Taiwan's Ministry of Finance announced Friday that the island's exports reached US$75.30 billion in July — the third-highest monthly total in the country's history — as integrated-circuit shipments hit an all-time high and AI server demand kept every major technology sector growing in tandem. The Ministry of Finance announced the figures at a Taipei press conference by Department of Statistics Director-General Beatrice Tsai, marks the 33rd consecutive month of year-on-year growth and lands two weeks after a new US tariff regime took effect — one that currently exempts semiconductors but whose companion overcapacity investigation, specifically targeting Taiwan's chip industry, has not yet issued its ruling.

The July figure, up 32.9% from a year earlier, trails only March's US$80.18 billion and May's US$78.48 billion all-time record. In the first seven months of 2026, Taiwan's total exports reached US$491.95 billion — up 44.7% year-on-year — producing a cumulative trade surplus of US$114.61 billion, itself up 63.5% from the same period in 2025. At the current pace, Taiwan is on track to surpass the government's full-year export forecast of US$894.5 billion, which the Directorate General of Budget, Accounting and Statistics revised upward in May 2026.

AI Servers and Chips Dominate July, Two Sectors Account for Nearly Four in Five Dollars

July's headline number rests almost entirely on two categories. Integrated circuits — the semiconductors that power every AI accelerator, GPU, and AI server — set a fresh monthly export record at US$26.30 billion, up 52.3% year-on-year, driving the broader electronic components category to US$27.73 billion (+50.5%). Information, communications, and audio-video products — the AI servers, switches, routers, and networking hardware that data centers worldwide are deploying at a pace that has surprised even the most optimistic analysts — contributed a further US$31.37 billion, up 29.5% year-on-year. Together, the two categories accounted for 78.5% of all July exports.

That concentration figure — nearly four of every five export dollars flowing through two technology categories — is the most important single number in the July data, though not for the reason the headline suggests. It is not merely a sign of strength; it is a structural measure of how exposed Taiwan's economy has become to a single demand vector. A meaningful deceleration in hyperscaler AI capex, a Section 232 action applied to servers, or an adverse overcapacity ruling that imposes tariffs on finished AI hardware would land on an economy that has steadily narrowed its cushion in traditional sectors. Transportation vehicles fell 0.7% in July; textiles dropped 9.1%. The AI premium has not lifted those sectors. It has crowded them.

Tsai attributed July's performance to three reinforcing factors: robust global demand for AI applications, the resilience of the broader global economy, and a seasonal dynamic in which international brands build up inventories ahead of major consumer electronics product launches. The third driver is worth watching: the consumer electronics pre-positioning visible in July's data suggests the ICT/AV category's strength has both a structural component (AI infrastructure) and a cyclical one (inventory builds) — an important distinction for anyone reading these numbers as a guide to Q4 AI spending.

Machinery exports rose 10.8% to US$2.55 billion, reflecting semiconductor firms purchasing equipment for ongoing capacity expansions. Electric equipment climbed 34.3% to US$1.69 billion, as countries building out power grids for AI data center deployments lifted demand for grid-related hardware.

Why These Numbers Are TSMC's Numbers

Behind Taiwan's export data is a single company that now functions as the world's most critical AI infrastructure supplier: Taiwan Semiconductor Manufacturing Company. TSMC fabricates the advanced processors at the heart of every major AI accelerator — from Nvidia's Blackwell-generation GPUs to the custom silicon designed by Google, Amazon, Microsoft, and Meta — and its capacity constraints have become the defining supply-chain story of the technology industry.

TSMC's second-quarter 2026 revenue reached US$40.2 billion, up 36% year-over-year, as the company raised its full-year growth guidance to above 40% — its second upward revision of the year. For the third quarter, TSMC guided revenue of US$44.6 billion to US$45.8 billion, implying roughly 37% year-over-year growth at the midpoint. The company also raised its 2026 capital expenditure guidance to US$60 billion to US$64 billion — up to 14% above its prior ceiling — with the additional spending flowing almost entirely into meeting AI-related demand.

