Oxford Economics Puts an Asterisk on Asia’s Factory Boom

Oxford Economics has Asia-Pacific industrial production growing 4.9% in 2026, a June upgrade it has already wrapped in 2027 caution. Chip plants in Taiwan and South Korea still carry the print. Gulf fuel costs, weaker car demand, and a slower AI spending impulse sit on the other side of the ledger.

Lead economist Toby Whittington and electronics economist Makoto Tsuchiya walked that split in a 17 June 2026 webinar. By 9-10 September the firm was marking down 2027 Asia GDP and treating thin Hormuz traffic as a multi-year base case, not a passing shock.

4.9% Growth, Then a Smaller 2027 Impulse

On 2 June the Industries team raised industrial production growth to 4.9% for 2026, up from 4.3% in April, after a stronger-than-expected first quarter. That is a 0.6 percentage point lift. The same note said supply disruptions from the US and Israeli war with Iran would slow growth for the rest of 2026.

The June upgrade, which we covered when Oxford Economics lifted the 2026 APAC industrial growth forecast, treated high-tech electronics as the offset. US hyperscaler capital spending was still feeding semiconductor lines in Taiwan and South Korea, and the wider parts chain around them.

Three months later the firm is not walking back the 2026 factory number in public teasers. It is changing what that number is allowed to mean. A 10 September briefing on Asia’s 12 major economies puts 2027 GDP at 4.2%, 0.4 percentage points slower than 2026’s 4.6%. Hardware demand stays high. It just adds less extra growth once the 2026 surge is in the base.

THE TWO VINTAGES

Those are different series. Factory output can still beat GDP when chip plants run hot and households do not. That is the 2026 pattern the September notes are already discounting for 2027.

Chip Exports Are Carrying Asia’s Trade

Betty Wang, lead economist on the firm’s Asia Chip Export Index, and assistant economist Jia Yu Lee built the case in an 18 May briefing. The index, which bundles integrated-circuit shipments from Taiwan, South Korea, mainland China, Japan, and Singapore, jumped in March by 80.7% year on year in value and 28.3% in volume. Non-tech exports rose 4.6% in the same month.

The value-volume gap is the widest in 15 years. In older phone and PC cycles, a price spike often marked the peak. Wang’s team said this cycle looks different because compute demand, not household device replacement, is setting the order book. South Korea’s semiconductor export prices were up 148.3% year on year in April, the seventh straight month of large gains, helped by high-bandwidth memory.

THE MAY CHIP SCORECARD

Those five chip exporters account for about 70% of Asian exports in the index. Semiconductor shipments from them have doubled their share against a decade earlier. Across 11 major Asian economies, leaving out Indonesia, electronics are doing the work that cars, chemicals, and consumer goods are not.

Wang’s transmission chain runs from token use to hyperscaler capex to accelerator sales to Asian foundry revenue. OpenRouter was handling about 80 trillion tokens a month, against 20 trillion a year earlier. Purchases of property and equipment at Microsoft, Amazon, Alphabet, and Meta rose 70% year on year in the first quarter, more than 10 points faster than in the fourth quarter of 2025. Combined data-centre revenue at Nvidia and AMD more than tripled between the first quarter of 2024 and the first quarter of 2026.

At TSMC, high-performance computing was 61% of wafer revenue in the first quarter, up 10 points from 2024, while smartphones fell 9 points to 26%. Taiwan’s top five chip fabrication firms grew revenue 38.6% year on year in the first quarter, after 29.3% in 2025. SK Hynix nearly tripled first-quarter revenue and said high-bandwidth memory would stay tight for three years. The briefing’s working assumption was AI-related demand adding 10 to 15 points to Asian export growth through 2026.

Gulf Fuel Still Sets Factory Costs

The June industry note was blunt about who does not get that tailwind. Asian manufacturing still leans on Gulf hydrocarbons, which leaves it more exposed to a Hormuz shock than plants in Europe or the United States. Higher energy prices also cut household buying power, with the hit landing first on durable goods such as cars.

Whittington’s beat is the unfashionable side of that map: automotives, chemicals, and high-tech goods, after years running the firm’s oil, gas, and renewables forecasts from Singapore. Tsuchiya covers electronics, including semiconductors. Their August follow-up webinar kept APAC industry as the global outperformer for 2026, then added that momentum in traditional manufacturing was turning more uneven as energy costs and tariffs stacked up.

China is the awkward middle. The June briefing said advanced manufacturing should keep getting state support, while higher energy costs and softer foreign demand threaten export-heavy lines. The 10 September Asia note extended China’s housing correction through 2030, arguing that fading stimulus and a weaker property market outweigh the regional lift from Beijing’s manufacturing push.

The Asian Development Bank, in its July 2026 outlook, also tied a weaker developing-Asia call to prolonged disruption to energy and supply chains after the Middle East conflict. Factory PMIs in several economies had already dipped in March and April, at the peak of the first energy wave, before a May bounce.

Why the AI Boom Stops at the Fab Gate

The chip cycle is not a factory-town cycle. Wang’s May paper already listed energy security and critical industrial gases as constraints that were not fully in the baseline. Early pressure showed up in higher energy import bills for South Korea and Taiwan. Korea and Japan leaned back on coal. Taiwan looked toward nuclear. Singapore shopped for other LNG. Malaysia had already started restricting non-AI data centres because the grid and water could not take more load.

But supply-chain sustainability could become a binding constraint, particularly around energy security and the availability of critical industrial gases.

