How the AI Build-Out Rewired Global Goods Trade in 2025

In the second quarter of 2025, Taiwan's AI-related exports to the United States reached the equivalent of roughly 14 percent of Taiwan's entire gross domestic product, according to a Federal Reserve FEDS Note published in February 2026. One country. One quarter. One bilateral flow. Accounting for a share of GDP that a generation ago would have been reserved for a country's entire export basket.

That single data point is the cleanest signature of a larger shift: the capital expenditure wave behind artificial intelligence has stopped being a stock-market story and has become the dominant force reshaping what crosses borders. The World Trade Organization's latest outlook, published on 20 March 2026, reports that the share of AI-enabling goods in world trade climbed from around 13 percent in 2023 to nearly 17 percent by the end of 2025, with trade in these products growing 21.9 percent year-on-year in 2025 — a category that the WTO says "accounted for almost half of global trade growth in 2025."

The rest of the traded economy is slowing, fragmenting, and being taxed more heavily. The AI slice is growing, concentrating, and — so far — exempt from most of the friction. Understanding how that split formed, who benefits, and what could unwind it is now the central question for anyone modeling the next two years of cross-border commerce.

The Size of the Shift

Start with the headline numbers. World merchandise trade volumes expanded by 4.6 percent in 2025, the WTO's March 2026 outlook confirms. That was well above the organization's own forecast from a year earlier, and the quarterly data are consistent with the upgrade: in its Q3 2025 statistics release, the WTO reported that the US dollar value of merchandise trade reached an all-time high, with dollar-value growth of 7.5 percent year-on-year in Q3 2025 and volume growth of 3.6 percent year-on-year.

Underneath that headline, the composition change is where the story lives. The WTO's October 2025 update tracks roughly 100 AI-related product lines — a basket that runs from graphics processors and advanced memory to networking switches, servers, optical transceivers, and semiconductor manufacturing equipment. Trade in that basket reached USD 1.92 trillion in the first half of 2025 alone, up from USD 1.61 trillion in the first half of 2024. Extended to a full-year view and widened slightly, the March 2026 WTO outlook puts the 2025 total value of AI-enabling goods trade well above levels that, as recently as the WTO's World Trade Report 2025, were summarized as "USD 2.3 trillion in 2023."

The scale matters because it forces a reinterpretation of every aggregate. A global trade forecast that treats AI-enabling goods as a rounding error will miss both the upside and the concentration risk. The WTO's current baseline for 2026 merchandise trade growth is 1.9 percent, with a 2027 pickup to 2.6 percent. Strip out AI-linked hardware, and the non-AI trade baseline almost certainly implies something closer to a stall.

Data Centers as a Trade Policy

The AI-goods surge is not a technology story that happens to show up in trade statistics. It is a consequence of one specific capital expenditure decision, replicated at scale across a small set of US-headquartered hyperscalers.

The Federal Reserve's FEDS Note puts the direct figure bluntly: "U.S. data-center spending alone is expected to exceed half a trillion dollars in 2025." That capex has to land somewhere in physical form — racks, accelerators, high-bandwidth memory, power-delivery components, cooling infrastructure, networking gear. Roughly none of it is manufactured entirely inside the United States. The result, the Fed notes, is that "AI-related trade" — using its narrower product definition — reached more than USD 272 billion in the first half of 2025, a 65 percent increase over the first half of 2024.

The Fed's measurement methodology differs from the WTO's — it uses a tighter product list focused on imports directly tied to US data-center deployment, which is why its share of total merchandise trade runs closer to two percent compared with the WTO's nearly 17 percent. The two frames are complementary, not contradictory. The Fed is measuring the narrow corridor through which hyperscaler dollars become international shipments. The WTO is measuring the full set of goods whose demand curve has been bent by AI.

In practice, that corridor passes through a very short list of economies. The Fed's note describes the United States and China as having driven a sweeping expansion in AI-related computing capacity in 2024, with the US accelerating further through mid-2025. Beyond the two end-markets, the pass-through of US data-center capex into export revenue has landed heavily on Taiwan, South Korea, Vietnam, and Mexico — a finding that an Oxford Economics briefing describes as significant international spillovers transmitted through electronics imports, with the largest effects observed in China, Taiwan, South Korea, Vietnam, and Mexico.

The Asia Concentration

If AI goods are the engine, Asia is the piston. The WTO's March 2026 outlook reports that Asia contributed 71 percent of total merchandise trade growth in 2025 and accounts for 62 percent of all AI-enabling trade. The Q3 2025 statistics release shows Asian export volumes running 9.5 percent ahead year-to-date through September 2025 — several multiples of the growth rates recorded in any other region.

The concentration runs deeper than the regional aggregate suggests. TSMC in Taiwan and SK hynix and Samsung in South Korea sit at the apex of the advanced-logic and high-bandwidth-memory supply chains respectively. Vietnam and Mexico are increasingly the assembly and final-pack nodes where US-destined systems are integrated. Chinese factories still supply an enormous share of standard components — power subsystems, cabling, lower-memory tiers, basic electronics — even as the advanced-logic trade with the United States has been explicitly curtailed.

Three second-order effects deserve attention. First, the bilateral AI-goods flow is carrying exchange-rate and macro implications for the supplier economies that dwarf the usual cyclical drivers. The Taiwan-US AI export corridor hitting roughly 14 percent of Taiwanese GDP in a single quarter, per the Fed note, is the kind of number that redefines the country's balance-of-payments profile. Second, the concentration runs through a small number of firms whose capacity expansion timelines are already priced several years forward. Third, shock propagation in that network has a different geometry than in a diversified manufacturing base: a single fab disruption, export-control announcement, or shipping-lane incident cascades directly into global AI deployment schedules.

The Parallel Story: Trade Fragmentation

While the AI corridor has been widening, the broader trade system has been narrowing. The WTO's March 2026 outlook reports that US imports from China fell 29 percent in 2025 — a contraction on a bilateral flow whose absolute size still makes it one of the largest in the world. The same outlook records that the share of world trade conducted under most-favoured-nation tariff treatment fell from 80 percent in 2024 to around 72 percent by early 2026 — eight percentage points of the trading system moving out of the baseline rulebook in roughly a year.

The World Trade Report 2025 adds the regulatory counterpoint: quantitative restrictions specifically on AI-related goods rose from 130 in 2012 to nearly 500 in 2024. A large share of that increase is recent — export controls, foreign-direct-investment screens, outbound-investment limits, and due-diligence regimes layered onto advanced semiconductors, manufacturing equipment, and the inputs that feed them.

The coexistence of a booming AI-goods flow and a fragmenting trade system is not accidental. The WTO notes in its March 2026 commentary that key AI-enabling goods — chips, semiconductors, data transmission equipment — are "exempt from most new tariffs," a carve-out reflecting the practical reality that imposing import friction on the hardware running domestic AI strategy would be self-defeating. The fragmentation is happening around the AI corridor, not through it.

The Middle East Overlay

The March 2026 WTO outlook introduces a second downside scenario overlaid on the baseline: the ongoing conflict in the Middle East and its effect on energy and transport costs. Under an adverse oil-price scenario, the WTO estimates that merchandise trade growth could fall by 0.5 percentage points — from 1.9 percent to around 1.4 percent — with services trade slipping from 4.8 percent to 4.1 percent.

For AI-goods trade specifically, the channel is indirect but non-trivial. Electronics shipments do not run on oil prices the way bulk commodities do, but container shipping rates, air-cargo premiums (still the primary mode for high-value semiconductors), and insurance premia on Suez and Red Sea routings do. A protracted energy-price spike also tightens financial conditions broadly, which historically has compressed hyperscaler capex plans at the margin even when the strategic AI commitments remain unchanged.

The more acute risk is that the energy overlay combines with the tariff overlay. If the MFN share continues to erode in 2026 on the trajectory implied by the WTO data — from 72 percent toward the mid-sixties — and the Middle East adverse scenario materializes, the non-AI portion of trade would face simultaneous cost and policy pressure. A flat aggregate could mask a two-track system: AI goods still growing at double digits while everything else contracts outright.

The 2040 Extrapolation

The WTO's World Trade Report 2025 makes a longer-term case that the current surge is not a one-time infrastructure pulse. Under its modeling, successful diffusion of AI could lift global trade by 34 to 37 percent by 2040 across scenarios, with global GDP rising 12 to 13 percent over the same horizon. The report also quantifies the distribution conditionally: if lower- and middle-income economies close half of their digital infrastructure gap, incomes in those groups rise 15 percent and 14 percent respectively.

Those numbers are modeled, not observed, and the WTO is explicit that the outcomes depend on sustained digital-trade liberalization and infrastructure investment. But the direction of the modeling aligns with what the 2025 data already show: AI's trade footprint is expanding faster than the overall economy can absorb, which means the AI-trade share will keep rising unless something actively breaks it.

What Could Go Wrong

Four channels sit between the current trajectory and the 2026–2027 baseline.

Export-control escalation. The semiconductor export controls, foreign-direct-product rule extensions, and FDPR-equivalent measures from multiple jurisdictions have so far targeted the most advanced nodes. If the perimeter widens to include trailing-edge equipment or adjacent materials — substrates, packaging chemistries, rare gases — the supply-side effect on AI-goods trade volume could be material. The WTO has documented the direction of travel; the magnitude is the open variable.

Capex reset. The Fed's over-half-a-trillion-dollar 2025 data-center spending figure is concentrated in a handful of firms whose spending plans can be revised quickly if return expectations deteriorate. An investment-cycle pause of even one quarter translates almost mechanically into a multi-billion-dollar hit to Taiwanese, Korean, and Mexican AI export revenues, given the capex-to-trade pass-through visible in the 2025 data.

Energy and logistics shock. Beyond the WTO's modeled 0.5-percentage-point oil-scenario hit, a prolonged disruption of Red Sea or East Asian shipping lanes would impose out-of-cycle costs on exactly the corridors where AI goods flow. Most advanced AI hardware moves by air for the final legs, but inputs and finished systems at scale still rely on maritime routes.

MFN erosion acceleration. The eight-percentage-point drop in MFN share between 2024 and early 2026 was driven largely by bilateral tariff measures on specific product categories. If that pace continues — another six to eight points in 2026 — the compliance and routing costs on non-AI trade rise sharply, and the political pressure to extend friction into AI-adjacent categories will also rise.

Implications for the Next Two Years

For policymakers, the current configuration is unstable in a specific sense: a system cannot indefinitely sustain an eight-point drop per year in MFN coverage while also protecting a rapidly growing category from any friction at all. Either the carve-outs broaden (which would require renewed multilateral commitment to digital-trade rules) or the friction eventually migrates into the AI corridor (which would slow the component of global trade that is currently doing the heavy lifting).

For investors in trade-exposed sectors, the 2025 data argue for a barbell view. AI-adjacent capex exposures — semiconductor equipment, advanced packaging, networking, power infrastructure — are riding a flow that the Fed and WTO now describe in similar terms. Non-AI exporters face the opposite picture: a WTO baseline that implies sub-two-percent merchandise trade growth in 2026, with downside if the energy scenario materializes, and a shrinking protected zone inside the trading system.

For the trade-dependent economies themselves, the concentration risk is now measurable. Taiwan's position — with AI-related exports to the US alone touching roughly 14 percent of GDP in a single quarter per the Fed — is the clearest illustration, but South Korea and Vietnam sit on similar vectors of exposure. The upside case is explicit; the downside case is one hyperscaler capex memo away.

Key Takeaways

• AI-enabling goods climbed from around 13 percent of world trade in 2023 to nearly 17 percent by end-2025, growing 21.9 percent year-on-year in 2025 and accounting for almost half of global trade growth, per the WTO's March 2026 outlook.
• US data-center spending exceeded half a trillion dollars in 2025, pulling more than USD 272 billion of AI-related goods through cross-border trade in the first half of the year alone — a 65 percent year-on-year jump, according to the Federal Reserve.
• Asia accounts for 62 percent of AI-enabling trade and contributed 71 percent of total merchandise trade growth in 2025 (WTO); Taiwan's AI exports to the US alone touched roughly 14 percent of its GDP in Q2 2025 (Fed).
• The surrounding trade system is fragmenting: the share of world trade under most-favoured-nation treatment fell from 80 percent in 2024 to around 72 percent by early 2026, US imports from China fell 29 percent in 2025, and AI-specific quantitative restrictions climbed from 130 in 2012 to nearly 500 in 2024 (WTO).
• A Middle East energy-price scenario could shave 0.5 percentage points from the WTO's 1.9 percent baseline 2026 merchandise trade forecast, with compounding risk if MFN erosion continues (WTO).