Taiwan's Two Faces

For the past three years, the talk has been of "U.S.-China decoupling," but what has actually happened looks more like diversion. The International Monetary Fund (IMF) calls it a "structural diversion" of global trade and capital — in effect, a reshuffling of the supply chain's underlying order ^[24]. Research from the Centre for Economic Policy Research (CEPR) and the Rhodium Group likewise finds that most companies have not left China; instead they have adopted a "China+1" strategy, spreading part of their capacity to Vietnam, India, and Mexico to keep an alternate outlet open for the supply chain. But Rhodium also cautions that China's own supply-chain resilience remains strong, and that Chinese-invested firms are often behind much of the new diversification capacity in Southeast Asia — so the actual extent of "decoupling from China" should not be overstated ^[22][23].

The numbers sketch the shift. China's share of U.S. imports has fallen from 22% to 16%; U.S. imports from China dropped to $308 billion in 2025, the lowest since 2009 and roughly 40% below the 2018 peak ^[4]. Picking up part of that share are Vietnam (whose share grew 38%, drawing more than $36 billion in new foreign investment into electronics in 2025) and Mexico (whose share grew 52%). Tariffs are redrawing the map of the supply chain.

Diversification, however, has never been free. Studies estimate that dual-sourcing costs 5% to 12% more than betting everything on China alone — this "resilience premium" does not simply vanish; it gets distributed across every link of the supply chain as higher prices or thinner margins, ultimately absorbed jointly by consumers and manufacturers ^[24]. And the shortcut of "just changing the postmark" is getting harder to take: the United States has tightened rules-of-origin verification and made transshipment through third countries — using another country to launder the origin of goods — a priority enforcement target. Take polysilicon as an example: for specific items like this, what actually counts as "transshipment" has yet to be pinned down ^[25].

The cost of this migration also shows up in the global trade scorecard. The World Trade Organization (WTO) estimates that global goods trade volume shrank 0.2% in 2025, with North American exports plunging more than 10% ^[21]; independent think tanks assess that this round of U.S. tariffs is the largest tax increase since 1993, adding roughly $900 to the average American household's costs in 2026 ^[12]. Global trade is broadly slowing — and Taiwan is growing against that current. That is precisely the anomaly this report sets out to explain.

Companies haven't decoupled — they've just moved some eggs from one basket into a "friendly" basket. And the bill for that diversification is being split by the whole world.

In this great migration, Taiwan captured the richest slice. Start by following the money.

The story begins with America's cloud giants. Amazon, Microsoft, Google, and Meta together plan $635–665 billion in capital expenditure for 2026, the vast majority of it pouring into AI data centers and chips ^[9]. That astronomical sum flows first to NVIDIA — its data center segment has become NVIDIA's fastest-growing, highest-revenue business, and the core engine driving the expansion of the entire data-center semiconductor market.

And where do NVIDIA's chips come from? The answer is the physical endpoint of this money trail: more than 90% of the world's most advanced-node chips can only be made in Taiwan ^[10]. Taiwan captures more than 60% of global foundry revenue, and at the most cutting-edge nodes it comes close to a monopoly. TSMC's revenue in the fourth quarter of 2025 reached $33.7 billion, up more than 20% year-on-year, with high-performance computing (HPC) alone accounting for 55% of that quarter's revenue ^[6]; its 2026 capital expenditure is projected to reach as high as $56 billion, a record high, with 70–80% going to advanced processes ^[5]. Even in advanced packaging (CoWoS) — the chokepoint for AI chips — TSMC plans to push monthly capacity to roughly 130,000 wafers by the end of 2026, more than half of it claimed by NVIDIA alone.

Trace this chain to its end and the mystery of Taiwan's doubled surplus resolves itself: it was not won through cheap labor or currency subsidies, but by sitting at the cash register of the global AI arms race. The more the world wants AI, the fuller Taiwan's order books get — this is both its firmest source of confidence and its hardest-to-replace bargaining chip.

But what makes Taiwan shine is also precisely where it is most fragile. The key point is this: Taiwan's advantage and its risk are the same thing.

Lay out Taiwan's economic structure and semiconductors account for roughly one-fifth of GDP ^[11], while electronic components and ICT products alone make up a staggering 78.5% of total exports ^[10]. This degree of concentration is extremely rare among major economies. By the first quarter of 2026, Taiwan's year-on-year GDP growth had rocketed to an astonishing 13.69% — but the Chung-Hua Institution for Economic Research (CIER) poured cold water on the number: nearly 60% of that growth came from a single sector — advanced semiconductors ^[11].

