chokepoints, advanced computing chips, electronic-design-automation software, semiconductor manufacturing equipment, and the components that went into that equipment, and that together these amounted not merely to slowing Chinese technological catch-up but to actively degrading Chinese capability in a sector. The line that traveled fastest was Allen’s claim that the rule was structured around strangling large segments of the Chinese technology industry, with what he allowed himself to call an intent to kill. The phrase was provocative on purpose. It was also descriptive. The rule did not seek to slow China’s leading-edge progress. It sought to set it back, and keep it there.

The phrase the broader trade and policy community settled on, gradually across the following weeks, was an unprecedented act of economic statecraft. The Cold War’s COCOM regime, which had governed Western technology transfers to the Eastern bloc, had been narrower in target and broader in coalition. The Iran sanctions and the Russia sanctions assembled after the 2022 invasion of Ukraine had been broader in target but thinner in technological depth, mostly financial flows and dual-use items. The October 7 rule, by contrast, reached into the technical guts of a single industrial sector and reorganized global participation in it around a national-security perimeter set by Washington. A Taiwanese executive, quoted anonymously by the Financial Times in the days after the announcement, compared the rule’s effect on the supply chain to an earthquake. The metaphor stuck.

What the rule did not do was, in its own way, as informative as what it did. It did not name any Chinese company in its core threshold-based provisions. The rule’s authors had deliberately avoided the Entity-List approach that had defined American sanctions against Huawei, ZTE, and the various procurement networks of the previous decade, in part because they had concluded, in the Huawei case, that named-entity perimeters could be ducked through corporate restructuring. The October 7 thresholds applied to any Chinese fab making 14-nanometer logic, 18-nanometer DRAM, or 128-layer NAND, regardless of its corporate form, regardless of whether it was state-owned or private, and regardless of whether the United States had formally identified it. The rule also did not block legacy semiconductor production. The mature-node analog, microcontroller, and power-management chips that dominated Chinese fab capacity in absolute terms were not affected. Chinese assembly, packaging, and test continued. Chinese wafer fabrication at older nodes continued. The rule was surgical at the leading edge and silent below it, a structural choice the drafters had defended in interviews as the small yard’s first principle.

It also did not, in October 2022, reach high-bandwidth memory, co-packaged optics, or gaming GPUs, all of which the industry expected would become Chinese workaround targets. Those gaps would be patched in subsequent rules across 2023, 2024, and 2025, in a ratchet pattern that neither side would call by its name. October 7 was the foundation. The rules that followed were extensions and tightenings, each one written against a specific workaround the previous perimeter had failed to capture.

Within seventy-two hours of the announcement, the secondary economy that builds itself around any consequential American regulation had assembled around October 7. Kevin Wolf, the former BIS Assistant Secretary who had drafted earlier-generation export-control rules under Obama and was now in private practice at Akin Gump, hosted an emergency podcast with the China-policy analyst Jordan Schneider that was downloaded a quarter-million times in its first week. Wolf called the rule a fundamental shift in the definition of national security as it pertained to commercial commodities and services. The major trade-law firms produced client alerts running into thousands of words, most of them organized around the same nine-rule structure the document itself had imposed. The analytic community processed the rule into chokepoints and precedents. The industry processed it into decisions: which orders to fulfill before October 21, which engineers to bring home, which line items to write off, which products to redesign.

The redesigns came fast. Nvidia, by mid-November, had a draft specification for what would publicly become the H800, a Hopper-class GPU in which the NVLink interconnect bandwidth had been deliberately throttled below the rule’s 600-gigabyte-per-second threshold, while the on-chip compute throughput remained intact. The A100 received a parallel treatment in the form of the A800. Both products were marketed solely to Chinese customers, on the understanding that they cleared the bandwidth-threshold provision while preserving most of the underlying training utility. Within Nvidia’s data-center sales organization, the China-specific SKUs were treated as a temporary holding action, an effort to retain the Chinese market until either the rule changed or the customers migrated to substitute architectures. Inside BIS, where the H800 specification leaked through unofficial channels almost immediately, the response was patient and brief. The agency would let the H800 ship and would write a tighter rule the following year. The implicit doctrine across the rest of October was that the perimeter was provisional. The next perimeter would be drawn against whatever workarounds emerged from this one. The export-control regime had become a process, not a single act.

