The World Grows 3%, Taiwan Grows 9.64%: In the Tug-of-War Between AI and War, How Long Can This High Growth Last?

Before discussing high growth, it helps to clear up the figure most likely to cause confusion.

The 14.55% published by the DGBAS is the preliminary year-on-year growth rate of real GDP for the first quarter of 2026 — it describes something that has already happened, though the statistic may still be revised later. 9.64% is the forecast for full-year 2026, built on assumptions about exports, consumption, investment, and prices for the remaining three quarters. ^[3]

That same DGBAS forecast already has a slowdown written into it: 12.65% growth projected for the first half and 6.94% for the second half. ^[3] In other words, even if the full year does end up close to 9.64%, the growth path does not hold the first quarter's pace in every quarter. A low base at the start of the year, plus rising AI hardware shipments and prices, pushed the early figures especially high; once the base rises in the second half, the year-on-year rate naturally comes down.

This distinction has direct consequences for decision-making. If the government treats the full-year forecast as a stable new normal, it may overestimate future tax-revenue growth; if companies simply extrapolate the first-quarter year-on-year rate, they may plan plants and staffing around an unusually steep demand curve; and if households read the headline figure as a promise that every industry will get a synchronized pay raise, they will be mistaking a statistical forecast for a personal income guarantee.

High growth is a real signal, but it is first and foremost a map that has to be read in pieces.

What a global average most easily hides is the divergence taking place among individual countries.

The IMF notes that 2026 global growth is being dragged down at one end by war, energy, and inflation, and propped up at the other end by technology investment. ^[1][2] Economies with a high share of energy imports and little fiscal room have to pay more for oil and gas while lacking the budget to subsidize households and businesses over the long term; economies positioned in the AI hardware, software, data-center, or power-equipment supply chain, by contrast, are receiving a fresh round of capital spending.

So "the world is growing only 3%" does not mean every country is slowly walking at 3%. Some economies may be close to stagnation, while some technology-exporting economies are running well above the average. Taiwan, South Korea, Japan, and the U.S. market carry greater technology exposure, and their stock markets and investment are also more readily supported by AI expectations; on the other hand, should a profit re-rating occur in these markets, the wealth effect and corporate investment could reverse together just as readily. ^[2][7]

Here, the average sits like the spot where the rope stops after a tug-of-war. It tells you where things ended up, but not how hard each side was pulling.

A Taiwanese company that looks only at the global 3% will underestimate the strength of orders in the AI supply chain; one that looks only at Taiwan's 9.64% will underestimate the headwinds from energy, interest rates, and non-tech demand. The correct reading puts both forces into the operating plan at the same time.

First-quarter data gives a clear answer.

The DGBAS notes that demand for AI and related infrastructure drove first-quarter real exports of goods and services up 35.76%, and manufacturing up 26.18%; semiconductors, computers, and electronic and optical products were the main sources. Gross capital formation also grew 5.92%, reflecting increased investment in machinery and equipment, intellectual property, and transport equipment. ^[3]

The Ministry of Finance's (MOF) merchandise-trade statistics follow the same line. Exports in June 2026 came to US$74.83 billion, up 40.3% year on year; first-half exports came to US$416.66 billion, up 47.1% year on year. ^[4] Imports likewise surged, with June imports of US$62.63 billion, up 51.8% year on year, part of which reflects raw materials, equipment, and components purchased for subsequent production. ^[4] A rise in imports is not necessarily bad news — it can mean factories are restocking inputs for future shipments.

The real engine is not a single end product but an entire set of capital goods spanning wafer fabrication, packaging and testing, memory, server assembly, switches, cooling, power supplies, and equipment investment. As companies worldwide build out compute capacity at the same time, Taiwan is not supplying one item — it is supplying most of the hardware needed to build a digital factory.

This also explains why the growth is so concentrated. Data-center investment involves large individual sums, long supply chains, and high technical barriers; as soon as a handful of global customers expand capital spending at the same time, export and manufacturing figures rise quickly. Conversely, as soon as those same customers shift their pace of expansion from "accelerate" to "maintain," Taiwan's year-on-year growth rate can also fall sharply, even while the absolute value of orders stays high.

This round of Taiwan's growth has at least three layers.

The first layer is shipment volume. Cloud service providers need more computing equipment, driving shipments of chips, servers, and networking hardware. The second layer is product prices and mix. High-end AI servers, advanced chips, and high-speed memory carry higher unit prices; even without a proportional increase in unit shipments, export value can rise simply because the product mix has upgraded. The third layer is investment to expand capacity. Companies purchase equipment, add R&D, and build factories in anticipation of future orders, giving current-period GDP another push. ^[3][4]

When all three layers rise at once, the headline numbers look excellent. But they do not reverse the same way.

