The Taiwan Triage

Position: The supply chains underpinning U.S. AI dominance are structurally more fragile than the market assumes. The Iran conflict is stress-testing all three at once.

Methodology: AI-assisted evidence infrastructure · Human-directed thesis · Primary-source verified

In July 2025, the Trump Administration issued its clearest signal in a generation: build AI infrastructure fast, build it big, build it here. Eight months later, three concurrent shocks are undermining that directive across fragile supply chains and testing an AI growth thesis backed by trillions in cumulative investment.

The first is silicon. Over 90% of advanced semiconductors are fabricated on an island that imports 97% of its energy and holds roughly 11 days of LNG reserves. Taiwan depends on Qatar for approximately 28% of its LNG — shipped through the Strait of Hormuz, which has been effectively shut since February 28. The Iran shock did not create Taiwan’s energy vulnerability. It stress-tested it in real-time — and the results are not reassuring.

The second is the U.S. power grid. Active natural gas capacity in the interconnection queue surged 72% year-over-year in 2024 to 136 GW, while solar, storage, and wind all declined. FERC-approved fast-track mechanisms in PJM, MISO, and SPP disproportionately favor gas. Three-quarters of MISO’s fast-track requests are gas capacity, including three 1.5 GW units for Meta data centers in Louisiana. The grid has locked in a gas dependency for the next generation of AI power.

The third is the industrial crowding-out. The same constrained grid must also power the semiconductor fabs, battery gigafactories, and manufacturing capacity the Trump Administration is reshoring — precisely to reduce its Taiwan exposure. Hyperscalers will outbid everyone else for electrons. The entities that get crowded out are the very industrial assets that were supposed to make America less dependent on the island the Iran shock just put at risk.

The consensus narrative entering 2026 was straightforward: the U.S. AI buildout is a generation-defining infrastructure cycle constrained primarily by domestic power generation capacity. Solve the megawatt problem, the thinking went, and the rest follows. U.S. economic growth is increasingly coalescing around this thesis.

That framing is incomplete. It treats the AI buildout as a domestic story — a matter of turbines, transmission, and permitting timelines. It ignores the supply chains upstream, offshore, and adjacent to the grid, upon which the entire buildout depends.

On February 28, U.S.-Israeli strikes hit Iran across 24 provinces, killed Ayatollah Khamenei, and triggered an effective shutdown of tanker traffic through the Strait of Hormuz. Iran’s Revolutionary Guard issued warnings prohibiting vessel passage, and within 48 hours, traffic through the strait dropped to near zero — not through a naval blockade, but through drone strikes that caused an insurance-driven shutdown. Brent crude surged past $80, European natural gas futures jumped roughly 30%, and QatarEnergy halted LNG operations at its two main facilities.

The market is pricing this as an oil story. It is also — and more consequentially for allocators with technology exposure — a chip story.

The broken assumption is that U.S. AI infrastructure risk is bounded by domestic variables: permitting delays, interconnection queues, generation capacity. In reality, the AI buildout depends on a global supply chain architecture that routes through some of the most geopolitically fragile chokepoints on earth. The Iran conflict is not a black swan. It is a stress test of vulnerabilities that were visible, documented, and unhedged.

Think of this the way the early internet era’s infrastructure investors learned to think about undersea cable routes. In the late 1990s, it was assumed that bandwidth was the constraint. The real constraint turned out to be the physical concentration of capacity through a handful of cable landing stations — a vulnerability that only became visible when cables were cut. The AI buildout faces an analogous concentration risk, except the “cables” are LNG tanker routes and the “landing stations” are Taiwan’s power grid and the U.S. interconnection queue.

Three supply chains are now flashing red, and none of them were designed to absorb a geopolitical shock of this magnitude simultaneously.

Over 90% of the world’s advanced semiconductors are fabricated on a single island that imports 97% of its energy. TSMC, the sole manufacturer of the most advanced AI chips — supplier to Nvidia, Apple, and Broadcom — currently consumes roughly 8–10% of Taiwan’s total electricity (S&P Global; IEEE Spectrum; Global Taiwan Institute). S&P Global projects that figure will reach 24% by 2030 as AI chip demand scales progressively more energy-intensive process nodes. A 2nm chip requires more than three times the energy of a 28nm chip, driven by the extreme power demands of EUV lithography.

Taiwan depends on Qatar for approximately 28% of its LNG imports, shipped through the Strait of Hormuz (New Lines Institute; S&P Global). Bloomberg reported this week that Taiwan is already scrambling for alternative LNG sources following the disruption. The Foundation for Defense of Democracies conducted a wargame simulating Chinese energy coercion of Taiwan and concluded that a sustained LNG squeeze would cut the island’s grid capacity in half, forcing a triage between powering hospitals and powering chip fabs (FDD, Nov 2025). CSIS wargames concluded that under blockade conditions, Taiwan’s LNG would run out in under two weeks.

The Iran shock is a preview. If a Middle East conflict can disrupt 28% of Taiwan’s LNG supply and send the island scrambling for alternatives within 48 hours, a deliberate Chinese energy quarantine — executed through coast guard patrols, insurance manipulation, and cyber operations against LNG terminal logistics — would be catastrophic.

The U.S. interconnection queue holds enormous nominal volume — nearly 2,300 GW of generation and storage capacity as of end-2024 — but the queue is a composition and speed problem, not a volume problem.

Active natural gas capacity in the queue surged 72% year-over-year to 136 GW in 2024, while solar fell 12%, storage fell 13%, and wind fell 26% (LBNL Queued Up 2025 Edition). FERC approved fast-track interconnection mechanisms in PJM, MISO, and SPP that disproportionately favor gas plants through operational flexibility and the ability to repurpose retiring coal sites. Three-quarters of MISO’s fast-track requests are gas capacity, including three 1.5 GW units for Meta data centers in Louisiana (Earthjustice; Sierra Club).

The system has locked in a gas dependency for the next generation of AI power while simultaneously constraining the renewable alternatives that would diversify the fuel mix. If the Administration’s July 2025 directive was to build fast, the grid’s revealed preference is to build gas.

The Administration’s AI ambitions sit alongside an equally aggressive reshoring agenda: semiconductor fabs that each consume a nuclear reactor’s worth of power, battery gigafactories, and the first sustained U.S. electricity demand growth in two decades. TSMC’s Arizona fab alone will require 1,200 MW at full build-out. The semiconductor sector has attracted over $200 billion in private investment for new U.S. facilities; the EV battery sector over $75 billion since late 2022.

These three demand sources — data centers, fabs, and manufacturing — are now competing for the same constrained grid. Only one of them can afford to pay whatever it takes.

The single geographic chokepoint where all three supply chains converge is Taiwan. The island is being squeezed from two directions at once: the Iran shock disrupts the energy that powers its fabs from the south (Qatari LNG transiting Hormuz), while China tightens the aperture from the north. In 2025, Chinese cyber units averaged 2.63 million intrusion attempts per day against Taiwan’s critical infrastructure — a 113% increase from 2023 — with energy facilities experiencing the most significant year-on-year surge (Taiwan National Security Bureau).

The supply chain fragility extends beyond physical infrastructure into the AI technology stack itself. On February 27, the Pentagon designated Anthropic — the first AI lab to deploy models on classified military networks and arguably the most successful enterprise AI company globally — a “supply chain risk to national security” after the company refused to allow unrestricted military use of its technology (CNBC; NPR; Bloomberg). Hours later, OpenAI announced a replacement deal.

The message to the AI industry is unmistakable: compliance with the Administration’s demands is the price of participation. This is not merely a policy story. It fractures the domestic AI supply base at precisely the moment when unified capability matters most.

Three orders of impact present distinct market mechanics. The broader market appears to be pricing Order 1 while largely mispricing Orders 2 and 3.

U.S. gas-heavy utilities appear structurally insulated from near-term downside. The downstream AI hardware supply chain, however, exhibits significant asymmetric risk. If the Iran conflict disrupts Qatari LNG for even two weeks, Taiwan’s grid enters crisis and TSMC’s most advanced fabrication lines halt.

Portfolios reliant on continuous silicon delivery face acute operational exposure. The global supply of advanced chips depends entirely on an import-dependent, fossil-fueled island grid facing a potential disruption to 28% of its gas supply — while simultaneously absorbing 2.63 million daily cyberattacks on its energy infrastructure from a nation that has publicly committed to annexation.

While the U.S. grid will constrain the AI buildout to a degree, it is far more likely to constrain the industrial reshoring that the Administration needs to reduce its Taiwan exposure. Hyperscalers are price-inelastic. They will absorb whatever capacity costs the market demands. Battery plants, semiconductor fabs, and manufacturers will not.

The structural winners in the next cycle are not the utilities building gas plants for tech, but the companies selling energy independence to everyone tech left behind.

Utilities are financing 40-year thermal assets against 15-year hyperscaler PPAs, underwriting a stranded-asset crisis in slow motion. When hyperscalers migrate to SMRs, geothermal, or whatever reaches cost parity — and they will, because they are already investing in all three — the ratepayer inherits the liability. If this scenario plays out, there may be an opportunity for utilities to pivot caseload gas to meet demand from industrial capacity assuming a durable reshoring trend materializes.

The irony is structural: the AI buildout that the Administration is fast-tracking will monopolize the grid capacity that the reshoring agenda needs to reduce the very Taiwan dependency that the Iran shock just exposed.

The market frames the Iran conflict as an oil price story and the AI buildout as a domestic power story. Both framings miss the structural reality. The supply chains underpinning U.S. AI dominance — silicon fabrication, grid capacity, and the industrial base — converge on a single geographic chokepoint (Taiwan) and a single fuel dependency (natural gas) that the Iran shock is stress-testing in real time. The Administration’s own reshoring strategy to reduce Taiwan exposure is being crowded out by the AI buildout it is simultaneously fast-tracking. The system is not failing at any single point. It is failing at the seams where three supply chains meet — and the Iran shock is pulling all three at once. The allocators who see this as a multi-chain propagation event, rather than a commodity price spike, will position ahead of the repricing.

About Aroko: Aroko provides strategic advisory and capital allocation intelligence at the intersection of energy transition, technology infrastructure, and geopolitical risk. Our analytical process combines proprietary evidence infrastructure with human-directed thesis formation. Every keystone claim is verified against primary sources, and all editorial judgment and capital allocation framing is conducted by Aroko’s team. The Letter is published biweekly for institutional allocators.

Disclaimer: This article is for informational purposes only and does not constitute financial, investment, legal, or tax advice. The opinions expressed regarding macro trends and infrastructure investments are solely those of the authors. Past performance does not guarantee future results. Readers should consult with a qualified financial professional before making any investment decisions.