Chairman and CEO C.C. Wei told investors on the company's second-quarter earnings call that demand for AI infrastructure has exceeded the company's own projections. At the company's annual shareholders meeting in June, Wei described advanced packaging capacity as extremely tight and sold out through 2026.

High-Performance Computing — the segment anchored by AI accelerators for cloud data centers — rose 20% sequentially in Q2 2026 and now accounts for 66% of TSMC wafer revenue. Smartphones, which generated the largest share of TSMC's revenue as recently as 2022, fell to 22%. That revenue-mix shift is the clearest single measure of how thoroughly AI has restructured global semiconductor demand — and how directly Taiwan's trade numbers reflect the decisions of a small number of US hyperscalers.

What CoWoS Packaging Means for the Supply Chain — and Why It Still Can't Keep Up

Understanding why the AI-driven export cycle propagates through every layer of Taiwan's technology supply chain — not just chip fabrication — requires understanding what happens between a wafer coming out of TSMC's fab and an AI accelerator landing in a data center rack.

TSMC's proprietary CoWoS (Chip-on-Wafer-on-Substrate) advanced packaging technology places a logic die — the GPU or AI ASIC — and high-bandwidth memory (HBM) stacks on a silicon interposer, enabling data to flow between them at speeds and densities impossible with conventional 2D packaging. Nvidia's Blackwell-generation GPUs require memory bandwidth exceeding 9,000 gigabytes per second; physically achieving that bandwidth requires the logic and memory dies to be separated by microns rather than millimeters. Only CoWoS does that at production scale today.

TSMC has been expanding CoWoS capacity from approximately 35,000 wafer starts per month in late 2024 toward a targeted 120,000 to 140,000 wafer starts per month by year-end 2026 — a roughly fourfold increase, according to TrendForce capacity analysis. Even at that scale, CoWoS remains the primary bottleneck for AI chip delivery, because specialized equipment required for packaging facilities — thermal compression bonders, ultra-precise pick-and-place machines — carries 12 to 18-month lead times that capital alone cannot compress. Nvidia alone holds approximately 60% of TSMC's CoWoS allocation, with the top three customers collectively accounting for more than 85% of available capacity.

The supply constraint runs deeper than packaging. Under the CoWoS interposer sits an ABF (Ajinomoto Build-up Film) substrate — the organic circuit board that connects the completed chip package to the server's main board. Ajinomoto controls more than 95% of the global supply of ABF film and raised its prices 30% effective the third quarter of 2026. Lead times for ABF substrates have extended from a normal 12 to 16 weeks to 20 to 30 weeks, with Taiwan's major substrate makers — Unimicron, Nan Ya PCB, and Kinsus Interconnect Technology — operating at full capacity with AI chip programs pre-allocated through year-end. Ajinomoto's investment in a third ABF production facility targets a completion date of 2032, which means the material-level constraint cannot be resolved this decade regardless of capital commitment.

This is the engineering reality that explains why July's strong export data runs across multiple sectors simultaneously — and why no single capacity addition at any one layer will resolve the AI chip supply constraint. The system is constrained at a layer below the layer most observers are watching.

Where US-Taiwan Trade Policy Stands After July

Taiwan's record-setting export performance is occurring in a more complicated trade environment than the headline numbers alone suggest, and the composition of that environment matters.

On the favorable side: the United States and Taiwan reached a deal in January 2026 that set a 15% tariff rate — down from a 32% reciprocal tariff imposed under the Trump administration — and that deal has provided meaningful relief for some Taiwanese exporters. On July 24, the Section 301 Forced Labor tariffs took effect, applying a 10% duty (net of most-favored-nation rates) to certain goods from Taiwan. That rate was the most favorable tier among major trading partners — lower than the 12.5% imposed on Japan and South Korea, and far lower than the rates applied to China and Vietnam. Critically, semiconductors and 1,909 product categories covering Taiwan's most important technology exports are currently on the exemption list.

