Tech's AI buildout is being funded like an infrastructure supercycle
Google, Meta, Microsoft, and Amazon have poured $112 billion into AI infrastructure in the last three months. That level of capex needs fresh capital. So they're leaning hard on debt, asset-backed structures, and off-balance-sheet financing to keep scaling.
For finance teams, this isn't a curiosity. It's a new credit complex forming around data centers, power contracts, and chip supply-at a speed that tests balance sheets and market plumbing.
How they're funding it
We're seeing bonds secured by data centers, special purpose vehicles (SPVs), and private credit stepping in where traditional lenders would typically hesitate. Blackstone is raising $3.46 billion via a data center bond structure, and Meta used an SPV to finance $30 billion for a new site.
Morgan Stanley estimates private lenders will need to supply $800 billion over the next two years to satisfy demand. This isn't a one-off. It's a pipeline.
Why this matters for credit markets
Only about 3% of consumers are willing to pay for AI services today. That means the revenue engine must come from enterprise spend, developer tooling, and embedded AI inside cloud contracts. If profits don't cover capital costs, debt loads can spill risk into broader credit markets.
The Bank of England flagged this explicitly: if hyperscalers can't earn their cost of capital, contagion risk rises. Timing and unit economics matter more than the narrative.
Bank of England Financial Stability Report
The circular cash loop
Hyperscalers invest in AI firms like OpenAI. Those firms then funnel spend back into cloud, GPUs, and power bought from the same hyperscalers. It's a loop that boosts bookings and, in turn, equity prices-until it doesn't.
OpenAI's contracts for roughly $1 trillion in compute have already locked in more than 20 GW of capacity. That demand signal propels new builds and adds momentum to stock valuations.
Deal flow to watch
Banks including Sumitomo Mitsui, BNP Paribas, and Goldman Sachs are backing an $18 billion loan for OpenAI's "Stargate" data center in New Mexico. Beyond that, there's a $38 billion push for sites in Texas and Wisconsin.
The scale, tenor, and collateral packages on these deals are setting templates for the next wave of financings across the sector.
Practical takeaways for finance teams
- Credit investors: Track AI capex intensity vs. cash generation, coverage ratios including off-balance-sheet obligations, and sensitivity of EBITDA to GPU and power pricing. Scrub disclosure for SPV use, recourse, and cross-default linkages.
- Lenders and private credit: Tighten covenants on incremental secured debt, SPV leakage, and permitted liens. Negotiate step-in rights on data center collateral, minimum power availability, and take-or-pay compute contracts.
- Corporate treasury: Hedge rate and power risk tied to long-dated AI buildouts. Align capacity reservations with signed demand. Use prepayments and offtake agreements to reduce funding costs.
- IR and strategy: Separate "AI narrative" from unit economics. Report paid conversion, AI revenue mix, cloud gross margins with AI workloads, and utilization of reserved compute.
- Leading indicators: Data center occupancy, PPA strike prices, GPU lead times, cancellation rates on capacity, and disclosure of AI-related prepayments in cloud contracts.
- Stress test: +100-200 bps rates, 20-30% slower AI revenue ramp, delayed chip supply, grid constraints, and tighter high-yield windows. Map second-order effects across suppliers and customers.
What could break
- Funding gaps if private credit or bank balance sheets hit limits at the same time the bond window narrows.
- SPVs drifting into de facto recourse, pulling liabilities back onto parents under stress.
- Power bottlenecks delaying revenue realization on committed capex.
- Policy shifts on energy subsidies, export controls, or AI safety rules increasing cost of capital.
What could help
- Faster enterprise adoption with usage-based pricing tied to clear ROI.
- Government support for energy and grid upgrades that lowers build costs.
- Vendor financing, strategic equity, and longer-term customer prepayments to smooth cash needs.
Action steps this quarter
- Inventory exposure to data center ABS, AI-linked SPVs, and hyperscaler credit across all portfolios.
- Re-underwrite models with explicit power and GPU price paths; add downside cases for slower uptake.
- Ask for clearer disclosure: SPV recourse, compute utilization, AI gross margin bridges, and contract durations.
- Set position limits on unsecured tech debt where AI capex is a material driver without matching cash flows.
- Track regulatory and grid updates in key build states (NM, TX, WI) as timing catalysts.
Valuation reality check
US tech stock valuations have surged on AI infrastructure momentum. That's fine as long as utilization and pricing catch up with spend. If not, equity beta will transmit into credit spreads-fast.
Sam Altman has defended plans for $1.4 trillion in AI investments and asked for government support. That request alone tells you how capital hungry this cycle is.
Sources
Tooling for finance teams
If your team is pressure-testing AI budgets and use cases, a curated list of AI tools for finance can help ground assumptions in real workflows and costs.
Your membership also unlocks:
