Every major war game, every Pentagon study, every think tank report, every newspaper op-ed has been built on this framing. The order of battle has been catalogued. The casualty estimates have been run. The amphibious doctrine of the People's Liberation Army has been studied in detail. The defensive geography of the island has been mapped. The reaction times of US carrier strike groups have been timed. The whole apparatus of contemporary American strategic analysis has spent the better part of a generation getting very good at answering a question that the people who matter in Beijing are not, in fact, asking.

The actual question is much narrower and much more uncomfortable. The actual question is what happens to the disposition of Taiwan Semiconductor Manufacturing Company, and specifically to the leading-edge production that flows out of perhaps fifteen specific buildings in Hsinchu and Tainan, and the integrated workforce of roughly seventy thousand engineers and technicians whose accumulated tacit knowledge is the only thing in the world that currently makes those buildings function. Once you set the strategic question that way, almost everything in the conventional analysis becomes a category error, and the situation is revealed to be more dangerous than the war games suggest, not less, because the action that would deliver the actual Chinese strategic objective is much smaller than an invasion, much harder to deter, and much more available in the present window than the official discourse is willing to admit.

This piece is about the gap between the question the analysts are answering and the question the strategic situation is actually presenting. It is not a piece about whether China will or will not act. It is a piece about what acting would look like, and why the conventional framework cannot see most of the available options.

What TSMC Actually Is

It is necessary to be precise about TSMC because the conventional analysis treats it as a factory, and TSMC is not a factory in any sense that the word usually carries. It is closer to a unique cultural artefact, a piece of accumulated industrial knowledge that has no real precedent in the history of manufacturing, and that cannot be replicated by the means that would normally suffice to replicate a factory.

TSMC produces approximately 90 percent of the world's semiconductors at process nodes of 5 nanometres and below. At 3 nanometres, the figure is effectively 100 percent. The 2 nanometre node, which TSMC is in the process of bringing into volume production, has no commercial competitor. Apple's processors, Nvidia's AI accelerators, AMD's CPUs and GPUs, Qualcomm's mobile chips, the radar processors in the F-35 fighter, the guidance systems for the most advanced US precision munitions, the chips inside the AI training infrastructure that the US technology sector is currently betting its future on, all of these come from a small number of buildings in Taiwan, produced by a workforce that took thirty years to assemble and that cannot be reconstituted on any timescale that matters strategically.

The fabs themselves are visible. The lithography machines from ASML are visible. The supplier ecosystem is visible. The actual asset is none of these things. The actual asset is the integrated process knowledge, distributed across the heads and hands of the workforce, embedded in undocumented routines and tacit adjustments and the kind of know-how that only accumulates through years of running the equipment at the edge of what physics permits. This is what makes the fabs work, and this is what cannot be captured by stealing the buildings, or even by stealing the equipment.

Intel knows this. Samsung knows this. Both have spent enormous sums trying to catch up to TSMC at the leading nodes, and both are years behind, with declining yields and chronic difficulty in matching TSMC's process maturity. The TSMC fab in Arizona, opened with great fanfare under the CHIPS Act, is producing 5 nanometre chips in low volumes on a process that depends on continuous knowledge transfer from Taiwan. It is years away from matching even the current Hsinchu output, let alone whatever node TSMC has moved to by the time it would need to. The Japanese fab in Kumamoto is similar. The new European facility in Dresden is further out. The replacement leading-edge production base does not exist outside Taiwan, and on current trajectories will not exist at strategically relevant volumes until the late 2020s at the earliest, and even then only at process nodes that will be a generation behind whatever TSMC is doing in Taiwan by then.

This is the central fact. The strategic question is not about an island. It is about fifteen buildings.

What "Taking TSMC" Actually Means

The conventional analysis assumes that any Chinese action against TSMC requires capturing and operating the fabs. Once that assumption is in place, the strategic problem becomes very hard for Beijing, because the fabs are uniquely vulnerable, the workforce would resist or flee, the supply chain depends on Western inputs that would be cut off, and the operational complexity of running a leading-edge semiconductor fab without the institutional continuity of TSMC's corporate structure is essentially insurmountable. The conventional analysis concludes that China cannot actually benefit from taking TSMC, and therefore that China will not act against TSMC, and therefore that the situation is stable.

