Week 9: Geopolitics and National Security (EE 292P — Atoms, Bits, and the National Interest)

Date: March 3, 2026
Instructors: Ali Keshavarzi, Joe Malchow
Guests: Eyck Freymann (Hoover, Columbia), Steve Blank (GSB)

Lane 200–205
Stanford University

The central argument

Taiwan’s importance to the United States is not primarily territorial, military, or even about chips — it is structural. Freymann put it directly: “The reason Taiwan really matters to the United States is not just the territory, it’s not just the chips — it’s because of what Taiwan represents in the structure of the global economy.” The situation is one of triangular interdependence between the US, China, and the broader Indo-Pacific electronics supply chain. Losing Taiwan would not just cost the fabs; it would hand China the ability to weaponize Japan, South Korea, and Southeast Asia against American re-industrialization.

Freymann’s definitional frame for the session: economic statecraft is a country’s pursuit of geopolitical objectives using economic power, where economic power is a function of a country’s attractive and coercive influence over domestic and international market participants in the context of its international relations. The entire session is about how those two instruments interact — and where the US currently stands without a plan.

Why Taiwan matters — three layers

Chips. TSMC’s fabs produce over 90% of the world’s advanced logic chips and 99% of the Nvidia GPUs that train frontier AI models. Freymann: “If we lost access to these fabs today, the global economy would tumble something like 5% of GDP — but it would feel worse than that, because as we all know the US economy is increasingly a story about AI.” He continued: “The NASDAQ would fall off a cliff. There would be a chain reaction of financial calamity that would touch the United States and every major allied country.” Keshavarzi added the compute framing: the entire AI data center buildout — running large language models from Anthropic, DeepSeek, and others — has existed at scale for roughly two years. “No chips, no data centers, no new generation of large language models.” The dependency is new, concentrated, and almost entirely unhedged.

Geography. The first and second island chains have been the geographic basis of US power projection in Asia since MacArthur articulated the concept in the 1940s. Freymann: “China’s Navy, although by fleet size is the largest in the world, is actually hemmed in on these very shallow waters close to the mainland coast.” The two usable exit routes into the Western Pacific are the Miyako Strait and the Bashi Channel, running just north and south of Taiwan respectively. “As long as the US has positions in the first island chain, it can effectively cut those off and hem China’s Navy in next to the coast where it’s easier to destroy.” Taiwan is not a formal treaty ally — Japan and the Philippines are — but its geographic position is what keeps China’s submarine-launched ballistic missile deterrent contained.

Economic order. This is Freymann’s central claim and the one he spent the most time on. “What China wants is not just the geography. China wants to control the economic order across this whole region.” The supply chains that matter run beyond Taiwan: Japan’s specialty chemicals, South Korea’s high-bandwidth memory, Southeast Asian components. “If China can flip Taiwan and in the process gain the ability to tell South Korea, ‘We actually don’t want you sending that high-bandwidth memory to the United States — you know, to go into the TSMC fab in Arizona — don’t do that’… if China can weaponize these regional economies against us, it will become impossible for the United States ever to re-industrialize.” The implication for defense: long-range precision munitions, drones, ships, any complex manufactured system — “it’s going to be very difficult to build that in the United States without access to our trading partners in this region.” Freymann’s summary: “It’s not just about the island, though it matters. It’s not just about the chips, though it matters. It’s about the broader structure of international order.”

Xi Jinping’s strategy: incremental coercion, not shock and awe

The visible signals of Chinese intent are unambiguous. In Inner Mongolia, about four and a half hours northwest of Beijing, stands a perfect replica of Taiwan’s presidential office building. Across the street: a replica of the Ministry of Foreign Affairs. Across from that: a highway cloverleaf matching the entrance to Taiwan’s main airbase. “This isn’t a movie set,” Freymann said. “This is a training ground for invasion.” In 2017, Xi visited the facility, addressed PLA troops in fatigues from an open-top Jeep, and declared: “Always march where the party points, follow the party’s orders” — with bombers and jets overhead. “The signals internally could not be clearer.”