Volume can slow as customers work through inventory; prices can fall as supply and demand loosen or new competitors enter; and investment can naturally taper off once capacity has been built out. None of these three shifts means AI demand has disappeared, yet each can bring the year-on-year growth rate for GDP and exports down quickly.

So judging how far this high growth can run cannot simply be a matter of asking "will AI keep growing?" The sharper questions are: how much new compute demand are customers actually adding? How is unit hardware pricing and product mix changing? And how fast do suppliers still need to expand capacity? These three answers will not all arrive at the same time.

Strong headline growth also creates a gap between the statistics and how things feel on the ground.

First-quarter private consumption grew 4.74%, which is not weak; the DGBAS attributes this to information and communication, entertainment, transport, outbound travel, and the fees and wealth effect that came with an active stock market. ^[3] But this pace is still clearly below that of exports and manufacturing. High growth first shows up in the capital-intensive, export-oriented tech chain, and only gradually spreads outward through wages, dividends, tax revenue, procurement, and consumption.

That spread does not happen evenly. The DGBAS reports that in May 2026, the average regular earnings of full-time domestic employees was NT$52,164, with a median of NT$41,050. ^[6] These two figures cannot be compared directly with the GDP growth rate, nor can they be used to conclude who has "taken" the growth; they simply remind us that average output, average pay, and the pay of a typical employee are, from the start, three different questions.

A company making advanced-packaging equipment may be racing to hire engineers and add capital spending, while a shop serving local consumers may instead be feeling rent, electricity, wages, and raw-material costs. Both companies sit inside the same 9.64% economy, yet their operating weather is completely different.

If policy communication only publishes the headline growth figure, people who have not benefited from it may come to feel the statistics are misleading; if companies use headline growth to estimate their own demand, they may mistake someone else's business cycle for their own order book.

The optimistic view of the AI boom rests on solid ground.

First, demand is not coming only from consumer-electronics upgrade cycles — it comes from businesses and governments building new computing infrastructure. Data centers, model training and inference, cloud services, sovereign AI, and enterprise applications all require sustained investment in hardware, software, power, and talent. ^[1][8]

Second, Taiwan holds technical and manufacturing-scale advantages across several key segments. Advanced process nodes, packaging and testing, server design and manufacturing, and component integration are not things that can be fully replicated in the short term with subsidies alone. Even if customers push to localize the supply chain, they still need Taiwanese companies to participate.

Third, if AI genuinely spreads into more industries, productivity gains could give the investment cycle its next wave of demand. IMF analysis holds that whether the benefit proves durable depends on whether AI spreads from a handful of large tech companies into ordinary businesses, and on whether power, digital infrastructure, skills, and financing can keep pace. ^[8]

So calling every dollar of AI investment a bubble is just as lazy an answer. Orders, revenue, and infrastructure are genuinely happening; Taiwan is genuinely earning income, jobs, investment, and tax revenue from it. The point of risk management is not to deny the upward trend, but to avoid mistaking that trend for a straight line with no volatility.

In its July outlook, the IMF also laid out the other side: if markets mark down expectations for AI profits and productivity, technology investment could contract quickly; economies with higher tech exposure and greater market concentration could also see a stock-price correction drag down consumption through the wealth effect. ^[2][7]

The point here is to separate "the technology is useful" from "every single investment in it will pay off."

The internet genuinely changed the world at the end of the twentieth century, but telecom and internet investment at the time still saw overbuilding and a valuation correction. AI could just as easily satisfy both conditions at once: raising productivity over the long run, while in the short run leading some companies to chase uncertain returns at excessive cost. As soon as expectations shift from "more is always better" to "prove the revenue first," the growth rate of capital spending will fall.

Taiwan is taking on the hardware orders, so it will see purchasing changes earlier than the end applications will — but it cannot control when those end applications actually generate cash flow. If large customers demand price cuts, delay equipment delivery, or adjust their product roadmaps, the pressure will travel down through chips, packaging, servers, and components.

What the pessimists are really warning about is not a specific crash date, but the fact that Taiwan has converted global tech companies' investment decisions into its own exports, equipment investment, tax revenue, and stock-market wealth. The deeper the connection, the faster the transmission.

AI investment and the energy shock are not two unrelated lines.

Data centers need large, stable amounts of power. Rising energy prices raise the operating cost of server halls and can also make it harder for central banks to cut interest rates quickly; with rates staying higher, companies use a higher discount rate when calculating the returns on long-term investment. Even if chip demand stays unchanged, data-center siting, financing, and grid construction can still be delayed. ^[1][2][9]

Taiwan is also an energy-importing economy. Rising international oil and gas prices affect domestic companies through power generation, transport, chemicals, and input costs. The government can smooth prices temporarily, but subsidies and the finances of state-run enterprises are not unlimited; the cost is simply borne by a different account.

