Good news from the World Bank usually comes without a health warning attached. This week's did not. The bank lifted its 2026 growth forecast for East Asia and the Pacific to 4.5 percent, with Vietnam jumping a full point to 7.4 percent, and credited the upgrade almost entirely to one thing: the global scramble for the chips, servers and electronics that feed artificial intelligence. Then, in the same report, it explained why that is precisely the problem.
The dependence is startling once you see the numbers. AI-related goods accounted for more than half of all export growth across most of the region, and more than 70 percent in Malaysia, the Philippines, Thailand and Vietnam. Six economies shipped $1.4 trillion of them in the year to April. Strip out AI, the bank notes, and trade growth has been "weak or negative." South Korea's exports leapt 83.5 percent in September to a record, with chips half the total; two firms, Samsung and SK Hynix, make up 43 percent of the country's benchmark stock index. An entire region's prosperity is leaning on a single, narrow bet.
What makes the bank nervous is the shape of the spending behind it. AI-related capital expenditure has reached roughly 6 percent of US GDP, matching the dot-com peak of 2000, and it is still accelerating. Of the $2.9 trillion in AI investment planned through 2028, some $800 billion is expected to come from private credit, a corner of finance the bank calls "less visible" and untested by a severe downturn. AI lending there has already jumped to 34 percent of activity. The Bank for International Settlements drew the same comparison in June, likening the boom to the "manias" of history. HSBC now sees early signs of cooling: softening electronics prices, weaker new orders, a pullback in Asian shipments.
If the Bank's tone is cautionary, former BitMEX chief Arthur Hayes is simply blunt. Speaking in Singapore, he argued that "humanity is wasting multi-trillion dollars" building AI data centres, and that the result will follow the script of every great technology rollout: overbuilding, a crash, then a bailout. "If you study financial history," he said, "it always is overbuilt. There always is a crash, and there always is a bailout." His reasoning is almost mechanical. End users like OpenAI, Anthropic and SpaceX do not yet make money, he notes, yet once the data centres under construction are finished, their builders will come looking to be paid for all that committed compute. He puts the reckoning at late 2027 or 2028.
It would be easy to file Hayes under permabear, and he would not entirely object; he happily concedes the bull case, that AI could become useful enough in the next year to make its backers profitable, and he says he would not short the sector. But the interesting thing is how closely his crash timeline tracks the World Bank's own warning about when planned capacity lands and demand must finally justify it. The pessimist and the multilateral institution are looking at the same calendar.
The deeper point is not whether a bust is coming but who absorbs it. A correction in Silicon Valley is a story about valuations. A correction that runs through Vietnamese factories, Korean chip lines and Malaysian bank balance sheets is a story about livelihoods. The World Bank's estimate is sobering: a one-point slowdown in US growth knocks 0.6 of a point off other emerging markets, with the hit to investment roughly twice as large. The AI boom has quietly become load-bearing for economies that never chose to gamble on it. That is the risk no benchmark captures.