But Freymann’s argument is that the invasion scenario is not Plan A. It is a threat held in reserve. “Xi Jinping is trying to use every lever of his national power to resolve this without a war, because a war would be messy.” The reason a direct strike is unattractive: “If they shoot first at Americans, they make it easy for us. Potentially things that wouldn’t otherwise be possible become possible.” China’s preferred approach is to move incrementally — “rather than through one shock-and-awe move” — transforming the economic order of the region from an open one to a closed one without ever crossing the military tripwire.

On why Xi would choose to act at all, Freymann offered two theories. The peaking-power theory — act before China’s window closes — he found unpersuasive: “All the evidence suggests he thinks China isn’t peaking.” The more plausible trigger is domestic: “If he thinks he’s in trouble at home… there is a protest movement where people are holding up signs calling for his downfall… he may need to wag the dog.”

The quarantine scenario

This is the live operative scenario, and Freymann walked through it in detail. “The quarantine scenario is not a blockade. A blockade says no one in or out — this is Stalingrad, this is medieval.” The quarantine is a legal action, a Coast Guard action.

How it would work: one morning the Chinese Coast Guard posts a notice that contraband — fentanyl, weapons — is being smuggled into Taiwan Province, which is subject to Chinese customs law. Ships are boarded and inspected. Clear papers? On your way. Contraband? Come clear customs on the mainland. “China would not be saying ‘we’re declaring war.’ They would not be saying ‘we are blockading you.’ They’re saying: we are just asserting the right, as a matter of law, to control who and what comes and goes from Taiwan.”

The genius of the scenario is what happens next. Freymann put it as a class question: do the fabs stay on or shut down? The answer, confirmed by students: they stay on. “TSMC shareholders wouldn’t let them shut down the fabs. So that’s how China gets the fabs intact.” US arms shipments to Taiwan? “Terribly sorry, it’s a violation of our customs law. Please clear customs on the mainland.” American efforts to evacuate process engineers and EUV machines? “Terribly sorry, we have an export control regime. You’re not allowed to send that to the United States.”

The result: “This is the scenario that’s really dangerous — it pushes the burden of escalation onto the United States.” The US president faces a binary choice. “Do you want to blow up the fabs? Do you want to take a step which leads to an immediate global financial crisis? What US president would want to push that button?” Freymann: “It’s well understood in deterrence theory that you want to push the burden of escalation onto your opponent. This is basically textbook.”

There is historical precedent for a president invoking emergency financial powers in exactly this kind of moment. In August 1914, Treasury Secretary McAdoo pressured New York to keep the Stock Exchange closed to prevent gold flight at the outbreak of WWI — detailed in William L. Silber’s When Washington Shut Down Wall Street (Princeton UP). The question for a future US president is not whether such powers exist; it is whether using them against China would destroy American allies along with the adversary.

Economic MAD and its limits

The theory of weaponized interdependence holds that mutual economic entanglement deters both sides from pulling the trigger. Freymann acknowledged the logic: “The basic idea is maybe we can work all this out without actually having to have the war and the crisis… China is so dependent on networks that the US controls, and the US is so dependent on networks that China controls, that neither of us actually wants to blow those up. That’s the theory of economic mutually assured destruction. And there’s something to it.”

But the theory is eroding. China is actively building insulation. The Economist reported in October 2025 that “China’s secret stockpiles have been a great success — so far. Xi Jinping is desperate for trump-proof access to food, fuels and metals.” This is direct evidence that China is preparing to sustain an economic confrontation, not avoid one. Meanwhile the US is decoupling at a slower pace. “As decoupling proceeds, the mutual deterrent weakens.”

