HSBC's 37% AI Risk: Why Overspending Fear Is Rewiring the Market Now

HSBC's overspend regime is shifting the market from AI demand to AI payback

The market is no longer debating whether AI demand is real. It is asking who pays for it, and whether today's winners are already financing tomorrow's problem. HSBC's model puts a 37% probability on hyperscaler overspend as the dominant AI regime. That matters because narrative share can change market flows faster than earnings do.

Investors are now testing who gets paid, not just who is exposed

In plain English, this is a payback trade. For years, the market rewarded almost every AI link in the chain. Now investors are asking whether current prices already assume years of heavy capex and strong returns on capital lofty valuations. That shifts the focus from "who has demand" to "who sits on the winning side of spending discipline." In HSBC's framework, the overspend regime still favors data-center and semiconductor names, but it does so at the expense of the capex spenders.

The selloff showed how quickly fear moves

Loss aversion and recency bias appear to be taking over from pure AI excitement. After global semiconductor stocks sold off sharply, the weakness spread quickly through sensitive markets. EM Asian equities fell more than 2% on Wednesday, and South Korea's KOSPI dropped 5.35% that day.

If the debate keeps moving from adoption to accountability, capital is likely to keep favoring proven AI suppliers while putting more pressure on the companies expected to carry the spending burden.

The AI winners are narrowing, not disappearing

The debate is no longer about whether the AI buildout continues. It is about who gets paid while the bill gets bigger.

Under HSBC's leading regime, the AI trade gets more selective

Under HSBC's leading hyperscaler overspend regime, the AI trade does not die; it narrows. In that setup, tech hardware returns an annualized 12.3% and semiconductors 11.8%, but those gains come at the expense of the capex spenders. The market is still pro-AI, but more hostile to what it sees as inefficient spending.

That distinction matters more than the headline fear. A market worried about overspending can still send specific subsectors higher; it just becomes stricter about who sits on the winning side of the chain. In practice, that favors scarce bottlenecks and proven suppliers over companies expected to carry the heavier spending burden. That is why exposure to Taiwan and South Korea can still work in this regime, but mainly for the firms viewed as powering the AI infrastructure boom, not for the broader market on faith alone.

Why the market keeps wobbling

The latest volatility showed how quickly the market can turn when a broad AI rally starts facing tougher questions. Major semiconductor indices fell more than 10% in the week, erasing about $1.3 trillion in market value. Forbes described the move as being driven by fears around AI infrastructure returns, high valuations, and a hawkish Federal Reserve, rather than a loss of demand.

That helps explain the wobble. Investors do not appear to be rejecting AI itself. They are rejecting overpayment for it. In that context, sharper rotation and faster punish-reward cycles are exactly what you would expect when narrative and valuation matter more than simple exposure.

What to watch as rotation replaces broad AI optimism

What looks priced in is not an AI demand collapse. It is a harsher test of who gets paid during the buildout. The earlier EM Asia and Korea stress showed how quickly fear spreads once investors start looking through a financing lens instead of a growth lens renewed concerns over weaker AI spending. The latest chip selloff fits the same pattern: it was driven by high valuations, reassessing the price they are willing to pay, and concerns about AI infrastructure returns, not by clear evidence of collapsed demand.

The practical setup is relative, not absolute

The practical trade is not simply "long AI." It is the spread between AI infrastructure suppliers and the firms carrying the heaviest spending burden. HSBC's leading regime still favors data-center and semiconductor names at the expense of the capex spenders. In plain English, the market is rewarding scarce bottlenecks while putting more pressure on the companies likely to foot the bill.

The next test is management commentary on capex discipline

The next test is management commentary around AI capex discipline and expected payback. That is where the narrative can harden into price. If companies can defend spending and returns, the market may keep rewarding scarcity. If not, the rotation into more selective AI winners could deepen.