The conventional analysis is wrong because the assumption is wrong. China does not need to operate TSMC. China only needs to deny TSMC to the United States. The two objectives have radically different operational requirements, and the second is much easier than the first, and the second is sufficient for the actual Chinese strategic interest.

Denial achieves what Beijing actually wants. It removes the leading-edge chip supply from the United States and its allies. It cripples the US ability to produce next-generation military systems that depend on those chips. It cripples the US technology sector that has bet its capital on continuous access to TSMC's output. It puts the global AI race, which is the central technological competition of the coming decade, into a configuration where China retains its own production capacity (which is at trailing nodes but exists) and the United States loses access to anything more advanced than what it can produce domestically (which is not much). It does all of this without requiring any Chinese soldier to operate any Taiwanese equipment, without requiring any acquisition of Taiwanese personnel, without requiring any continuity of TSMC's corporate structure. It only requires that the production stops flowing.

Once you set the objective as denial rather than capture, the available mechanisms expand dramatically. Each of them sits below the threshold the conventional analysis treats as the trigger for full US military response, and each of them is much harder to deter than an invasion, and each of them is currently within Chinese capability.

The Mechanisms Below the Threshold

The first mechanism is a maritime quarantine. Chinese naval and coastguard vessels begin stopping and inspecting ships entering and leaving Taiwanese ports under some pretext acceptable to Beijing's domestic and international audiences. The pretext might be customs enforcement, or biosecurity, or smuggling interdiction, or the enforcement of Chinese legal claims that Beijing has been quietly filing for years. The effect is the same regardless of the pretext. Container shipping insurance rates spike within days. Major shipping lines reroute or suspend service to Taiwanese ports within weeks. Inbound shipments of ASML lithography parts, specialty chemicals, ultra-pure neon (the largest source of which is Ukraine, and which is itself disrupted), and the rest of the supply chain that TSMC depends on cannot reach Hsinchu and Tainan. Outbound shipments of finished wafers cannot reach Apple, Nvidia, AMD, or the US defence industry. The fabs are still standing. They are not producing for anyone outside Taiwan. The US has no chips. The global tech economy enters crisis. No invasion has occurred. The American political class is left with the question of whether to start a war over the right of foreign-flagged container ships to enter Taiwanese ports, and the answer to that question, when actually faced, is almost certainly no.

The second mechanism is a cyber operation against TSMC's industrial control systems. Semiconductor fabs are extraordinarily sensitive to environmental disruption. A sustained vibration, a temperature variance of a few degrees, a contamination event in the cleanroom, a corrupted process recipe, any of these can ruin a wafer batch worth tens of millions of dollars and require weeks of recovery. A coordinated cyber attack against the industrial control systems of one or more TSMC fabs could reduce production to zero for an extended period without leaving any visible signature of state action. China has the capability for this kind of operation. The deniability is built into the nature of cyber attribution, which the United States has spent years failing to make stick against either Russia or China at the political level. The US response options are limited because the attack is below the threshold of conventional war and because the alternative (escalating to kinetic response over a cyber action whose attribution is contested) is politically and militarily implausible.

The third mechanism is a targeted physical strike on the fabs themselves, presented not as an invasion of Taiwan but as a limited Chinese military action against a specific strategic threat. A small number of cruise missiles directed at the cleanrooms of TSMC Fab 18 in Tainan, where the leading-edge production happens, would render the facility non-functional for months and possibly years. The cleanroom environments take that long to recover from any breach of integrity. China would take the political and diplomatic cost of the strike. It would not take the political cost of an invasion of Taiwan, because no invasion would have occurred. The US would be left with the choice of full-scale war over a building that no longer produces anything anyway, or de facto acceptance of a fait accompli that destroys the strategic asset but leaves the territorial status of Taiwan technically unchanged. Faced with that choice in real time, and constrained by the industrial limitations we will return to in a moment, the US response would almost certainly be sanctions and rhetoric rather than war.

The fourth mechanism is a hybrid action combining several of the above. Cyber operations to weaken the supply chain and degrade the fabs' operational integrity. Naval pressure to interdict logistics. Political subversion of Taiwan's internal cohesion through information operations and the cultivation of pro-Beijing factions inside the Taiwanese political system. Limited kinetic strikes against selected nodes of the production chain. All of it calibrated to achieve TSMC denial without crossing whatever specific threshold the United States has tacitly committed to defending. The threshold is probably "an amphibious assault on the island of Taiwan." Anything below that threshold is much harder for the United States to respond to with full conventional force, especially in the context of an industrial base that, as we will see, cannot sustain such a response anyway.