Put another way: strip out semiconductors, and Taiwan's 2026 growth story would immediately look far less bright. That also explains why forecasts for 2026 diverge so sharply across institutions — the official Directorate-General of Budget, Accounting and Statistics optimistically projects above 9%, while more conservative institutions like CIER put the full-year figure closer to 4% ^[11]. The gap comes down to how much longer the AI demand curve can keep climbing, and whether 2025's extraordinarily high base will flip into a headwind for 2026.

When an island's growth, exports, and even its currency all hang on a single curve, "shining" and "risk" become two faces of the same coin. Should AI capital expenditure suffer a bubble-style correction, the sharper the shine today, the more violent the pullback could be tomorrow. This is precisely the structural challenge that a highly concentrated economy must confront squarely.

The price of being a beneficiary is buried in another set of numbers — and this set is the real reason Taiwan ended up at the negotiating table.

In 2025, the U.S. goods trade deficit with Taiwan nearly doubled to $146.8 billion, up 99.1% year-on-year from just $73.9 billion in 2024 ^[1]. That leap pushed Taiwan into position as the fourth-largest source of the U.S. trade deficit, trailing only China ($202 billion), Mexico ($197 billion), and Vietnam ($178 billion) ^[4]. Taiwan simultaneously became the fifth-largest U.S. trading partner and its ninth-largest export market.

Here is the cruel symmetry: to an administration that treats the "trade deficit" as a metric to attack, the more successful Taiwan's exports to America become, the larger that deficit grows — and the more likely Taiwan is to be targeted with tariffs. The same AI wave that sent Taiwan's exports soaring also pushed it squarely into Washington's crosshairs. A deficit that doubled in a single year was precisely the leverage U.S. negotiators held when they put the "investment for tariff relief" deal on the table ^[4].

What emerges is a paradox: Taiwan was not dragged to the negotiating table because it was weak, but because it was too successful. Its bargaining chip (irreplaceable chips) and its soft underbelly (a glaring surplus) are one and the same thing. Understanding this is the key to understanding the contentious agreement that followed.

2025 into 2026 was a year of violent tariff swings, with the rules changing overnight. For any Taiwanese exporter, the biggest enemy that year was not the tariff rate itself, but the uncertainty.

The fuse was lit on April 2, 2025. The Trump administration rolled out "reciprocal tariffs," imposing a baseline 10% on the world and scaling up to as high as 50% depending on the size of the trade deficit; China's "reciprocal tariff" rate spiked to 125% within days, and after fentanyl-related tariffs were later layered on top, it briefly reached as high as 145% ^[12]. Taiwan was briefly placed on the high-tariff list, and the most sensitive category — machine tools and hand tools — saw its tariff jump from an original 4.4% straight to 32%, all but cutting off the lifeline for thin-margin traditional manufacturers ^[17].

The deadlock did not break until January 2026. The U.S. and Taiwan signed an agreement: Taiwan committed to massive U.S. investment and credit guarantees in exchange for the reciprocal tariff coming down from 20% to 15% ^[7]. The headlines only remembered "20 down to 15," but three fine-print provisions determined who actually won and who actually lost.

Provision one: 15%, and "non-cumulative." This 15% rate is calculated on a "non-cumulative" basis — meaning it does not stack on top of other existing tariffs. For traditional manufacturers previously squeezed by multiple overlapping rates, briefly facing as much as 32%, this was real relief, pulling them back from the cliff's edge ^[17].

Provision two: Section 232 looks frightening, but its scope is narrow. For semiconductors, Washington separately imposed a 25% tariff under the Section 232 national-security statute — a fearsome number on paper. But read the notice closely and it targets only "high-end chips" meeting specific computing-performance and memory-bandwidth thresholds, with broad exemptions for U.S. data centers, startups, R&D, and public-sector use — what actually gets hit is a small slice of commercial imports ^[12]. The blade looks wide, but its edge is narrow.

Provision three: trying to launder your origin by routing around it? That road is being closed. The agreement also came down hard on transshipment: goods routed through a third country to disguise their origin and dodge higher tariffs will face steeper duties. Although the term "transshipment" still has no precise legal definition for specific items such as polysilicon, under tightened rules-of-origin verification, the old trick of simply "slapping a label on in Vietnam" has largely stopped working ^[25].

The real signal of this agreement isn't the tariff rate — it's that it rewards genuinely relocating capacity, and punishes merely changing the postmark.