The financial chokepoint question — how much leverage the US actually has — is the subject of Edward Fishman’s Chokepoints: American Power in the Age of Economic Warfare (NYT Bestseller, endorsed by Paul Kennedy as “one of the most important books on economic warfare ever written”). The answer may be more than commonly assumed.

But the harder question is Freymann’s slide 13 formulation: what if we don’t want economic MAD? Nicholas Lambert’s Planning Armageddon: British Economic Warfare and the First World War (Harvard UP) offers the historical parallel. Britain planned a comprehensive financial war against Germany before WWI — targeting City of London credit relationships with German banks — and nearly refused to use it at the moment of crisis because the weapon was self-destructive. That is the exact dilemma: the chokepoint weapon exists, but deploying it against China could trigger a global depression that hits American allies as hard as the adversary.

The congressional verdict is unambiguous. The House Select Committee on the Chinese Communist Party stated in December 2023: “The United States lacks a contingency plan for the economic and financial impacts of conflict with the PRC.” Freymann and Hugo Bromley wrote the closest thing to one that currently exists: “On Day One: An Economic Contingency Plan for a Taiwan Crisis,” CSIS Occasional Paper №4, July 2024, published by the Hoover Institution’s China’s Global Sharp Power Project.

Deterrence and autonomous warfare

Drone warfare has restructured the cost calculus of conventional conflict. Keshavarzi framed the economics: “Imagine the plane that you’re flying is $10 million, and the missile that you’re going to shoot at it to bring it down has also a cost. If that thing that is flying is — like what Iran is doing — these cheap drones, then you don’t want to use your fancy missiles that cost millions of dollars to bring them down. So the war has an economy.” Blank added the quantity dimension: Iran fired close to several thousand drones in the last few days of escalation; the same pattern runs in Ukraine every day for four years.

The 90-mile crossing matters. Blank: “So that ‘irrelevant’ conversation over here now gets to be quite relevant if you’re a Chinese fleet attempting a manned landing. Now all of a sudden, if Taiwan decided to look at Ukraine and Iran and go, ‘Huh — perhaps instead of buying F-35s, we should buy 10,000 drones that can take out multiple waves of ships crossing the straits.’ You can’t hide those things. They won’t teleport.” This is the “porcupine strategy” — a credible independent defense that doesn’t depend on US intervention. Blank also cited the Gaza lesson: “If you decide to dig deep on some of those islands, it’s really hard to remove people from tunnels 50 feet underground.”

Freymann’s caveat: military deterrence is necessary but not sufficient. The quarantine scenario is designed precisely to avoid the military tripwire. Even a fully porcupined Taiwan is still vulnerable to a legal Coast Guard action that never fires a shot.

On AI and the strategic balance, Freymann made a specific prediction: “Frontier AI is first going to affect the strategic balance when you have end-to-end automatable cyber agents. Because once you have that, the scope of what you can do in cyberspace multiplies. You can potentially get into your adversary’s nuclear systems and everything else.” If TSMC is taken off the board, the US and China both pour limitless capital into rebuilding compute. “After day one of a Taiwan crisis, the world has changed… there’s a rupture and then we’re racing for superintelligence with truly existential stakes.” Rebuilding fab capacity might take five to seven years; but that may not be the most important question.

Process knowledge and the TSMC problem

The class session produced a detailed discussion of why TSMC’s position is harder to replicate than it appears. Keshavarzi: “A lot of TSMC’s secret sauce — what you describe as their IP, and it is IP — is not formulas written in a notebook in a great big safe. It lives in the heads of the people: the process engineers on the line.” The specific mechanisms:

Yield. TSMC yields on advanced nodes are substantially higher than those of Huawei’s chip arm, SMIC, or other PRC manufacturers. Keshavarzi: “When you reduce the yield, it becomes much more costly per chip to produce — because you have this very capital-intensive equipment, your machines are running all the time, but a certain amount of the output has to be destroyed.” The yield gap is the productive gap.