The Strategy the United States Has Already Half-Acknowledged

The most revealing fact about the entire situation is that the United States has, in public statements that have not received the attention they deserve, already conceded that the strategy is denial rather than capture. Senior US officials have indicated, in language ranging from the diplomatic to the explicit, that the United States would prefer to see TSMC's facilities destroyed rather than fall into Chinese hands. There have been statements about the willingness to physically destroy the fabs in the event of a successful Chinese invasion. There have been indications that TSMC's own leadership has prepared contingency plans for the destruction of the leading-edge equipment. There have been Taiwanese statements, with varying degrees of denial, that the institutional knowledge of the workforce would not survive a Chinese capture in a usable form.

All of this is the United States telling Beijing, in the language of strategic signalling, that the United States understands the actual game. The actual game is denial. The United States is not threatening to fight a war to keep TSMC in Taiwanese hands. The United States is threatening to ensure that, if TSMC stops being available to the United States, it stops being available to anyone, including China. This is a declaration that the strategic objective on both sides is the disposition of the production, not the territorial sovereignty of the island. And once both sides are openly playing for denial rather than capture, the threshold for action drops dramatically, because either side can achieve its objective with much less than would have been required to capture and operate the asset.

This is also the moment to notice that destruction by China achieves the Chinese objective just as completely as capture would. If the United States cannot get the chips, the strategic position of the United States in the technology competition collapses regardless of who or what destroyed the fabs. China would prefer to capture and operate, because that would allow Chinese exploitation of the asset. China would settle for destruction, because the absolute level of Chinese capability is less important than the relative position of China versus the United States, and the relative position improves equally whether China gains the capability or the United States loses it. The conventional analysis treats the destruction option as a Chinese loss because it focuses on absolute Chinese gain. The strategic reality is that destruction is a Chinese acceptable outcome, and any analysis that treats it as a deterrent against Chinese action is mistaking the structure of the game.

The Industrial Constraint That Removes the Deterrent

The conventional Taiwan analysis assumes that the credible threat of US military intervention deters Chinese action. This assumption rests on the further assumption that the United States is materially capable of conducting a sustained conventional war in the Western Pacific. That further assumption has been quietly false for some years and has become loudly false in the last eighteen months.

The United States military depends on a small number of critical mineral inputs that the United States does not control. Antimony, used in ammunition primers and armour-piercing rounds, is roughly 48 percent produced by China and effectively 0 percent produced domestically in the United States. Gallium, used in the GaN radar arrays of every modern American air defence and fighter aircraft system, is approximately 98 percent Chinese-produced. Germanium, used in night vision and infrared optics, is dominated by China. Tungsten, used in armour-piercing kinetic energy penetrators, is over 80 percent Chinese-produced. Rare earth elements, used in the permanent magnets of every guided munition the United States deploys, are 70 to 90 percent refined in China depending on the specific element. China has, in the last two years, imposed export controls on antimony and gallium specifically, citing national security concerns, and the United States has not been able to substitute its way around these controls at any meaningful volume.

This means that the United States cannot sustainably produce the ammunition, missiles, and electronics required for a major war without Chinese inputs that China would not provide in the event of a conflict with China. The Ukraine war demonstrated the production ceiling for 155 millimetre artillery shells: the United States entered the conflict producing about fourteen thousand shells per month, struggled to ramp toward a target of one hundred thousand per month that has slipped repeatedly, and remains a small fraction of Russian production at the same calibre. The Iran war, in early 2026, demonstrated the higher-tier limits: Patriot, THAAD, SM-3, and SM-6 interceptor stocks were depleted faster than they could be replaced, exposing the same constraint at the level of strategic missile defence.

In a war for Taiwan, the United States would need to expend interceptors against Chinese ballistic and cruise missile salvos at rates that the production base cannot replenish. It would need to expend precision munitions against Chinese naval and air forces at rates that the production base cannot replenish. It would need to provide air defence and resupply to Japanese, Korean, and Filipino allies. It would need to do all of this against an adversary that has been preparing for exactly this scenario for two decades and that controls the inputs the United States needs to fight the war at all. Senior US officials know this. Pentagon planners know this. The CSIS war games of the last few years, which are the most detailed publicly available simulations of a Taiwan conflict, all show the United States either losing the war outright or winning it at a cost (multiple carriers sunk, hundreds of aircraft lost, tens of thousands of casualties in the first weeks) that the political system would not be able to sustain politically even if the industrial base could sustain it materially, which it cannot.

