HSBC Says 37% Odds AI Overspend Fear Is Swapping Winners, Not Killing the Cycle

HSBC's top AI regime is about who pays, not whether AI is real

Investors are no longer debating whether AI demand exists. They are debating who gets paid when that demand starts to look economically fragile. HSBC's framework matters because it turns that vague anxiety into a market map. The model's top regime is hyperscaler overspend, assigned a 37% probability. In practice, the market is asking less "Is AI real?" and more "Who carries the bill, and who gets cut if returns disappoint?"

In that regime, infrastructure names can win while spenders lag

In the overspend scenario, HSBC says data-center and semiconductor names can outperform at the expense of the capex spenders. That is the key positioning split: investors still want AI exposure, but they are preferring the pickaxe sellers and infrastructure players over the companies most exposed to budget scrutiny.

That is why the timing feels urgent. EM Asia has already shown how quickly sentiment can turn against the spending side. The KOSPI closed 5.35% lower in one session and was down more than 20% from its late-June peak. Even Samsung forecasting a 19-fold jump in second-quarter operating profit did not stop the sell-off. The pressure has been heaviest where investors expect AI budgets to be challenged, not across the entire AI complex.

If hyperscaler spending stays firm, the crowd could rotate back. If capex doubts deepen, that split should sharpen further.

The same capex boom is feeding two different reads

The debate is not whether AI spending is still happening. It is about who gets repriced first if investors become more skeptical about returns on that spend. That is why the same capex boom is producing two opposing stories at once: bears see a valuation reset on the spending side, while bulls see supplier demand still holding up.

The bear case is a re-rating, not a demand collapse

The bearish case sounds rational because it is not really claiming AI demand has broken. HSBC says the sell-off reflects reassessing lofty earnings expectations. That is a more subtle move. When a trade has run on aggressive forward assumptions, investors do not need a demand collapse to sell; they only need evidence that earnings may not scale as quickly as valuations did.

The ownership pattern matters too. HSBC says the pullback has been shaped less by broad conviction selling and more by foreign investors reducing exposure while domestic retail investors buy. That argues against full capitulation. A true cycle break usually shows weaker participation across the board, not one group offloading shares while another absorbs them. Even in Korea, the sharp moves looked more like expectation-setting than final proof that the AI cycle is over.

The bull case is that the build-out still looks early

The stronger evidence still sits on the spending side. HSBC expects global AI capex to rise from less than USD 400 billion in 2025 to more than USD 1 trillion by 2028, with Asia becoming a central part of the expansion. That is not the profile of a boom that has already fully reversed.

Micron is the clearest near-term check. HSBC said the company's quarter showed record financial results and a strong outlook, even as Federal Reserve policy concerns and leveraged ETF positioning weighed on sentiment. The key point is that price action can weaken for macro and positioning reasons while underlying AI demand still holds.

What decides the debate from here

The practical takeaway is to stop trading AI as one basket. If hyperscaler budgets stay firm, today's fear swing may prove to be a painful but correctable re-rating of the payer complex. If budgets start to crack, the market's nerves may have been an early warning rather than just psychology.

Position for the narrative swing, not the AI slogan

The edge now is not predicting the next headline. It is positioning before the market decides which AI theme deserves the multiple.

HSBC's framework matters because narrative risk is about sequence. The model still sees hyperscaler overspend as the dominant narrative, but it also assigns meaningful odds to AI positioning capitulation and China competition concerns. Investors can be right about AI in general and still lose money if they own the wrong bundle when sentiment flips.

Favor broader AI enablement over a narrow winner set

The more durable approach is to own parts of the ecosystem that can benefit if the narrative broadens faster than expected. HSBC says AI earnings growth is broadening within and beyond the IT sector. That matters because, once AI stops being viewed as just a tech story, the market may start rewarding enableers such as power, copper, industrial logistics, data-center real estate, and the financial players supporting the build-out.

Fear can also reverse faster than expected. HSBC says continued strength in the AI trade could be one of the market surprises in H2 2026. If that happens, the first move may be a snapback in the beaten payer complex, followed by a wider ownership base as optimism regains control.

What to watch now

Signposts that the rotation is working: - earnings strength spreading beyond semis and hyperscalers, consistent with broader AI earnings growth - fewer headline-driven swings between AI clusters, suggesting the market is moving from fear to allocation - fundamental checks such as Micron's record financial results and a strong outlook holding up as sentiment improves

Invalidation - when this positioning idea breaks: - the market slips into AI positioning capitulation, with semis down and defensives outperforming - capex scrutiny starts hitting the whole chain, not just the spenders - fundamentals in key suppliers weaken enough to overturn the view that AI-related investment remains resilient

The window is now because narrative shifts do not wait for perfect clarity. They often happen in the gap between weak sentiment and improving proof.