HSBC CEO warns the AI boom won’t make every data center a good bet

HSBC CEO Georges Elhedery described AI as transformational for the bank but urged investors to be selective about data-center infrastructure, assessing tenancy agreements and contract pricing. Global data-center spending could reach $7 trillion by 2030, while JPMorgan estimates $4.1 trillion of AI-related debt issuance by the end of the decade. Allianz adviser Mohamed El-Erian said hyperscalers raised $200 billion in bond financing last year and could raise $1 trillion this year, adding to competition for financing as sovereign yields reach multi-decade highs.

Analysis

The more useful signal is about underwriting dispersion, not whether AI demand is real. If capital costs stay high while power, construction, or delivery constraints delay capacity, projects with short, weakly committed, or aggressively priced leases face refinancing and valuation pressure first. That can transmit losses through private credit and bank lending; it does not establish that HSBC has material impaired exposure. HSBC’s stated selectivity may limit downside, while its own data-center operations create a potential conflict between financing and operating risk that warrants exposure disclosure. Separately, bank productivity gains are a plausible medium-term margin opportunity, but adoption costs and realized savings—not management’s ambition—will determine value.

Near term (days to 1–3 months), rising bond supply and sovereign yields can pressure long-duration infrastructure credit even if tenant demand holds. Over 6–18 months, lease quality, power availability, project completion, and refinancing terms should separate durable assets from speculative builds. A broad short of AI infrastructure is premature: hyperscalers may sustain spending, and the article provides no borrower-level credit or HSBC exposure data. The contrarian risk is that investors treat all data-center leases as equivalent; the opposite error is assuming every project is fragile. Thesis weakens if financing spreads stabilize alongside firm lease commitments and on-time capacity delivery; it strengthens with spread widening, delayed projects, or deteriorating lease terms.

Key Decisions for Investors

• No immediate directional trade on HSBC, JPM, Allianz, or LSEG from this interview alone; the article offers no new company-specific earnings, exposure, or credit data.
• Favor relative credit selection over an AI-infrastructure basket: prefer borrowers with demonstrable long-duration tenant commitments and funded power/capacity plans; avoid adding to opaque, highly levered project credit until lease and refinancing details are verified.
• For the next 1–3 months, monitor hyperscaler bond supply, credit spreads, and sovereign yields. Treat simultaneous spread widening and higher yields as a trigger to reduce long-duration infrastructure-credit risk, not as proof that AI demand has rolled over.
• For HSBC, request disclosure on data-center loan concentrations, underwriting standards, and operational exposure before changing the position. Reassess if future reporting shows weakening collateral/lease quality or if realized AI productivity fails to translate into operating-cost improvement over the next 6–18 months.