Yet just as everyone thought the rules had settled, part of the foundation was pulled out from under them. On February 20, 2026, the U.S. Supreme Court ruled that the president lacks authority to impose tariffs under the International Emergency Economic Powers Act (IEEPA); the administration was forced to fall back on another, temporary statute (Section 122) — one that itself expires on July 24, 2026 ^[12]. In other words, the "certainty" Taiwan purchased with an enormous bargaining chip is standing on a floor that could shift again at any moment.

This agreement has therefore triggered two starkly different readings in Taiwanese society and international commentary.

The first reading is the "drain" thesis. In this view, the $250 billion in U.S. investment, combined with the American claim of moving "40% of the supply chain" to the United States, is not ordinary corporate globalization but the forced offshoring of Taiwan's core industrial capability. What is truly unsettling is not just capital flowing out, but whether capacity, talent, R&D, and bargaining power will eventually follow. If the most advanced technology and critical supply chains gradually shift westward, will the "silicon shield" that Taiwan has long relied on — the strategic interdependence built on the world's inability to tolerate instability across the Taiwan Strait — be eroded? Beijing's language about the deal "hollowing out Taiwan's economy" is precisely this anxiety pushed to its most extreme form ^[16].

The second reading is the "insurance" thesis. In this view, building factories in the U.S. is not being hollowed out — it is converting Taiwan's irreplaceability into deeper market ties and security commitments. Record exports to the U.S., a widening surplus, and America becoming a critical market together show that Taiwan is not conceding ground out of weakness, but being asked, after becoming too successful, to trade investment for more stable trade terms. As for the most explosive claim — moving "40% of the supply chain" — Taiwanese officials have publicly stated this is impossible; it looks more like an opening bid at the negotiating table than an industrial fact already on the ground ^[15].

The gap between these two readings is not only about the data — it is also about values. How you view "binding to the U.S." often determines in advance whether you read it as protection or dependency, as risk diversification or industrial hollowing-out.

But the question truly worth asking is not which slogan to pick, "drain" or "insurance," but what Taiwan is actually moving out, and what it is keeping. If what moves offshore is marginal capacity, while what stays behind is the most advanced processes, R&D capability, and supply-chain decision-making power, then the U.S. investment looks more like risk diversification. If, over time, even the technological lifeline, the center of gravity for talent, and R&D nodes gradually shift westward too, then the "drain" thesis would become a genuine structural warning.

So the key to judging this dispute is not political rhetoric, but what is happening on the construction site in Arizona: is what's being moved out there Taiwan's peripheral capacity — or the true summit of the Sacred Mountain?

TSMC's investment in Arizona has grown from an initially announced $12 billion to $165 billion, making it the largest greenfield foreign direct investment in U.S. history, and it has secured $6.6 billion in direct CHIPS Act grants ^[19]. Laying out the blueprint, TSMC plans to build six wafer fabs, two advanced packaging plants, and one R&D center there — an imposing lineup. Looking at that figure alone, the "drain" thesis seems to hold up.

But the devil, as always, is in the timetable. The first fab already reached volume production on the 4-nanometer process by the end of 2024, with yields matched to the Taiwan parent fabs, and monthly capacity around 90,000 to 100,000 wafers by the end of 2025; but the second fab (3nm) will not have tools installed until summer 2026, with production not starting until 2027, and the third fab (2nm/A16) has only just broken ground ^[20].

What this timetable reveals is critical: the most cutting-edge processes and R&D remain firmly rooted in Taiwan for the foreseeable future; what moved to the U.S. first is mature advanced-process technology trailing Taiwan's home fabs by at least one generation. As of mid-2026, what Taiwan has moved offshore is "marginal capacity," while what it has kept is the "lifeline." This is strong counter-evidence against the "drain" thesis, and support for the "insurance" thesis.

Of course, this is not a settled conclusion. The third fab is accelerating, and AI demand keeps pulling domestic U.S. manufacturing forward; over the long run, whether the lifeline eventually follows the capacity westward remains an open question. But at least for the foreseeable future, that construction site in the desert tells us: the summit of the Sacred Mountain is still on the island.

Export strength carries another, easily overlooked side effect: the exchange rate. This is the year's other hidden wound for Taiwan.

In early May 2025, the New Taiwan Dollar surged 8% to 10% against the U.S. dollar within just two trading days — the sharpest move in decades; by mid-year it touched a three-year high of roughly 29.85 to the dollar, a complete reversal from its depreciation track in 2024 ^[13]. On the surface, a strong currency is a medal for a strong economy; but for exporters, it is a whip.