The deterrent that the conventional analysis assumes is therefore not a deterrent. It is a threat that everyone involved knows cannot be made good on at the scale that would be required. China's strategic planners have done the same analysis the CSIS war games have done, and they have reached the same conclusion, which is that the United States is not in a position to fight a sustained war over Taiwan and that a Chinese action below the invasion threshold would face only diplomatic and economic consequences, not military ones.

Why the Window Is Now

The Chinese strategic window is open and closing. It is open because the alternative semiconductor production base outside Taiwan does not yet exist at leading-edge volumes, and because the United States is currently overstretched in multiple theatres simultaneously, and because the political bandwidth and interceptor inventory required to deter Chinese action is being absorbed by the conflicts in the Middle East. It is closing because the Arizona TSMC fab, the Ohio Intel fab, the Dresden European fab, and the Kumamoto Japanese fab are all under construction with the explicit purpose of building redundancy that would reduce the strategic value of an action against TSMC. None of these will be at full leading-edge production before the late 2020s, and most will be a process node or two behind whatever TSMC is doing in Hsinchu by then, but they will exist, and their existence reduces the payoff for a Chinese action.

The closing of the window is partial and slow. It does not fully eliminate the strategic value of TSMC denial. But it changes the calculus, and the optimum point for action, from a strict cost-benefit perspective, is the moment when TSMC is still uniquely valuable and the United States is still uniquely incapable of responding militarily. That moment is the present moment. It will not last.

The Iran war (which is at this writing the dominant theatre absorbing US military attention, interceptor stocks, naval assets, and political bandwidth) is not just a context. It is an accelerant. It exposed the industrial limit on which the entire deterrent rests. It demonstrated that the United States cannot replace high-end interceptors at the rate they are consumed in actual combat against a regional adversary, never mind a peer competitor. It absorbed strategic attention that would otherwise be available for the Indo-Pacific theatre. It produced the conditions under which Chinese action faces the lowest interference cost it has faced since the Cold War. If you were running the strategic planning in Beijing, the question would not be whether to act. The question would be how much further the American position could be permitted to deteriorate before the action becomes unnecessary because the United States has effectively disqualified itself from contesting the outcome.

What the Conventional Analysis Cannot Say

The conventional analytical literature on Taiwan is extensive, well-resourced, and almost entirely useless for the present moment, because the conventional analysis is structurally unable to make the moves that the strategic situation actually requires. The reasons for this incapacity are political and institutional rather than analytical, and they are worth naming explicitly.

The conventional analysis cannot say that the United States may have already strategically lost TSMC, because the analysts who would have to say it are funded by institutions whose continued functioning depends on the assumption that the United States is still the dominant power in the Western Pacific. To admit that the dominance has already eroded past the point of recovery would be to admit that the policy framework the institutions have been advocating for is no longer viable, and the institutions are not built to admit that.

The conventional analysis cannot acknowledge that denial is a Chinese acceptable outcome, because doing so would force the conclusion that there is no defensive posture available that can preserve the asset, which would force the further conclusion that the entire strategic premise of US Indo-Pacific policy needs to be reconsidered. The reconsideration is too large a project for the current analytical class to undertake without losing access to the political conversations that fund their work.

The conventional analysis cannot acknowledge the industrial constraint, because acknowledging it would require admitting that decades of policy decisions about the American defence industrial base have produced a country that cannot fight the war it is committed to deterring. The political and bureaucratic actors who would have to make that admission are the same actors who took the decisions that produced the situation, and they are not in a position to indict themselves.

The conventional analysis cannot model the sub-invasion mechanisms because the war games are organised around invasion scenarios, and the institutional inertia of the war-gaming apparatus makes it very difficult to add new scenario types that fall below the threshold of what the existing apparatus is built to model. CSIS, RAND, and the various Pentagon offices that produce the studies have invested enormous human capital in invasion scenarios. The sub-invasion scenarios would require a different kind of model and a different kind of analyst, and the necessary work has not been done at the scale that would be required.