Inside the Chinese semiconductor system, the immediate operational effects were severe but recoverable. The leading-edge fabs lost their American service teams. Production lines slowed. Tool calibrations drifted. Equipment that needed parts from American vendors waited. By Caixin’s reporting in the closing months of 2022, YMTC had cancelled close to seventy percent of its outstanding orders with the Chinese equipment maker Naura, not because Naura’s tools were unsuitable but because the cancellations were part of a broader retrenchment as the company adjusted to the new regulatory reality. Across the system, contracts were suspended pending license review. The work moved, in many cases, to Chinese domestic equipment vendors, which were not capable on October 7 of substituting for Applied Materials and Lam and KLA at the leading edge but which by October 8 had several years of state-funded runway to attempt the substitution. By November, the trade press in Beijing had begun describing what it called the indigenization shock, which was not, in retrospect, a shock at all but a violent acceleration of trends that had begun with the original Huawei FDPR in 2020 and had been steadily building since.

Inside the United States, the operational effect on the equipment trio was sharp and durable. Applied Materials reduced its global workforce. Lam Research did the same. KLA absorbed its losses with less drama and pivoted toward European and American demand from the CHIPS-Act-funded buildout. The collective lobbying line from the equipment industry across the following year, made publicly through the Semiconductor Industry Association and privately in bilateral meetings with Estevez and Kendler, was that the rule needed allied alignment to be effective and that without alignment the principal effect would be to transfer market share from American to Japanese and Dutch and increasingly Chinese vendors. The argument was not wrong. It did not slow the rule’s enforcement. The collateral damage to American equipment makers was treated as a cost of the policy, not a reason to abandon it.

The political balance inside Beijing tilted in the opposite direction. Across the autumn of 2022 and the winter of 2023, the State Council, the Ministry of Industry and Information Technology, and the various provincial-level industrial-policy bodies began rewriting their semiconductor strategies around the new perimeter. The Big Fund’s third tranche, which would be announced in 2024, doubled down on equipment substitution. Domestic EDA companies, scattered and weak in early 2022, began consolidating. The Hsinchu poaching networks intensified, although now under tighter Taiwanese export-control attention. The most consequential move inside Beijing, made quietly through party rather than government channels, was the political elevation of Huawei to the operational center of the indigenous chip strategy. The company that had been the first target of the FDPR-based American perimeter would, by 2024, become the most credible Chinese competitor to Nvidia at the AI-accelerator tier, on a road map that started inside the Mate 60 launch the following August and extended into the CloudMatrix systems of 2025 and beyond. The October 7 rule had been written to deny China a capability. It also, by sealing off the alternative, removed the last remaining political ambiguity in Beijing about how much money and how much state will should be poured into building that capability domestically.

What was not in dispute, by the end of October 2022, was that the assumption on which the global chip industry had organized itself for the previous quarter-century had been retired. That assumption had been that semiconductor markets were, in a meaningful sense, global, and that the leading edge of the industry, while concentrated in a handful of companies and countries, was open in principle to any national champion that could pay for the equipment, the talent, and the customers. After October 7, the assumption no longer held. The leading edge of computation was politically demarcated, the instruments drawn from American export-control law and applied with American extraterritorial reach. Whichever side of the perimeter a country sat on, the next decade’s argument would be about whether the line moved, and how.

The drafters at BIS who had finished the document in the small hours of October 7 went home the following weekend with no public visibility and no clear sense, beyond their own immediate reading, of what they had done. The rule was, in its plain language, a regulatory text. It was also the moment at which the chip war stopped being conducted through implication and started being conducted through perimeter. The implication had run for four years, since the first Huawei moves in 2018. The perimeter was visible on October 7. Once it was visible, the rest of the decade had its shape.