Lab · AI Credit Stress Tape

Who funds the AI build-out — and at what price.

The build-out has stopped being paid for out of cash flow. Capex now exceeds operating cash generation at several of its largest spenders, and the gap is bridged with debt — which makes the credit market the constraint that decides how long the build runs. This board sorts the issuers into six funding models, scores each rung from filings, and reads the credit environment in percentiles rather than levels. Because spreads are historically tight, a board that reported levels would print "no stress" every day and tell you nothing. Environment as of 2026-07-23; issuer figures are trailing four quarters of filings.

Stress sits at

T3 — Project / secured funded

Stress at the project-funded rung is normal cycle behaviour: high-beta breathes. Watch whether it climbs.

Operating rungs, trailing 4 quarters

131.7 $bn free cash flow

198.3 $bn net new debt

How to read it. The ladder is ordered by funding dependence, because that is the path stress travels. Pressure on the project-funded rung is ordinary — high-beta breathes. Pressure on the corporate-debt rung means the thesis is running. Pressure at the balance-sheet rung would be a regime change: the market questioning build-out financing itself. The single most useful signal is not a level at all — it is a name migrating between rungs.

The funding ladder

Balance-sheet funded

5/100

MSFT · GOOGL · META · AMZN

Structure: Capex is paid out of operating cash flow. Debt is opportunistic — maturity management, tax, timing — not necessity.
Where the risk sits: Not solvency. The risk is shareholder patience: these names are punished for capex in the equity, not the credit. The credit danger is behavioural — the cash cushion erodes as capex compounds and they start behaving like T2 while still rated AA.
The tell: Operating cash flow ÷ capex crossing below 1.0 — the moment capex stops being self-funded.

OCF÷capex 1.44 new debt÷capex 0.24 interest coverage 108.3×

Debt-funded corporate

48/100

ORCL

Structure: Capex exceeds cash generation persistently; the gap is bridged with senior unsecured, full-recourse corporate debt — on balance sheet, rated, publicly priced.
Where the risk sits: Rating migration. This rung sits just above high yield, and investment-grade to high-yield is a cliff, not a step: IG-only mandates become forced sellers, index exclusion follows, cost of capital jumps.
The tell: Net new debt against negative free cash flow, a falling interest-coverage trend — and a switch to preferred or convertible issuance, which is the debt market saying "not at this price".

OCF÷capex 0.57 new debt÷capex 0.72 interest coverage 4.5×

Private credit / SPV

not scored

ARCC · OBDC · BXSL · OWL · APO · ARES (proxies)

Structure: JV and SPV vehicles, data-centre ABS and CMBS, direct lending from the private-credit complex. The debt sits at the vehicle, not the corporate.
Where the risk sits: Invisibility. This layer appears in none of the corporate metrics above, so any ladder built only on filings systematically understates build-out leverage — while this is the most levered and least transparent rung, where stress can surface with no warning in corporate balance sheets.
The tell: Data-centre ABS spreads, deterioration across the private-credit complex, and JV/VIE language appearing in 8-K and 10-Q filings.

Measured by proxy only. Operating-company metrics do not describe this rung — the vehicles lend rather than build, and their leverage sits off the corporate balance sheet entirely. This board therefore understates total build-out leverage, and that is the honest state of it.

Project / secured funded

90/100

CRWV · NBIS · WULF · APLD

Structure: Debt secured against specific assets — GPUs, buildings, power contracts — and frequently against a single customer contract. The financing follows the contract, not the company.
Where the risk sits: Counterparty and collateral value, not leverage. The pledge is depreciating compute whose worth depends on the next silicon generation: a lender who underwrote against Hopper-era hardware holds different collateral in a Rubin world, with Feynman already on the calendar.
The tell: New debt ÷ capex above 1.0 — borrowing more than they invest, i.e. funding operations; and interest not serviced from the business at all.

OCF÷capex 0.19 new debt÷capex 1.29 interest coverage -2.5×

Vendor / circular financing

not scored

NVDA (proxies)

Structure: The supplier finances the customer — equity stakes in its own buyers, cash prepayments, equipment leasing, vendor credit.
Where the risk sits: Circularity, not leverage. The seller funds the buyer of its own product, so part of reported revenue growth is self-financed. If the customer breaks, the supplier is hit twice — revenue and the stake.
The tell: Growing investment and prepayment positions alongside rising customer concentration.

Measured by proxy only. Operating-company metrics do not describe this rung — the risk here is circularity rather than leverage, which cash-flow ratios cannot see.

Alongside the ladder

Landlords (own cohort)

own cohort

DLR · EQIX

Structure: REITs. They do not fund AI capex — they fund buildings and lease them long-term. The debt is mortgage-like, laddered, investment grade.
Where the risk sits: Rates, not AI. Their credit responds to the 10-year and cap rates rather than token demand. They carry the highest leverage on the board BY CONSTRUCTION, so tech thresholds manufacture false alarms. The genuine AI risk is tenant concentration and re-leasing: if a T3 tenant fails, the landlord holds a purpose-built shell.
The tell: Spread against a REIT cohort rather than tech — that isolates the data-centre premium from generic REIT rate risk.

OCF÷capex 1.31 interest coverage 3.1× Normal for a REIT — never compare to the rungs above.

Issuer by issuer

Trailing four quarters from SEC filings. Flags fire on the funding structure, not on price.

Issuer	Rung	OCF÷capex	FCF $bn	Net new debt $bn	New debt÷capex	Interest cover	What it says
MSFT	T1	1.75	72.9	6.0	0.06	52.7×	funds its investment from cash flow
GOOGL	T1	1.40	53.3	70.1	0.53	296.0×	external funding doing the work
META	T1	1.64	48.3	-2.6	-0.03	51.0×	funds its investment from cash flow
AMZN	T1	0.98	-2.5	63.0	0.42	33.7×	capex not covered by operating cash flow · drifting toward debt-funded (T2) behaviour
ORCL	T2	0.57	-23.7	40.1	0.72	4.5×	capex not covered by operating cash flow · external funding doing the work
CRWV	T3	0.36	-10.6	9.5	0.57	-0.4×	capex not covered by operating cash flow · external funding doing the work · interest not serviced from the business · interest coverage falling two quarters running
NBIS	T3	0.55	-2.5	7.5	1.37	-5.0×	capex not covered by operating cash flow · borrowing more than it invests — funding operations · interest not serviced from the business
WULF	T3	-0.13	-1.7	5.2	3.48	-1.7×	capex not covered by operating cash flow · borrowing more than it invests — funding operations · interest not serviced from the business
APLD	T3	-0.02	-1.8	-0.5	-0.26	-2.9×	capex not covered by operating cash flow · interest not serviced from the business
DLR	T4	1.83	1.1	0.9	0.67	2.7×	not scored on these metrics
EQIX	T4	0.79	-1.0	3.4	0.71	3.5×	not scored on these metrics
ARCC	T5	—	1.1	1.9	—	2.5×	not scored on these metrics
OBDC	T5	—	1.1	-1.8	—	1.2×	not scored on these metrics
BXSL	T5	—	-0.6	0.6	—	1.2×	not scored on these metrics
OWL	T5	23.03	1.3	0.6	10.82	3.9×	not scored on these metrics
APO	T5	—	6.0	4.8	—	28.6×	not scored on these metrics
ARES	T5	25.50	1.6	0.7	10.30	1.9×	not scored on these metrics
NVDA	T6	19.12	119.1	0.0	0.00	544.6×	not scored on these metrics

The credit environment

ICE BofA option-adjusted spreads. The percentile is against three years of the same series — the honest way to read a tight market: a spread can be historically low and still be moving fast.

IG corporate OAS

0.79%

20 days+3 bp

60 days-2 bp

3-year percentile22th

BBB OAS

0.98%

20 days+3 bp

60 days-3 bp

3-year percentile16th

B OAS

2.94%

20 days-3 bp

60 days-11 bp

3-year percentile29th

High-yield OAS

2.77%

20 days-1 bp

60 days+0 bp

3-year percentile17th

BBB − IG 0.19% +0 bp 20d Quality compression inside investment grade

B − BBB 1.96% -6 bp 20d The junk premium — what T3 pays over T2-grade credit

HY − IG 1.98% -4 bp 20d The market-wide risk appetite in credit

Treasury 10y / 2y 4.69% / 4.33% 2026-07-24 The risk-free leg — every spread above sits on top of this

What this board does not measure

Single-name credit default swaps. Not available from any data source we hold — CDS is an institutional product. Where a rung's market-implied cost is quoted here, it is the rating cohort's spread, not the issuer's own.

Issuer bond spreads. Awaiting FINRA TRACE, the tape of actually traded corporate bonds. Until then the funding side is measured from filings and the market side from cohorts.

Distance to default. The divergence signal — equity rising while credit quality deteriorates — needs a structural default measure. Buildable from data we already hold; not yet built.

SPV and ABS leverage. The private-credit rung sits off corporate balance sheets, so every filings-based ladder — including this one — understates total build-out leverage.

Sources: issuer figures computed from SEC filings (trailing four quarters, normalised statements, cross-checked against as-reported filings); credit environment from ICE BofA option-adjusted spread series via FRED; Treasury curve from vendor data. Interest coverage is operating income divided by interest expense. Landlord and proxy rungs are deliberately excluded from the stress score — REIT leverage is normal by construction, and financial vehicles do not fund capex from operating cash flow. This is an investment diary, not investment advice.