This creates an asymmetry: the gains from AI orders are concentrated in part of the tech chain, while an energy shock spreads through electricity, transport, and prices to a much broader range of companies and households. Looking only at total exports shows the updraft; looking only at non-tech companies' costs misses the tech boom. Both are present at once — that is the complete picture of Taiwan's business cycle in 2026.

Judging whether high growth will continue requires separating at least five sets of signals.

The first set is customer capital spending. Watch the actual capital expenditure, free cash flow, and next-quarter guidance of major cloud and AI companies — not just their sweeping multi-year investment pledges.

The second set is the quality of Taiwan's orders. Beyond export value, watch order visibility, the rate of cancellations or delays, days of receivables, inventory, and capacity utilization. If revenue growth comes with slower collections, the risk is already moving.

The third set is price and product mix. A rising share of high-price products can push exports up, but if average selling prices turn weaker first, nominal exports may slow even while unit shipments keep growing.

The fourth set is energy and interest rates. Oil and gas prices, long-term bond yields, and data-center power bottlenecks will together affect the threshold for capital spending. ^[1][2]

The fifth set is the degree of spillover. If non-tech investment, private consumption, median wages, and services demand keep pace, that means AI income is gradually turning into broader domestic demand; if there is no spillover over the long run, headline growth is that much more dependent on a single external cycle.

No single indicator can declare the cycle over. Only when all five sets of signals turn weak at the same time is that clear evidence for cutting risk.

At the national level, the most important task is converting this temporary excess growth into long-term capacity. Fiscal planning should not treat all the tax revenue that comes with high growth as a permanent base; a sounder approach is to put part of it into the power grid, talent, the social safety net, and productivity in non-tech sectors, and to keep a buffer for when the cycle turns.

The government should also break its business-cycle dashboard down into finer pieces. Presenting AI-related exports, non-electronics exports, private consumption, fixed investment, median wages, and industry employment alongside GDP would let society understand how "growth is very high" and "it doesn't feel that way" can both be true at once.

At the level of industry intermediaries, trade associations and banks should group companies by customer concentration, cash-conversion cycle, and degree of capacity expansion. High-growth suppliers need stress tests, not just more loans; non-tech companies need market development and productivity improvement, not to be masked by the overall boom.

At the SME level, do not use 9.64% to estimate your own revenue for next year. The most reliable data is still your own order book, gross margin, collections, and inventory. Before expanding capacity, test at least three things: can you survive if your largest customer cuts its order by 20%? Would you still have positive cash profit if the average selling price fell 10%? If new equipment takes half a year longer than planned to reach full utilization, is there enough capital to hold on?

The greatest value of high growth is not simply earning more for one year — it is buying time to fix weaknesses.

Upside scenario: AI applications spread faster, corporate revenue and productivity keep pace with hardware investment, and the energy shock eases quickly. Taiwan's exports and investment stay at high levels, and growth spreads into services and non-tech sectors. In this scenario, the priority is avoiding infrastructure bottlenecks and talent crowding-out so that supply capacity can keep up.

Baseline scenario: AI demand continues, but the growth rate of capital spending gradually normalizes; the year-on-year rate in the second half is lower than in the first half, exports remain high, and how different industries feel diverges further. In this scenario, companies do not need to slam the brakes, but should shift capacity expansion from "chasing orders" to "watching utilization and cash flow."

Downside scenario: The war or the energy shock drags on, interest rates fail to come down, AI profit expectations are marked down, and tech stocks and investment are re-rated together. Taiwan could face slower exports, a weaker wealth effect, and elevated non-tech costs all at the same time. In this scenario, fiscal buffers, the quality of bank lending, and corporate cash positions will matter more than the annual growth target.

None of these three scenarios is a prediction. Their purpose is to let policymakers and companies know in advance: when a given signal appears, which specific action to take.

Global 3.0% and Taiwan's 9.64% are not contradictory.

The former describes the world average after war, energy, technology, and fiscal pressure have offset one another; the latter describes the acceleration Taiwan gets from standing at the core of AI hardware investment. Taiwan's advantage is real, and the sensitivity that comes with high concentration is equally real.

There are two especially dangerous readings. One calls the high growth fake, and in doing so fails to see Taiwan's real strength in the global tech supply chain. The other treats the high growth as the new normal, and in doing so forgets that capital spending, prices, energy, and valuations all move in cycles.

The more mature approach is to acknowledge, while riding the wave, that you are rising — and to measure the waterline for when the tide goes out at the same time. The question Taiwan should be asking is not "can 9.64% last forever," but rather: how much capacity has this stretch of high growth left behind for the state, for industry, and for companies to draw on the next time the shock comes.