Is AI Running Out Of Money?

Megacaps funding AI expansion have issued nearly $250B in debt this year, but investors are increasingly cautious as demand/payback lags. Amazon plans $200B in AI and robotics spending this year versus $140B cash reported at Q1 end, while Oracle’s $70B AI investment has coincided with a 28% share decline this year. Skepticism is rising that AI data centers won’t quickly offset costs (risk to debt premiums) alongside growing local opposition—reported $130B of data center construction blocked so far this year.

Analysis

The market is likely underpricing how quickly AI capex can shift from an equity-duration story to a credit story. The first-order loser is the hyperscaler with the weakest margin for error on incremental AI spend relative to revenue visibility; the second-order loser is the supply chain around it, because any spread widening or rating pressure forces procurement discipline before the core businesses feel it. AMZN can absorb more of this than smaller peers because of its cash engine, but the opportunity cost is real: more AI spend means slower buybacks and lower FCF conversion, which matters if multiple support is already stretched.

The cleaner implication is not "AI demand dies," but that project timelines become more lumpy. If local permitting, power, and water pushback slows data-center builds, revenue recognition for semiconductor, networking, and power-infrastructure vendors gets pushed right while bond markets reprice the funding model left. That creates a setup where current winners in compute hardware can still outperform on near-term orders, but the beneficiaries of grid buildout and power availability may do better over 6-18 months than the data-center owners themselves.

The contrarian read is that this may be less about absolute debt levels and more about financing quality: if these borrowers can still issue long-dated paper at modest spreads, the equity selloff in the capex-heavy names can overshoot. The real falsifier is not the headline spending number; it is whether AI monetization shows up in operating margin and net revenue retention over the next two quarters. If it does, debt becomes a levered growth tool; if it doesn’t, the market will start discounting a slower, more selective buildout across the whole AI stack.

Key Decisions for Investors

• Short ORCL on rallies or buy 3-6 month put spreads: it has the weakest revenue-to-capex symmetry in this group, so any widening in credit spreads or guidance disappointment should hit the multiple first; target 15-20% downside, stop if the stock reclaims its post-selloff range on improving backlog commentary.
• Prefer a relative long in power/grid beneficiaries over data-center operators for a 6-12 month horizon (e.g., VST or CEG vs DLR): if AI buildouts proceed, the bottleneck shifts to electrons, not chips; if permitting delays persist, the pair still benefits from deferred but not canceled demand.