Dormitory culture. “In TSMC’s fabs, professionals who’ve done this for decades will essentially live in dormitories on the campus. They’re available 24 hours a day, seven days a week.” Keshavarzi compared the talent dynamic to Nike in Portland: TSMC in Hsinchu is the only serious employer for that graduate cohort. “When you’re here in the Bay Area, you can jump from Meta to Google to whatever. It’s Taiwan and TSMC. Any graduate in that area, it’s a dream job. Here, Intel is having a hard time even hiring people.”

Non-static lines. Malchow: “The logic chips are generally a custom business. The lines get, in a sense, decomposed and recomposed quite often… they are like fine harmonic devices that once stopped are quite a pain to get going again. And they do stop — because there’s a new design next year. And the literal space on the PCB is a different size. And you have to reconfigure everything.” Even a complete recipe set from today’s process would not reproduce tomorrow’s line.

The student discussion surfaced the ASML dependency (lithography machines, manufactured in the Netherlands, take roughly two years to build and cannot simply be airlifted and restarted elsewhere) and the question of whether buying TSMC’s securities gets the US what it actually needs. Malchow’s answer: “Buying TSMC — buying the securities of the company — it’s not the same as buying the actual physical assets. The physical assets are lithography machines and upstream things like highly passivated optics. But even then, you also cannot simply airlift those machines, bring them somewhere else, and expect them to work the same.”

The submarine parallel illustrates the same structural problem at scale. The US is producing 1.2 nuclear attack submarines per year while retiring approximately 1.5 — the fleet is contracting. Keshavarzi: “Congress is now pouring money into it. We don’t have the skilled workers to make more, faster. And we’re going to have a shortage of over 120,000 technical experts in submarine manufacturing.” The nuclear attack submarine fleet is “the tip of the spear of the US force structure in the Pacific” — they “either kick the door down or they fail in the first 48 hours.” The bottleneck is not appropriations. It is embodied process knowledge of the exact same kind TSMC holds.

Industrial policy: what would it take

Blank: “The CHIPS Act was the first real national industrial policy in 75 years in the US. We’re now doing national industrial policy by domains — critical minerals and others — but we still don’t have an integrated strategy. Unlike China.” The diagnosis of the gap is empirical. House Select Committee on the Chinese Communist Party, December 2023: “The United States lacks a contingency plan for the economic and financial impacts of conflict with the PRC.”

Blank’s recommended reading is Apple in China, which documents Apple outsourcing manufacturing to China as a cost decision while China played a long-term strategic game. “China was actually playing Apple for the long term. Basically built an infrastructure that strengthened China but also got us to where we are for semiconductors.” His argument: “The claim is, ‘Well, we could never get all that stuff back.’ And you know what — it’s bullshit. You could actually read about what they did and go, ‘We could do our version of this back if we decided to do that.’ It really is a reverse playbook.”

Freymann’s caution: “Industrial policy can work — China has proved that it can work, and Taiwan has proved that it can work. But it’s a road littered with burning wrecks on both sides of the road. And the US track record in recent memory is not very good.” The structural constraints Freymann identified: China owns the land and can compel relocations. The US cannot.

Several constraints the class worked through:

Talent gravity. The student asking Keshavarzi whether the problems are solvable with enough money got a sharp response: “Are you going to join Anthropic after you graduate, or start your own company and have those stock options — or are you going to go to a manufacturing fab where you’re on call 24/7 with a pager? The pager goes off at 3 AM because the EUV machine cannot wait. Are you really going to do that, regardless of how much I pay you? I pay you $500,000 a year salary.” The point being that the incentive structure of the US tech labor market actively pulls against the manufacturing workforce the fabs require. Keshavarzi suggested, citing TJ Rogers, that stapling a green card to every US PhD would at least create the option for the 80% of EE PhD graduates who are non-American.

