In partnership with
One workspace where buyers, sellers, and AI close deals together
Instead of running your deals across scattered email threads and forgotten attachments, Aligned brings everything into one shared deal workspace.
Buyers get a single place to evaluate, loop in their team, and say yes. Sellers get visibility into what is actually happening when they're not in the room and AI Deal Insights that help you act accordingly, before it's too late.
Best part? Your first deal room takes only minutes to stand up.
D E E P D I V E
ISSUE 33 COMPANION · FRIDAY 16 OCTOBER 2026 · R. LAURITSEN
The Spread Test
A practical check on whether an AI company's contracts can carry its debt, with worked examples you can rerun on your own holdings.
Two companies reported on the AI build-out on Wednesday. ASML, which makes the machines that print the most advanced chips, said it generated €11.4bn of free cash flow in the quarter, holds €10.8bn of net cash, and can't build enough for 2027 or 2028. CoreWeave, which rents out Nvidia GPUs, reported about $104bn of backlog at the end of June, before more than $25bn of further commitments, and watched its 2032 bonds slide below 80 cents on the dollar.
Both sell into the same ambition. The difference sits underneath. One pays for growth with money customers have already handed over; the other with money it has to keep borrowing. Last week's deep dive asked how long each part of the build-out is promised for. This one adds the question lenders asked this week: what does the money cost, and does the contract cover it?
Hargreaves Lansdown · ASML cash and outlook · 14 Oct | CoreWeave · Q2 backlog · 11 Aug | SiliconANGLE · CoreWeave selloff · 14 Oct
THE FUNDING LADDER
Start at the top of the chain and work down. At each step, ask who puts up the money to build, and what they pay for it.
Layer and example | Who funds the build | Price of that funding, where public | This week |
Chip tools · ASML | Its own cash flow; €10.8bn net cash | No borrowing needed: €11.4bn Q3 free cash flow | 2027–28 supply short of demand; 2026 range raised to €44bn–45bn |
Foundry · TSMC | Its own profits: about $23.6bn in Q3 | No borrowing needed for a $60bn–64bn 2026 capex budget | Record profit; US shares −1.8% |
Model lab, the buyer · Anthropic | Private funding rounds: $125bn raised | Equity at $965bn in May; IPO talk of $1.5tn–2tn | Q3 revenue +45%, a slowdown from Q2; 80% of $518bn commitments fixed |
GPU landlord · CoreWeave | Debt and leases: $35.1bn and $16.5bn at June | Unsecured bonds yield over 15%; record CDS of 1,194.5bp | −15.5% Wednesday, +5.9% Thursday |
GPU landlord, diversified · SpaceX | $93.5bn cash at June, plus about $40bn of debt in talks | Five-year CDS about 194bp on 7 Oct, a record then | No change to notice periods disclosed |
Robot fleet · Tesla | Its own balance sheet | No fleet debt; cost is its shareholders' capital | Cybercab fleet past 300 in Austin, by its own count |
'Its own cash flow' means no fresh borrowing was needed for this build, not that the company carries no debt. Figures as reported.
ASML · Q3 release · 14 Oct | SiliconANGLE · TSMC profit · 15 Oct | Yahoo Finance · TSMC capex · 15 Oct | Crypto Briefing · Anthropic capital raised · 1 Oct | Reuters · Commitments · Sep | roic.ai · CoreWeave balance sheet · 14 Oct | Investing.com · ICE CDS record · 15 Oct | MarketWatch · Bond yields · 14 Oct | Reuters · IPO valuation talk · 15 Oct | TIKR · SpaceX cash · 7 Oct | KuCoin · SpaceX CDS · 8 Oct | Motley Fool · Tesla fleet · 15 Oct
My reading: the higher up the chain, the less a company needs anyone's permission to grow. The landlords sit near the bottom, and their revenue is the most heavily contracted. That's the uncomfortable bit. Contract quality and funding cost are separate questions, and this week the market only asked the second.
THE SPREAD TEST
For any business renting out AI capacity, three numbers decide whether the contract pays for itself. First, the contract yield: yearly rent minus running costs, divided by what the equipment cost. Second, the cost of debt: not the coupon on old bonds, but what new money costs today, read from where the company's bonds trade. Third, the repayment schedule: how fast the debt must be paid down while the equipment is still earning.
The spread is what's left of the contract yield after interest and repayments. If it's thin, the company needs every dollar of every contract, every year.
Here's CoreWeave on its own reported numbers. In the second quarter, net interest was $640m. Adjusted EBITDA was $1.51bn, which covers that interest about 2.4 times. Adjusted operating income was $128m: about 0.2 times. Same company, same quarter.
The gap between those two numbers is mostly depreciation, the GPUs wearing out. EBITDA ignores what it costs to replace them; operating income charges for it on CoreWeave's own schedule. If GPUs earn for less time than that schedule assumes, even 0.2 times flatters. A longer life helps, but replacement spending still has to be paid for. And neither ratio tells you whether a particular contract covers its own financing over its life. That needs contract-level terms, which aren't public.
WORKED EXAMPLE ONE: SAME CONTRACT, DIFFERENT MONEY
All figures hypothetical, and the same landlord as last week so you can compare. It buys $10bn of GPUs. One customer pays $3.6bn a year; running costs are $0.6bn. The loan is repaid in five equal annual instalments covering interest and principal. This time the customer stays. Only the interest rate changes.
Interest rate | Debt service a year | Cash after debt | Rent needed to break even |
6% | $2.37bn | +$0.63bn | 83% of the contract |
8% | $2.50bn | +$0.50bn | 86% |
12% | $2.77bn | +$0.23bn | 94% |
15% | $2.98bn | +$0.02bn | 99.5% |
Per year, with the customer paying in full. Illustrative only; ignores taxes, fees and empty months.
