Wall Street just started doubting AI spending. Data center construction is riding on the answer.
IBM's stock crashed 25% and tech had its worst week since April 2025 after investors began questioning whether hyperscaler AI capex pays off. Data centers are now more than a fifth of all nonresidential building starts — funded increasingly by debt, not cash.
IBM's stock dropped 25% on July 14 — its worst single-day fall on record — after warning that corporate clients were shifting IT budgets toward AI infrastructure. That single earnings warning helped trigger the worst week for tech stocks since April 2025, with the Nasdaq 100 down more than 4% and chip stocks off roughly 10%, as investors openly questioned whether the money hyperscalers are pouring into AI actually pays off. For most industries that's a market story. For construction, it's a question sitting directly underneath the biggest growth driver the industry has: data centers, now more than a fifth of all nonresidential building starts nationally, increasingly built on borrowed money rather than cash.
What actually happened on Wall Street?
IBM's warning wasn't really about data centers — it was about enterprise clients diverting software and consulting budgets toward AI infrastructure spend. But it landed in the middle of a broader, building investor question: Alphabet, Microsoft, Amazon and Meta are projected to spend roughly $725–785 billion on capital expenditures in 2026, with Moody's pushing 2027 estimates toward $1 trillion across the six largest hyperscalers. Investors have tolerated that pace as long as AI revenue kept climbing to match it. The week of July 13–17 was the first time in months that tolerance visibly cracked, with tech leading a broad market selloff on doubts about payback timelines.
Why does a stock selloff matter to a contractor pouring a data center slab?
Because the way these projects get funded has changed. Hyperscaler capex used to run largely off internal cash flow. It increasingly doesn't. Morgan Stanley forecasts global AI-related debt issuance will hit roughly $570 billion in 2026 — about four times last year's pace — as companies borrow against future AI revenue to keep building at this speed. That's a different risk profile than a project self-funded by a company sitting on tens of billions in cash. A debt-financed project depends on lenders and bondholders staying confident. When that confidence wobbles, as it did last week, the projects most exposed aren't the ones already poured and steel-topped-out — they're the ones still lining up financing for their next phase.
What should a GC or sub check before taking on more hyperscale backlog?
| Question to ask on a data center pursuit | Why it matters |
|---|---|
| Is this project funded off the hyperscaler's own balance sheet, or through a debt-financed special-purpose entity? | Balance-sheet-funded projects are far less exposed to a bond-market wobble than SPV-financed ones |
| What share of your current backlog is data center work? | Concentration risk compounds if financing conditions tighten across the sector at once |
| Are contract milestones tied to project financing closes, or independent of them? | Financing-contingent milestones can slip even when your crew and schedule are ready |
| Does the contract include payment or parent guarantees beyond the project entity? | A guarantee from the hyperscaler itself is worth more than one from a single-project LLC |
Should a mid-size GC or sub walk away from data center work?
No — the underlying demand is real, and it's still growing fast: data center construction spending was up 23% year over year as of May 2026. The point isn't to avoid the sector, it's to underwrite the financing the same way you'd underwrite an owner's creditworthiness on any large speculative job. CAB covered how a rejected air-quality permit can freeze a $165 billion data center project overnight — that's a supply-side risk you can track through public filings. This one is harder to see coming, because it shows up in bond markets, not zoning boards.
The takeaway
Ask who's actually funding the project before you commit more backlog to it — a hyperscaler's balance sheet and a debt-financed SPV are not the same counterparty, even when the sign on the fence says the same company name.
Forward this to the person on your team who's still arguing AI is overhyped. Subscribe at constructionaibrief.com.
Next time a data center pursuit lands on your desk, ask one more question before you bid it: whose balance sheet is actually paying for this.
- Why did IBM's stock crash affect confidence in AI data center spending?
- IBM fell 25% on July 14 — its worst single-day drop on record — after warning that clients were redirecting IT budgets toward AI infrastructure (servers, storage, memory) and away from IBM's software and consulting business. It wasn't a data center story on its own, but it crystallized the question investors had already been circling: is the money pouring into AI infrastructure actually generating returns, or just moving around the balance sheet?
- How much of nonresidential construction is data centers right now?
- Data center starts have accounted for more than one-fifth of all nonresidential building starts nationally over the trailing 12 months, according to ConstructConnect's July 2026 report. Census Bureau data put data center construction spending at a $59.3 billion seasonally adjusted annual rate in May 2026, up 23% year over year — roughly 8% of all private nonresidential construction spending.
- Is hyperscaler AI capex funded by cash or debt?
- Increasingly debt. Morgan Stanley forecasts global AI-related debt issuance will reach roughly $570 billion in 2026, about four times the pace of borrowing over the same period in 2025. Moody's has raised its 2026 capex projection for the six largest hyperscalers to about $785 billion, heading toward nearly $1 trillion in 2027 — a spend rate that increasingly outpaces what those companies generate in free operating cash.
- Does a stock selloff actually threaten a data center that's already under construction?
- Not directly, and not on projects already fully funded and under contract. The exposure is on the next wave: projects still in financing, especially ones structured through debt-funded special-purpose vehicles rather than a hyperscaler's own balance sheet. If bond investors get pickier or pricier about AI-linked debt, the projects most likely to slip, re-scope, or stall are the ones whose funding was never fully committed in the first place.
- What's different about this risk compared to data center permitting or moratorium news?
- Permitting fights and local moratoriums are a supply-side risk — a government body blocking or delaying a project that otherwise has funding. This is a demand-side and capital-markets risk — whether the money behind the project keeps flowing at all. A GC can win every zoning fight and still see a project slow down because the owner's financing got more expensive or harder to close.