But the January deal explicitly excludes electronics, semiconductors, and goods subject to Section 232 national security review — meaning the categories responsible for 78.5% of Taiwan's July exports are not covered by the 15% rate. And a separate Section 301 investigation — launched in March 2026, targeting structural excess manufacturing capacity in 16 economies including Taiwan, specifically naming semiconductors, electronic products, and IT products as sectors under scrutiny — was expected to conclude by the end of July. No public announcement of a ruling from that investigation has been confirmed as of August 7.

Taiwan Institute of Economic Research economist Liu Pei-chen has said that the overcapacity probe poses a bigger source of uncertainty for Taiwan than the Forced Labor tariff, adding that advanced-node foundries should be less affected while mature-node foundries face greater exposure. Chang Chien-yi, TIER president, noted that Taiwan exports relatively few semiconductors directly to the United States — many are shipped to other countries for assembly into finished technology products — but that the impact could become more significant if Washington imposed tariffs on finished products containing chips not manufactured in the United States.

Director Tsai acknowledged at the August 7 press conference that geopolitical tensions and US trade policy uncertainty represent the primary risks to Taiwan's otherwise strong export outlook.

For the United States specifically: US-bound goods reached US$23.24 billion in July, up 25.2% year-on-year — but that 25.2% rate was the slowest pace of growth in 18 months, attributable in part to a decline in exports of computer peripherals and components to the US market. China and Hong Kong remained the second-largest market at US$19.17 billion.

What the Ministry Expects Next

The Ministry of Finance is projecting that August exports will continue the expansion, forecasting US$76.6 billion to US$78.9 billion — up 31% to 35% year-on-year. If that forecast holds, Taiwan's cumulative eight-month export total for 2026 will approach or exceed US$570 billion, maintaining the pace needed to reach the full-year targets that several institutions revised upward in the spring.

The structural AI demand that is driving this trajectory shows no clear sign of abating. Taiwan's government revised its 2026 GDP growth forecast to 9.64% in May 2026 — its fastest projected annual expansion in 16 years — driven primarily by exports. The same hyperscaler capital expenditure commitments that produced Taiwan's record months in Q1 2026 are still running through the supply chain. TSMC's Q3 revenue guidance alone ($44.6 billion to $45.8 billion) implies Taiwan's foundry revenue for the quarter will sustain the export volumes the Ministry of Finance is projecting.

The question that the August, September, and October data will answer is whether the AI infrastructure cycle is entering a digestion phase — a period in which hyperscalers absorb the capacity they have built before ordering more — or whether the demand for AI compute continues to surprise to the upside. TSMC's management has consistently guided toward the latter view. The July trade data, the third-best month in Taiwan's history, is consistent with that view holding for now.

Can This Pace Continue? The Risks Are Real

No economic reading of these numbers is complete without naming the structural risks that sit alongside the structural strengths.

Taiwan's economy — in which exports represent roughly 70% of GDP — is more exposed to a single demand cluster than at any point in its postwar history. The 78.5% export share concentrated in two technology categories means that any demand shock in AI capex would arrive in an economy with limited absorptive capacity in other sectors. Traditional industries — transportation vehicles, textiles, chemicals — have not kept pace with the AI-driven surge, and their relative contribution to exports has narrowed.

The pending Section 301 overcapacity investigation is the single most consequential unresolved regulatory risk. If that probe results in tariffs applied to finished technology products containing Taiwan-fabricated chips — the scenario TIER's Chang Chien-yi identified as the more significant exposure — the impact would arrive directly in the export categories that have driven 2026's records. The investigation explicitly names semiconductors and electronics as sectors under scrutiny.