Betty Wang and Jia Yu Lee, Oxford Economics, 18 May 2026 briefing

That is the catch inside the bright spot. The plants that print AI accelerators need stable power, water, and specialty gases. They do not automatically hire more auto workers in Thailand or chemical operators in coastal China. Marco Santaniello and Paul Moroz, writing on 9 September, said AI’s global mark is still clearer in capital spending than in productivity, and that demand for AI hardware has helped supplier economies in Asia while US firms booked the equipment and software bills.

Their baseline has generative AI lifting US productivity about 3.5% over the next decade and 4.5% in the long run, on the cautious side of published estimates that run from about 1% to more than 12%. Adoption in that model only reaches 80% of possible uses by 2040 in most advanced economies. Until workflows change, Asia’s gain is the wafer, the memory stack, and the server box, not a broad productivity wave on the factory floor.

Power is the other gate. New AI campuses are already bumping into grids, and gas-fired AI campuses face a short useful life if the electrons are not there. Nvidia’s Vera Rubin platform is billed at up to 35 times more throughput per megawatt, which helps, and over time better token-per-watt math could slow the pace of chip orders even if total compute keeps rising. Near term, the order book stays full. The employment multiplier does not.

Korea’s Semiconductor Export Machine

South Korea is where the concentration stops being a chart and starts being the whole export account. Jefferies, in a note circulating on 11 September, said semiconductor shipments jumped 193% year on year to $133 billion in the three months to August, a record 44% of total exports. Computers, electronics, and optical products contributed 1.9 percentage points of Korea’s 3.7% real GDP rise in the second quarter, which is 51% of the growth print.

Early September was more of the same. Korean export data for 1-10 September put chip shipments at $16.48 billion, up 270.1% year on year, and at 47.1% of all exports, 23.9 points higher than a year earlier. Cumulative exports had reached $709.4 billion by 5 September, just above the previous year’s full-year $709.3 billion.

HOW TIGHT THE KOREA BET IS

• Three-month chip haul: $133 billion, up 193% year on year, 44% of all exports through August, per Jefferies.
• Growth share: Electronics accounted for 51% of second-quarter GDP growth.
• Early September pulse: Chips were 47.1% of shipments in the first ten days, at $16.48 billion.

A country can run that hot while the rest of industry is flat. It cannot run that hot and call the result a balanced manufacturing recovery. If hyperscaler capex slows, as the 9 September global outlook already flags as the next phase of the AI boom, Korea feels it first and hardest. SK Hynix’s three-year HBM shortage call is the bull case. A disorderly end to the boom is the risk Oxford Economics has now put on the 2027 page.

Hormuz Traffic Stays Thin Into 2027

The energy side of the June webinar did not fade. Brent crude rose above $100 a barrel on 9 September 2026, the first time in six weeks, and traded around $105 the next day. An OPEC monthly report showed Saudi crude output at 6.238 million barrels a day, down 1.9 million barrels a day, the lowest since 1990.

Those prices sit on a physical gap, not a scare headline. Oil-shipment tracker Vortexa has about 10 million barrels a day still missing because of the Iran war, roughly a tenth of world demand. Kpler put August oil exports through the Strait of Hormuz at 2.2 million barrels a day, against about 17 million barrels a day in 2025. Those are different cuts of the same problem: barrels not arriving, and the strait itself still running far below last year’s pace.

HOW THE OUTLOOK TIGHTENED

1. 18 May 2026: The chip-export briefing treats AI compute as a durable tailwind and lists energy security and industrial gases as unmodelled risks.
2. 2 June 2026: The Industries team lifts 2026 APAC factory growth to 4.9% and warns the Iran war will slow the rest of the year.
3. 17 June 2026: Whittington and Tsuchiya take the split on air at 10:00 HKT.
4. August 2026: The follow-up session keeps APAC as the outperformer and says traditional manufacturing is turning uneven under energy costs and tariffs.
5. 9-10 September 2026: Global and Asia briefings mark down 2027 growth, hold thin Hormuz traffic through the end of 2027, and warn that hyperscaler investment growth is slowing.

The 9 September world outlook now treats simmering US-Iran tension and lower-than-normal Hormuz traffic through at least the end of 2027 as the most likely path. It also says that path need not send oil prices spiralling from here. The factory consequence is duller and more durable: a lasting fuel tax on chemicals, shipping, and car plants, while chip lines keep the regional average looking healthy.

The Consumption Multiplier That Did Not Show Up

The September Asia briefing’s real warning is not a chip crash. It is that the 2026 boom does not pass through to pay packets. Strategic spending on technology, power, defence, and supply-chain capacity is running at about 18% of regional GDP and 56% of total investment, a high plateau the firm thinks will hold in 2027. Import content and physical bottlenecks cap the near-term lift to jobs and incomes.

THE THREE 2027 ASTERISKS

• Thin domestic value: Intra-Asia trade looks dense, but parts cross several borders, so local income gains lag the shipment totals, with China still sitting at both ends of the chain.
• Import leakage: The capex wave buys a lot of foreign equipment, land, and power gear, so the multiplier inside each host economy stays small.
• Fading support: Consumption ran near 3.7% in 2026 on fuel subsidies and cash transfers; as those fade, and with China’s private consumption seen staying under 4%, households do not take the baton.

On 9 September the global desk put 2027 world GDP back in a 2.8% to 3.0% range and said risks that oil stress or a disorderly end to the AI boom could finish the cycle are growing. Hyperscaler liabilities and Chinese model competition point to slower investment growth and jumpy equity prices, not an automatic bust. For APAC factories, that is already enough. The 4.9% industrial print was real. The engine that produced it is a narrow set of wafer and memory lines whose extra impulse is fading in the 2027 arithmetic, while the energy bill stays in the model through the end of next year.