Asset life vs debt tenor: SpaceX's $40bn chip deal

October 7, 2026 · 9 min · Free, with full transcript

Can you fund four-year Nvidia hardware with thirty-year paper and keep an investment-grade rating? SpaceX is trying, with roughly $10bn of bank loans and $30bn of bonds led by Apollo, and the position here is that the structure sells residual value rather than credit. The episode walks through the BBB threshold, the June $25bn deal that drew $90bn of orders then traded near 85 cents with a 2.27 point spread, Michael Burry's $176bn depreciation claim, and the Bank of England's circular arrangements warning. You leave with a three-part test, life, lock and residual, plus the CoreWeave, Meta and Blue Owl precedents, an asset-life ladder to build, and the purchase obligations footnote to check.

Listen to the podcast

8 min

Chapters

Key takeaways

  • Fund an asset with debt no longer than the contracted cash flow that pays for it, plus only residual value someone has guaranteed in writing.
  • When economic life and accounting life differ by more than a year, use the shorter number for funding decisions.
  • Build an asset-life ladder with four columns, economic life, accounting life, contracted cash and funding tenor, and escalate any line where funding outlives contracted cash by over 24 months.
  • Put the uncovered refinancing exposure in the board pack as a number, not the coupon: on $30bn of four-year chips at 6% over ten years with half contracted, that is $15bn.
  • Ask your bond desk how much of your spread premium is disclosure quality rather than credit risk, since disclosure can be fixed within a quarter.
Read the full transcript

Host:You're listening to Leaders Insights. Today's subject: How do you finance $40bn of Nvidia chips without breaking your rating?. A chip you'll throw out in four years, paid for with a bond that matures in thirty. That's not financing, that's a bet dressed up in a credit rating.

Expert:And yet it's about to happen at $40bn, so either everyone's wrong or you're missing something. I'd argue it's a bit of both.

Host:Start me at zero. SpaceX wants roughly $40bn to buy Nvidia chips. Why can't they just write one big cheque to the bond market and be done?

Expert:Because the money isn't the hard part. Capital's lying around everywhere in AI right now. The hard part is one question: how long does the asset earn versus how long the debt lives, and who owns the gap when they don't match?

Host:Spell out the gap for someone who's never thought about it.

Expert:Picture buying a racehorse with a thirty-year mortgage. The horse is quick for three or four years, then it's a very expensive lawn ornament, but you're still paying the bank in year twenty-nine. A 2026 Nvidia rack is the horse. The bond is the mortgage.

Host:So why split it? The reporting says $10bn of bank loans and $30bn of investment-grade bonds — bonds safe enough for pension funds to hold.

Expert:Each layer does a different job. The $10bn of bank loans is the flexible layer — covenants, amortisation, a relationship manager you can phone when a chip order slips. The $30bn of bonds buys tenor and a buyer base: insurers and pension funds who can only hold paper rated investment-grade.

Host:And SpaceX sits at BBB — the second-lowest rung still inside investment grade.

Expert:That one notch is the whole ballgame. BBB lets those institutions in the door. Drop to high-yield — junk — and your $30bn tranche becomes a fantasy. So picking the rating and the story behind it isn't an investor-relations chore, it's a structuring decision.

Host:Here's where I want to push. Apollo's leading this — a private credit firm, not a bank syndicate. Why does that matter to a listener who thinks a lender is a lender?

Expert:Because of what's actually being sold. A bank syndicate underwrites whether the borrower can pay you back. Apollo's selling something narrower: what's a 2026 Nvidia rack worth in 2030? That's residual value — the leftover worth of the asset at the end — and private credit prices it for a living. Apollo's already financed these exact chips three times.

Host:Let me bring you three things people say about this stuff. Sort them for me — true, half-true, or wrong. First one: "The chips are collateral, so the loan is basically safe."

Expert:Wrong, and dangerously so. Nobody agrees what the collateral's worth. S&P says service lives past five years have held so far, but stays conservative. Google, Microsoft and Oracle depreciate over roughly six years. Michael Burry reckons the real economic life is two to three, and figures the industry's overstating profits by about $176bn across 2026 to 2028.

Host:That's not a rounding error between those views.