So the conventional analysis continues to answer the wrong question, in detail, with great rigour, in publications that are read by policymakers who have been trained to ask the same question. The wrong answer is reinforced through the citation networks and the conference circuits and the funding decisions, and the right question (what happens to TSMC, by what mechanism, on what timeline) is left to occasional flashes in the strategic literature that are not pursued because pursuing them would require the kind of conclusions the institutional structure cannot support.

This is the part of the situation that should worry the reader most. It is not that China has options against TSMC that the United States cannot deter. It is that the United States cannot even publicly discuss the options, because the discussion would force conclusions that the political and analytical class cannot afford to draw. The strategic situation is being analysed by people who are systematically prevented from saying what they see, and the result is a public conversation that reassures the political leadership about a stability that does not exist.

The Honest Closing

The strategic situation around Taiwan is more dangerous than the war games suggest, not less, and the reason is the inverse of the reason the official discourse implies. The danger is not that an invasion might come. The danger is that the actual Chinese strategic objective can be achieved without an invasion, by mechanisms the United States is structurally unable to deter, in a window that is currently open, against an industrial base that cannot sustain the war the deterrent assumes the United States can fight.

The fifteen buildings that produce the world's leading-edge semiconductors are the only assets in the contested space that actually matter for the strategic balance of the next decade. Both sides know it. Both sides have built their planning around it. One side has the capability to act and the window in which to act. The other side has a deterrent that depends on a war it cannot materially fight. And the conversation about this, in the public sphere of the country whose strategic position depends on the outcome, is being conducted in a vocabulary that systematically obscures the actual stakes because the actual stakes are intolerable to the people writing the vocabulary.

We do not know whether China will act, or when, or by which mechanism. We do not know whether the action, when it comes, will be a quarantine, a cyber operation, a kinetic strike, a political subversion, a hybrid, or some combination none of us have anticipated. What we know is that the conditions for action are present, that the cost of action is currently lower than at any previous moment in the strategic history of the relationship, that the window for action is closing as alternative production comes online elsewhere, and that the conventional analytical framework being used to reassure Western publics about the stability of the situation is built on assumptions that have already been falsified by visible facts.

The conversation that needs to be happening, in the publications that the political class actually reads, is the conversation about what the United States does in a world where the leading edge of semiconductor production is no longer reliably available to it, by means that fall below the threshold of conventional war. That conversation is not happening, because the people best placed to lead it are the same people whose careers depend on not reaching the conclusions it would force. So the conversation has to happen somewhere else, in publications that are not bound to the political and institutional incentives that govern the major think tanks and the major newspapers, and that can say plainly what the analytical class can only hint at.

The conventional analysis is going to be wrong, one way or the other, within the present decade. Either it will be wrong because nothing happens and the deterrent works in ways the framework cannot explain, in which case it was wrong about the mechanism even if it was right about the outcome. Or it will be wrong because something happens that the framework was not built to model, in which case it was wrong about both the mechanism and the outcome. The question for the reader is which kind of wrongness to bet on, and the honest assessment of the present configuration suggests that the second is much more likely than the first, and that the response of the United States when the second arrives will be characterised by the kind of slow institutional shock that produces decades of policy failure rather than the kind of immediate crisis that produces decisive action.

The fab is the asset. The island is the cover story. The war is the thing that will not be fought, by means that cannot be deterred, in a window that is closing. The American century is ending in fifteen specific buildings, on a small island that the analysts are studying as if it were the territory that mattered, while the actual contest moves through the cleanrooms of the production lines none of the studies are looking at.

That is the situation. We can say it because we are not bound by the funding structures that prevent the analytical class from saying it. The reader can do with the information what the reader will. The strategic reality does not require anyone to believe in it before it operates.

This article draws on TSMC's published market share data and the Semiconductor Industry Association's annual reports, on the CSIS war games on Taiwan published between 2022 and 2024, on US Department of Defense unclassified reporting on critical mineral supply chains, on the Center for a New American Security's work on the defence industrial base, on the Royal United Services Institute's analysis of Western ammunition production capacity during the Ukraine war, on Chinese export control announcements concerning antimony and gallium, on public statements by US Department of Commerce officials on the CHIPS Act and Arizona TSMC fab status, and on the strategic literature on semiconductor supply chains from Chris Miller, Jon Bateman, Jeffrey Ding, and others. The interpretation is the author's own and goes beyond what most of these sources would be willing to state in their own publications, for reasons the article itself attempts to explain.

Overwatch Report is an independent publication. We have no financial positions in any entity mentioned.