At 8%, the landlord has a cushion. At 15%, it needs essentially the whole contract just to service the loan. One percentage point of discount to a customer who renegotiates, or one empty month, and cash goes negative. That's why bond prices matter even when no contract has changed.
Two caveats, and they cut in opposite directions. CoreWeave's existing bonds still pay their original coupons, around 9.6% to 9.75%; the 15% is what the market wants today, which matters for the next loan and any refinancing, not the last one. And CoreWeave reportedly priced a lease-backed bond in June with a coupon of about 7%. That compares a June coupon with an October yield, so it doesn't isolate the effect of the collateral. Still, the gap is wide enough to suggest lenders price the customers' contracts more kindly than the company. Last week's thesis survives. It just needed a second line.
WORKED EXAMPLE TWO: THE BUYER'S FIXED BILL
Hypothetical again. A model lab has committed to $50bn a year of computing it can't cancel. Its other costs, staff and sales mostly, run at 30% of revenue. The table shows yearly profit at different revenue levels.
Revenue | Compute (fixed) | Other costs (30%) | Profit |
$40bn | $50bn | $12bn | −$22bn |
$60bn | $50bn | $18bn | −$8bn |
About $71bn | $50bn | $21bn | Break-even |
$80bn | $50bn | $24bn | +$6bn |
$100bn | $50bn | $30bn | +$20bn |
Per year. Illustrative only; not an estimate of any company's costs.
Going from $80bn to $100bn of revenue, a 25% rise, more than triples profit. Going from $80bn down to $60bn wipes it out. That's operating leverage, and it's why investors fixate on a lab's growth rate. Anthropic's 45% quarterly growth is extraordinary; the slide from '$2tn' to '$1.5tn' talk shows how much of the valuation rides on it continuing.
Q3's $16.8bn annualises to about $67bn. Analysts floated a $110bn run rate by year-end after the meeting, but the reports don't define it. If it means quarterly revenue times four, Q4 would need to be near $27.5bn, roughly 64% above Q3. If it means December's monthly revenue times twelve, Q4 could come in well below that. Ask which, before using the number.
Google and Amazon, on the supplying end, get revenue that doesn't depend on that growth. Anthropic keeps the leverage, both ways.
IF RATES MOVE: THREE PATHS
Party | Ten-year back under 5%, Fed holds | Ten-year stays 5%–5.3% | Fed hikes, ten-year above 5.3% |
GPU landlords | Spread widens; refinancing back on | Thin spread; every contract counts | Refinanced debt at or above today's ~15% |
Model labs | Easier IPO window | Valuation leans harder on growth | Valuation talk falls further |
Self-funded fleets | Little direct effect | Little direct effect | Valuation multiples squeezed, not funding |
Tool and foundry suppliers | Demand-driven as before | Demand-driven as before | Customers' borrowing slows orders, with a lag |
A conditional scenario, not a forecast. Timing and size depend on terms that are mostly not public.
WHAT TO WATCH
When | Checkpoint | What it would tell us |
21 Oct | Tesla Q3 results | Separate robotaxi revenue or miles would make the self-funded fleet a measurable business. |
c. 25 Oct (reported) | Anthropic prospectus made public | The cancellable share and counterparties. A bigger fixed bill raises the growth bar. |
28 Oct | SK Hynix Q3 call | Lower 2027 investment would be the first crack in ASML's queue. |
28 Oct | Fed decision | A hike pushes landlords' cost of new money higher; a hold gives them room. |
16 Nov (projected) | CoreWeave Q3 results | Interest above adjusted operating income again keeps the case live; cheaper unsecured debt on comparable terms weakens it. |
TURN THE ANALYSIS INTO A DECISION
The bull case for the borrowers: Huang is right, CoreWeave's cost of capital falls as its customers prove good for the money, and this week's 15% marks the top. Nothing rules that out yet.
The thesis strengthens with a Fed hike on 28 October, a bigger fixed bill in Anthropic's prospectus, or another CoreWeave quarter of interest above adjusted operating income. It weakens if the ten-year settles under 5%, or CoreWeave raises unsecured debt at materially lower yields on comparable terms. Two weeks ago the question was whose balance sheet stands behind the payment. Last week, for how long. This week: at what price.
YOUR WEEKEND REVIEW One holding, one horizon: the next 12 months. 1 Cash generated before interest: adjusted EBITDA, less tax and upkeep spending on existing equipment (last four quarters is a fair proxy). Reported operating cash flow usually has interest taken off already; add it back. 2 Interest due over the next 12 months (last quarter × 4 is a rough start). 3 Principal due over the next 12 months, from the debt maturity table. 4 Cash on hand plus undrawn credit lines today. 5 The gap: lines 2 + 3 (plus any committed growth capex), minus lines 1 + 4. If positive, that much must be refinanced or raised. Price only that amount at today's bond yield; existing fixed-rate debt keeps its coupon. No traded bonds? Use the latest loan rate, marked stale. A large gap next to line 1 means size small, however good the contracts look. |
This companion supports Issue 33 of iPrompt Signals, the Friday briefing on AI and robotics investing. Forwarded this? Subscribe at [SUBSCRIBE LINK] to see this week's tripwires checked against the results.
Information and education only, not financial advice. Worked examples are hypothetical; Anthropic's figures come from press reports and a leaked prospectus. iPrompt Signals is not a registered investment adviser.
iPrompt Signals Deep Dive · Issue 33 · 16 October 2026 ·