AI debt passed $450bn and now your funding costs more

0:05Host:You're listening to Leaders Insights. Today's subject: AI debt passed $450bn and now your funding costs more. The Bank of England released the record of its 25 September financial stability meeting on Wednesday, and the phrase credit people keep rereading is "circular arrangements". Translate that for a finance team that will never build a data centre.

0:26Expert:Circular arrangements means money moving in a loop: a chip supplier or cloud provider invests in or lends to a customer, and that customer then buys capacity or hardware back, so one dollar can appear as revenue in one set of accounts and as an obligation in another. The Financial Policy Committee, the Bank of England body that watches risks to the whole system rather than inflation, said rising AI indebtedness combined with opacity and those arrangements "can complicate the assessment of risks and could amplify losses if expectations disappoint". That is their language, not mine.

1:03Host:Fine, but why should a CFO in packaging or healthcare services care what a UK committee thinks about Nvidia's customers?

1:10Expert:Because you borrow in the same pond. Reuters reported the committee's view that the likelihood that interconnected vulnerabilities crystallise has risen, and in the same record it said oil and gas price rises had pushed bond yields to levels not seen since 2008. When AI issuers take up that much of the available demand at the same time as yields are high, everyone queueing behind them pays for the crowding.

1:35Host:Give me the size.

1:37Expert:Global AI-related debt issuance was around $450 billion by early September, a Morgan Stanley estimate the committee cited, and The Next Web notes that is more than double the whole of 2025. Reuters also reported Morgan Stanley's forecast of up to $570 billion for 2026. The forward number is the one that makes treasurers sit up: The Tribune reported the committee citing JP Morgan analysts' estimate of about $4.1 trillion of debt-financed AI capex between 2026 and 2030.

2:12Host:Those are forecasts. Forecasts are free. What is already visible in prices?

2:18Expert:Fair challenge. Two things are observable now. First, in high yield, meaning bonds from issuers rated below investment grade, Reuters reported AI infrastructure supply has reached $40 billion this year against $12 billion for all of 2025, using BNP Paribas data. Second, who is paying what. SoftBank Group, rated BB+, which is one notch below investment grade, paid 8.625% on 3.5-year notes, 9.25% on 5.5-year and 9.75% on 7.5-year bonds, yields analysts told Reuters are normally paid by much weaker credits.

3:00Host:So the market is not closed. It is just expensive.

3:04Expert:It is expensive and selective. Reuters reported that leveraged finance buyers have clustered around double-B borrowers, and that demand is strongest for issuers with predictable revenue, long-term contracts, tangible assets and established customers. The structural detail I would underline for any borrower: many AI borrowers need heavy upfront investment before producing reliable cash flow, which makes their paper hard to absorb for managers of collateralised loan obligations, the funds that buy most leveraged loans and bundle them into securities.

3:41Host:Spell out why a CLO manager's portfolio rule is my problem.

3:44Expert:Because those managers are the marginal buyer for a large slice of floating-rate corporate debt. If their buckets are filling with AI names, or if they refuse to take them, the demand curve for the rest of the asset class moves, and your pricing conversation in January is not the one your banker promised you in June.

4:03Host:Where else is the pressure showing?

4:05Expert:In securitised markets, which pool contracted cash flows into bonds. The Real Deal reported Barclays data showing spreads on hyperscaler asset-backed securities at 150 basis points over the five-year Treasury earlier this month, 16 basis points wider than a year earlier. A basis point is one hundredth of a percentage point. Hyperscaler corporate debt was 31 basis points wider than a year earlier, at 165 over SOFR, which is the overnight dollar benchmark rate most floating loans now reference. One analyst's explanation in that piece was blunt: the hyperscalers are out raising funding, "the competition for capital is real", and that partly explains higher yields.

4:50Host:You said hyperscalers. Define it before someone writes in.

4:54Expert:The very large cloud operators building and renting out data centre capacity at industrial scale. Alphabet, Amazon, Microsoft, Meta, Oracle sit in that group.

5:04Host:Now the uncomfortable part. Is this the Bank of England being the Bank of England, warning about the last four crises it failed to predict?

5:12Expert:That is a reasonable suspicion, and I will separate fact from opinion. Fact: the committee itself said AI and semiconductor stocks fell sharply in July while market functioning stayed orderly, and it left its countercyclical capital buffer, the dial it turns to make banks hold more capital against the cycle, at 2%. A regulator genuinely braced for impact usually turns that dial. Opinion: the committee is signalling, not intervening, and it is honest about that. Governor Andrew Bailey wrote alongside the record that a more formal framework may emerge over time, but regulation is not the right place to start, and that understanding, testing and credible points of intervention must come first.

5:58Host:Then give me the strongest argument that this whole worry is overdone.

6:02Expert:The capacity argument, and it is a good one. Goldman Sachs' Amanda Lynam has said plainly that her team is not concerned about access to capital for this theme, and her team's arithmetic is that if the hyperscalers were simply sized like the three biggest US bank issuers, you would find roughly $510 billion of incremental capacity in the US investment-grade bond market alone. On that reading, what we are watching is a supply-driven repricing of spreads, not a solvency event.

6:32Host:And your honest read of that argument?

6:34Expert:Half persuasive. The same Goldman discussion concedes the thing that worries me most: capex and cash flow from operations are converging, so the excess liquidity cushion of past years is thinner. Capacity to absorb debt and the price of that debt are different questions. The market can swallow every deal and still hand you a hurdle rate two hundred basis points higher than your board approved last year.

6:59Host:What is the counterfactual that would prove the committee right?

7:02Expert:Slower productivity payoff. The committee said growth and fiscal outlooks partly rest on AI delivering significant productivity gains, and that reassessing those expectations could affect not only AI-related asset valuations but sovereign debt markets. Bailey made the same point in a BBC Radio 4 interview, saying the contribution may not be as great as financial markets expect and could take a lot longer to arrive. If that is right, the repricing does not stay inside the technology sector. It arrives as a higher risk-free rate in your discounted cash flow model.

7:37Host:Last question. One thing to do this week.

7:40Expert:Pull your maturity schedule for the next eighteen months and reprice every tranche at today's secondary market level plus fifty basis points, not at the coupon you are currently paying. Then take the gap to your audit committee with two numbers: the interest cost if you refinance at that level, and the capex you would cancel to protect the covenant. Ask your relationship banks one direct question while you are at it: are they aggregating related-entity exposure, so that a data centre financing vehicle counts against the same group limit as its parent? Reuters reported that some institutional investors are nearing single-name limits once debt issued through related structures is aggregated back to the same technology companies. If your lenders are near their limits on names you have never heard of, you will find out at pricing, and that is the worst possible moment.

8:31Host:This episode draws on Bank of England sees growing risk that dangers from AI and debt will materialise, ‘Not everybody always wins’: Bank of England’s Bailey on the AI boom, Bank of England says AI investment boom raises financial risks, debt financing could reach USD 4.1 trillion by 2030 - The Tribune, Bank of England warns of AI debt, US Market: AI credit surge tests lenders as future revenues remain uncertain, Analysis-AI Borrowers Face Tough Sell in Risky Corners of US Credit Market. We'll stop there. The CFO tools for making the call are at mba-training.com.