Expert:Apply that spread to a $30bn chip order and you've got tens of billions of collateral value that either exists or doesn't. "The chips are safe" is a sentence people say to avoid doing the arithmetic.

Host:Second belief: "It's investment-grade, so the market's comfortable."

Expert:Half-true. The market gave SpaceX access — the June bond, $25bn, drew around $90bn of orders. Access isn't the issue. Price is. Within weeks the paper sold off. The 2056 bonds have traded near 85 cents on the dollar, yielding about 2.27 points over Treasuries — that's junk-ish pricing on an investment-grade name.

Host:So the market said yes and then flinched.

Expert:It said yes to the company and no to the disclosure. Investors offered the earlier chip deals got a two-page term sheet. Two pages. For billions. That thinness has a price now, and you can read it in basis points.

Host:Third one, and this is the cynical take: "It's all circular — Nvidia funds the buyers who buy Nvidia chips, and it inflates itself."

Expert:True enough that a central bank said it out loud. Nvidia is both the vendor here and holds a large SpaceX equity stake. On 30 September 2026 the Bank of England's Financial Policy Committee — their financial stability watchdog — flagged rising leverage, thin transparency and what they literally called circular arrangements.

Host:So when Nvidia tells you how much demand there is for Nvidia chips—

Expert:You remember who's talking. That's a commercially interested number. Doesn't make it false. Makes it a sales figure wearing a forecast's clothes.

Host:Give me the clean version. If I'm a CFO and I never place $30bn, what's the actual test?

Expert:Three words on a whiteboard, in order. Life, lock, residual. Life: the honest economic life of the asset, not the number in your accounts. If the two differ by more than a year, the smaller one wins.

Host:Lock?

Expert:How much of the cash flow is contracted, by whom, for how long. CoreWeave closed an $8.5bn chip-backed facility in March because a named customer — Meta — had signed for at least $19bn. That's a lock. A general "we're buying chips" raise has none.

Host:And residual — the leftover value.

Expert:Who owns the asset at the end, and is it written down. Meta did the honest version on its $27bn Hyperion campus: Blue Owl's funds took 80%, Meta kept 20% and control, and Meta signed a residual value guarantee covering the first sixteen years. The debt left Meta's books. The risk did not.

Host:So the rule falls out of that.

Expert:Fund an asset with debt no longer than the contract that pays for it, plus only the residual someone's guaranteed in writing. Everything past that line is a bet on the refinancing market in 2036. Fine — but size it, and show the board that number, not the coupon.

Host:Walk me through the arithmetic, because "it's a bet" is easy to wave away.

Expert:$30bn of chips, four-year life. Straight-line, that's $7.5bn of value burning off every year. Fund it with ten-year bonds at 6% — inside what Meta actually paid in May. Interest is $1.8bn a year, $18bn over the life. Then the full $30bn principal lands in year ten, when the hardware is several generations into obsolescence.

Host:So to pay it back from the chips themselves—

Expert:The compute has to throw off roughly $7.5bn a year above costs and interest, in the first four years. If contracts cover half, then $15bn is pure refinancing bet. That $15bn is the number for the board pack. Nobody puts it there, because the coupon looks tidier.

Host:Where does this bite someone who runs a mid-cap company, not a near-trillion-dollar issuer?

Expert:Your version isn't a $30bn bond, it's a club facility where the covenants decide whether a delayed capex order becomes a polite waiver call or a default. And the landmine sits in your purchase obligations footnote — the non-cancellable commitments nobody reads until a lender does. SpaceX had about $28bn of those as of June, much falling due in 2027.

Host:One thing to do Monday morning. Just one.

Expert:Build an asset-life ladder. Four columns for every big capex line: economic life, accounting life, how much cash is actually contracted, and your funding tenor. Any line where the funding outlives the contracted cash by more than 24 months goes to the finance committee with a number attached.

Host:And if your spread's wider than your peers'?

Expert:Ask your bond desk how much of that premium is disclosure quality versus genuine credit risk. The credit part takes years to fix. The disclosure part you can fix in a quarter — and right now SpaceX is paying 2.27 points to prove it.

Host:This episode draws on Financial Times, CNBC Finance, Accounting Today, CFO Dive. That's the episode. If you want the structured version, the CFO track is at mba-training.com.

Read and go deeper