Trends›AI capex stays financeable
AI capex stays financeable
Through 2030 the AI build-out stays financeable — hyperscaler cash flow, debt and private credit keep funding ~$500B-1T/yr of capex without a credit seizure or a circular-financing unwind.
55 ↑2 confidence, computed 23 Oct 2026
- Financial
- High weight
What rests on thismarkets it underwrites or threatens
Evidencethe schedule the figure draws
The evidence, newest first
- Oct 2026
for · claim Private source - Jul 2026
against · primary Private source - Jul 2026
neutral · observable Oracle's CDS skyrocketed and credit spreads are widening, though $109B on-balance-sheet bonds and $65B off-balance-sheet private credit have been absorbed without dramatic spread widening as of May 2026 An AI debt wave meets uneven balance-sheet risk - Jun 2026
for · observable Private source - Jun 2026
neutral · observable Aggregate hyperscaler cash capex (~70%/yr) overtakes operating cash flow (~23%/yr) around Q3 2026, when aggregate FCF reaches zero — so the marginal build shifts to debt/equity (which the source notes is the plan, not a seizure). Epoch AI — Hyperscaler Capex vs Operating Cash Flow - Jun 2026
for · primary CoreWeave's 5yr CDS fell to 4.52pp (Jun 2026) from 8.81pp (Dec 2025) and Applied Digital priced a $1.59B junk bond at 7% (from ~10%), 5x oversubscribed — AI-infra financing stress easing, not seizing. CoreWeave credit rebound spurs cheaper funding (Bloomberg) - Jun 2026
for · observable The five largest cloud/AI firms sold $159B of bonds in the first five months of 2026 (+47% YoY); the IG market absorbed it with Aa names paying only 10-15bp concessions, the group holding >$1T cash. AI hyperscalers issue $159B in debt, +47% YoY - May 2026
against · primary FSB warns AI made up more than a third of private-credit deals in 2025 (from 17% prior 5yr) and a sharp correction in AI-infra lending could cause 'sizeable' losses for private-credit investors. Financial Stability Board — Vulnerabilities in Private Credit - Mar 2026
neutral · observable CDS spreads rose especially for hyperscalers with lower credit ratings, reflecting both volume of supply and uncertainties around projects' payoffs Financing the AI infrastructure boom: on- and off-balance sheet borrowing - Mar 2026
neutral · observable Hyperscalers are on a ~$1T multi-year borrowing path (Alphabet debt-financed ~15% of its $180B 2026 capex, Oracle ~50%); UBS sees cumulative public hyperscaler debt at $230-240B by end-2026 — deep capacity but rising, market-dependent leverage. Google, Meta, Oracle on a $1 trillion borrowing spree (Fortune) - Feb 2026
for · observable Oracle's Feb 2026 ~$25B multi-tranche IG bond sale cleared the public market, part of Oracle becoming one of the largest US high-grade issuers (~$120B of notes) to fund data-center capex. Oracle Corp Form FWP (FY2026 bond offering), SEC EDGAR - Feb 2026
against · observable Analysts identify $800B+ of circular vendor financing (Nvidia investing up to $100B in OpenAI, which buys Nvidia GPUs via CoreWeave), with OpenAI on track to lose ~$14B in 2026 — demand partly the same dollars looping, vulnerable to an unwind. AI's circular economy: OpenAI-Nvidia-Oracle financing loop - Nov 2025
against · observable Burry argues hyperscalers depreciate GPUs over 5-6yr vs a real 2-3yr life, understating depreciation ~$176B over 2026-2028 and leaving Oracle/Meta operating income >20% above economic reality. Michael Burry accuses AI hyperscalers of boosting earnings (CNBC)
What would move itthe measures watched
- watchAggregate hyperscaler free cash flow (operating cash flow minus capex), Big 5. OCF growing ~23%/yr vs capex ~70%/yr; the gap closing to zero/negative is the core stress signal. Aggregate FCF hitting zero ~Q3 2026 means the marginal capex dollar must come from debt, not cash.
- watchHyperscaler + neocloud debt issuance and IG bond absorption. Rising issuance is fine IF spreads stay tight and books oversubscribed; the tell is new-issue concession (10-15bp mild; 50bp+ or pulled deals = market saying no).
- watchAI/data-center CDS spreads (Oracle, CoreWeave) and HY data-center bond yields. CDS widening = market pricing default risk. Spreads tightening (CoreWeave 8.81→4.52pp) = bull; a re-widening above Dec-2025 highs or HY deals failing to clear flips it.
- watchPrivate-credit share of AI/data-center lending and concentration. AI = >1/3 of private-credit deals in 2025 (up from 17%); ~$200B+ outstanding. Watch redemption caps at Apollo/Blackstone and any single-name PC default — the FSB-flagged systemic amplifier.
- watchDepreciation / GPU useful-life disclosures and any write-down. Hyperscalers depreciate GPUs over 5-6yr; bears argue 2-3yr real life (~$176B understated depreciation 2026-28). The first material write-down re-rates the group's earnings and financeability.