Appreciation makes Taiwanese products more expensive in international markets and squeezes competitiveness. Many exporters and life insurers had not fully hedged — life insurers bore the brunt: a sharp NT dollar appreciation can immediately produce large unrealized currency-translation losses on their books, while for exporters the impact shows up mainly as compressed gross margins rather than directly booked losses ^[14]. The impact is especially pronounced for the electronic-components makers who dominate Taiwan's exports. The result is an ironic loop — the massive surplus that AI built up gets partially clawed back out of exporters' other pocket through currency appreciation. Buried inside the glittering bottom line is this invisible cost that never makes headlines.

Up to this point, this report has largely been about chips. But the group Taiwan's economy most needs to worry about is exactly the one the cameras never point at.

Many people do not realize that most of Taiwan's exports are not actually semiconductors. Servers, networking equipment, machinery, auto parts, petrochemicals, plastics and rubber — these non-semiconductor industrial goods are spread across manufacturing clusters all over the island, behind which sit tens of thousands of small and medium enterprises (SMEs). According to FPRI's analysis, the U.S. has been Taiwan's top goods export market since early 2024; for these SMEs, dependence on the U.S. market is generally elevated (observation/inference, confidence: low — the precise dependence ratio awaits official statistical verification), and when tariffs hit, they land directly, with no buffer ^[26].

Take Taichung's precision machinery cluster as an example. It is home to Taiwan's densest concentration of machine-tool and component makers — high export share, thin margins. When the reciprocal tariff briefly pushed the machine-tool rate from 4.4% to 32%, it was close to a matter of survival for these firms; now that the 15% non-cumulative cap has pulled them back from the cliff's edge, the Taiwan Association of Machinery Industry (TAMI) describes it as a crucial step that "pulls the competitive field back onto the same starting line as Japan, Korea, and Europe," and auto-parts makers are likewise gearing up to chase back orders now that the rate has stabilized ^[17].

There are also numbers pointing to resilience: a CPA Australia survey found that more than half of Taiwan's small businesses still grew in 2025, and operating confidence for 2026 hit its highest point since 2020, with more than 60% expecting growth ^[18]. The government has also used matching credit guarantees to help SMEs lower the risk of investing in the U.S. and navigating trade uncertainty.

But stopping the bleeding is not the same as healing. The fate of these clusters hinges on a tariff framework due to expire in July; they enjoy none of the "irreplaceable" bargaining power that semiconductors have, yet they must directly absorb the double squeeze of tariffs and currency appreciation. While all eyes look up at the Sacred Mountain, the shadow at its foot is the real test of Taiwan's economic resilience.

Chips decide how bright Taiwan shines. SMEs decide how much it hurts. Any policy that conflates the two is bound to end up hurting the latter.

Return to the contradiction we started with. Taiwan is neither a pure winner nor a drained victim — it is, in this era of de-risking, the party simultaneously receiving the greatest dividend and bearing the greatest strain. Facing this two-faced position, different actors need different stances.

For the state, the task is to convert this bargaining chip into "binding," not "hollowing out." U.S. investment does bind Taiwan more deeply into America's markets and security net, but holding onto domestic talent, energy, R&D, and the most advanced process nodes is a non-negotiable bottom line. "Moving 40%" can be a subject of negotiation; the lifeline cannot be surrendered.

For industry intermediaries (such as trade associations and startup federations), the task is to issue cluster-specific warnings rather than dismiss every member with one blanket line about "industry upside." The semiconductor chain is relatively protected, but the non-semiconductor SMEs across these manufacturing clusters are where the real pressure sits; helping them diversify away from single dependence on the U.S., open up third markets that are neither China nor the U.S., and reminding them to "genuinely relocate capacity, not just change the postmark," is the practical form of guidance they need.

For SMEs themselves, the task is first to treat "elevated dependence on the U.S. market" as a known, manageable risk to be quantified, to make full use of government credit guarantees, to get ahead of "China+1" orders from ASEAN, India, and Mexico, and at the same time to steer clear of the compliance minefield of pure transshipment. In an era when the rules can change overnight, betting everything on a single tariff environment is too dangerous.

And the shared deciding hand for everyone is a handful of timelines that have yet to play out: when China's advanced semiconductors will truly achieve self-sufficiency, what happens to that temporary tariff statute after July, and whether TSMC Arizona's lifeline will eventually follow its capacity westward after all. The answers to these three questions will decide which face of "Taiwan's two faces" ultimately wins out.

The answer has not yet been written. But the chips are already on the table — and what Taiwan is betting on this hand is not luck. It is time.