On the supply side, the CoWoS substrate and ABF material constraints described above represent an engineering ceiling that capital cannot immediately lift. Ajinomoto's third ABF plant does not come online until 2032. TSMC's Arizona packaging facilities — the first CoWoS-capable facilities on US soil — are not targeted for mass production until 2028. The near-term constraint is real, and it means that even robust demand may produce slower-than-expected revenue growth if the bottleneck tightens further.

Geopolitical risk is present but worth calibrating carefully. Taiwan's Pacific shipping lanes — the alternate routes that semiconductor exports use when the Taiwan Strait is deemed too exposed — have been the subject of increased Chinese coast guard activity in recent months, a development that analysts monitor as a coercive pressure tool rather than an imminent shipping disruption. The risk is real; its current magnitude does not match the severity of the supply-chain disruption scenario.

For investors who have watched the iShares MSCI Taiwan ETF outperform significantly in 2026 as markets priced Taiwan as the physical backbone of the AI economy, the July trade data confirms the thesis that drove that performance. It does not resolve the regulatory question that is still outstanding.

---

Frequently Asked Questions

Why did US-bound exports from Taiwan slow to their weakest pace in 18 months, even as overall exports hit a near-record?

July's US-bound shipments grew 25.2% year-on-year — strong in absolute terms, but the slowest pace of growth Taiwan has posted with the United States in 18 months. The Ministry of Finance attributed the deceleration primarily to a decline in exports of computer peripherals and components to the US market, suggesting that server and networking hardware (which is often shipped to other countries for integration before reaching US data centers) continued growing while consumer-oriented technology hardware slowed. This distinction matters for anyone reading Taiwan's US-trade figures as a direct measure of American AI investment: much of the AI infrastructure hardware that ends up in US data centers passes through intermediate manufacturing and assembly steps in countries other than Taiwan before final delivery.

Will the Section 301 overcapacity investigation affect Taiwan's semiconductor exports?

The investigation, initiated in March 2026 and expected to conclude by the end of July, has not publicly confirmed a ruling as of August 7. Taiwan's economists at the Taiwan Institute of Economic Research have said that if the probe results in action, advanced-node foundries — the segment that produces the cutting-edge chips responsible for most of July's IC export record — should be less affected than mature-node producers, because the overcapacity concerns are concentrated in older processes. The more significant scenario, per TIER president Chang Chien-yi, is if Washington eventually imposes tariffs on finished technology products containing chips not manufactured in the United States — a step that would reach directly into the AI server and networking hardware that drove July's ICT/AV export category.

What does TSMC's CoWoS bottleneck mean for how long AI export growth can continue?

CoWoS (Chip-on-Wafer-on-Substrate) is the advanced packaging step that physically integrates AI accelerator chips with high-bandwidth memory — without it, no Nvidia, AMD, or hyperscaler custom AI chip can be delivered at production scale. TSMC is expanding CoWoS capacity from roughly 35,000 wafer starts per month in late 2024 toward 120,000 to 140,000 by year-end 2026, a fourfold increase that still leaves demand exceeding supply. The deeper constraint is the ABF substrate material that sits beneath the CoWoS interposer — manufactured almost entirely by a single Japanese company, Ajinomoto, whose third production plant does not come online until 2032. This means that AI chip delivery timelines are constrained at a material level that no amount of foundry capital spending can quickly resolve, and that the export volumes Taiwan is currently posting reflect a supply ceiling as much as a demand floor.

What does Taiwan's export concentration mean for consumers buying AI-dependent products?

When 78.5% of a major exporting nation's outbound shipments flow through two technology categories — integrated circuits and ICT/AV hardware — the supply chain for those products becomes highly sensitive to any disruption at the top. TSMC's growing pricing power on its advanced process nodes, combined with rising substrate and packaging costs, tends to flow into retail prices for flagship smartphones, laptops, and AI-capable consumer devices within two to three product cycles. Consumers planning major technology purchases in the second half of 2026 are buying into a supply chain that is running near capacity at every layer. Prices for high-end consumer devices are more likely to rise than to fall over the next two to four quarters.