Capital structure. Malchow: “There is a fundamental difference between the financials of a software company and the financials of an industrial or manufacturing company. If you just imagine a smooth gradient from crazy profitable to break even to losing money — that’s what software looks like. In a highly CapEx-intensive business — I’m in the battery business for part of my portfolio — it is extraordinarily binary. You can make a lot of money and you can shift to losing all of your investors’ money and your plant very, very quickly.” He noted that Apple and Nvidia — companies we classify as manufacturers — don’t actually manufacture. “They’ve depended upon a very high-quality contract manufacturing layer for a long time. That absorption layer has removed an immense layer of manufacturing CapEx capital expenditure management talent from Western business.”

The “who is we” problem. Freymann: “You need an entity that’s going to be subsidized in perpetuity that’s not actually a private company. It seems like the US government would need to heavily fund this for it to be a reality.” On Intel specifically, the discussion surfaced reports that the Trump administration is seeking a golden-share equity stake — “because if things go in a very bizarre direction, they want the ability with a stroke of the pen to seize control of Intel and direct it to do what they want to do.” Freymann: “But in peacetime, if you do that, you ruin their P&L.”

On DoD acquisition, Blank offered a more optimistic read of recent changes: “Our Department of Defense — now called Department of War — basically acquisition has been taken over by private equity. And that’s really good. The Secretary of Army and Secretary of Navy are ex-VCs or private equity folks. They did a private equity roll — took all these silos of acquisition and contracting and deployment and put them all together. They created portfolio acquisition executives, PAEs, which are creating just enormous efficiencies. They basically took a shotgun to the FAR, the Federal Acquisition Regulations, and have thrown out thousands of pages of that.” But he identified the corresponding failure: “The people who are advising technology, with all due respect to VCs in the room, are venture capitalists, not scientists. VCs historically have thought that basic science and innovation came from the stork. Right now we’re kind of operating like that in the United States by actually shooting basic science.”

Closing frame

Both Blank and Freymann converged on the same terminal point. Freymann: “The question we face is: when do we pull the plug? If we do — break the glass, whatever your metaphor — what’s the plan?… If we have to hustle, and the world is different, and the only thing that matters is rebuilding this ecosystem in safe countries as fast as possible — what are the inputs we need? Some of the people in this room may have a role to play in that process.”

Blank: “Every one of you has the ability to go change the system and make it better. This class, I hope, made you think about the complexity of what’s in front of all of us. And that there are no easy answers.” He closed with the WarGames line — “the only winning move is not to play” — before noting: “In this case, we need some winning moves — at least, to decide not to play.”

Keshavarzi left a first-principles exercise: calculate how many TSMC wafers per month produce how many Blackwell dies, how many dies fill an AI data center, and what losing that supply chain means for the models running today. “That’s the math that every one of you should know at the tip of your tongue… Then you can start doing policy going, ‘Well, wait a minute — the CHIPS Act was to get fabs built here, but maybe we should also be thinking about — ‘“

Blank on the long-term technology bet: “Eventually people are going to realize that perhaps this rush to build data centers just to power large language models might not be the most efficient way to build these things. There might be different models for computing — neuromorphic chips, other architectural approaches. If in fact we lost Taiwan, all that research would be taken off at scale… That’s one possible path.”

Course notes and upcoming sessions

• Week 10 (next): Robert Burgelman (Stanford GSB) — strategic leadership, technology management. His book Strategy Is Destiny (foreword by Andy Grove) is the reference. Context: Intel’s arc from pinnacle to current position, and how firms manage the top of a technology curve.
• Reading list from this session:
• Apple in China — Blank
• Freymann & Bromley, “On Day One: An Economic Contingency Plan for a Taiwan Crisis,” CSIS Occasional Paper №4, Hoover Institution, July 2024
• Edward Fishman, Chokepoints: American Power in the Age of Economic Warfare
• William L. Silber, When Washington Shut Down Wall Street (Princeton UP)
• Nicholas Lambert, Planning Armageddon: British Economic Warfare and the First World War (Harvard UP)