AI capex is going on the credit card
Standard Chartered estimates debt will fund 40% of AI capex by 2027, up from 15% in 2025. PIMCO shows software loans bidding 8 points cheap into a $62 billion maturity wall. Operators themselves are scrutinizing the bill — Crusoe says the cheapest hourly quote doesn't buy the lowest total cost.
- Standard Chartered's Money Insights podcast estimates debt financing as a percentage of total AI capex is rising from 15% in 2025 to 25% in 2026 to 40% in 2027, while Big Tech's share of AI capex is expected to decline from 75% to 65%. m:…5642
- Rising funding risk was one of the main reasons the bank's AI confidence index was downgraded for October, the podcast says, alongside signs that AI lab revenue growth is moderating; the bank says it prefers US healthcare and US financials to high-beta AI exposure. m:…5642
- PIMCO reports software has been one of the worst-performing pockets of the leveraged loan market this year despite its more-than-10% weight in the Morningstar LSTA US Loan Index; the average bid price for software loans is about 8 points below non-software peers, with about $62 billion maturing in 2028 and 2029. m:…3575
- PIMCO sees no causal relationship between the recovery in software equities and relief for software loan issuers, and says refinancing costs remain prohibitively expensive for many issuers relative to the low-rate era. m:…3575
- Crusoe cites an independent Surf Report study finding the provider with the lowest quoted price did not deliver the lowest total cost, argues true AI-infrastructure cost combines quoted price, staff time, and useful work produced, and claims 56% better ROI than a hyperscaler over three years — a vendor framework with company-reported results. Self-reported m:…7896
The debt share of the buildout
Standard Chartered's Money Insights podcast, dated October 20, puts numbers on the shift: 'We estimate that debt financing as a percentage of total AI CapEx is rising from 15% in 2025 to 25% in 2026 this year to 40% in 2027,' says Fukien. Big Tech's share of AI capex is expected to fall from 75% to 65%, pushing spending toward smaller companies that the bank says face higher execution risks on data-center build-outs because of limited track records. m:…5642
Rising funding risk was one of the main reasons the bank's AI bubble meter flagged a downgrade of its AI confidence index for October, the podcast says, alongside signs that AI lab revenue growth is moderating after a strong second quarter. The bank's stated stance: it prefers US healthcare and US financials to high-beta AI exposure while it waits on the macro backdrop and earnings season. m:…5642
Where tech credit already cracked
PIMCO's October 19 credit note looks at the software corner of the leveraged-loan market: the sector has been one of the worst-performing pockets this year, and it matters because software carries more than 10% of the Morningstar LSTA US Loan Index. Spreads tightened over the summer, retracing about a third of this year's widening, then widened again from early September. PIMCO sees no real causal relationship between the broader recovery in software equities and relief for software loan issuers — optics improved, economics didn't. m:…3575
The numbers: the average bid price for software loans is still about 8 points below non-software peers, about $62 billion of software loans mature in 2028 and 2029, and refinancing costs remain prohibitively expensive for many issuers relative to borrowing costs put in place during the low-rate era. The market is split: some issuers extended maturities for just a 25-bp spread increase over the overnight rate, according to PitchBook reports cited by PIMCO, while others faced 150 bps and discounts to par. m:…3575
The TCO turn
Crusoe, a cloud provider, published a framework on October 19 arguing that headline GPU pricing misleads buyers: an independent Surf Report study found the provider with the lowest quoted price did not deliver the lowest total cost. 'The provider with the lowest quoted price was not the one that delivered the lowest total cost,' says Omar Lari. The company's thesis: true cost comes down to three things moving together — the quoted price, the staff time it takes to run and keep the system running, and the amount of useful work produced. m:…7896
Its own numbers are company-reported: 56% better ROI than a hyperscaler over three years; 99% of paid compute time kept productive in a stress test with injected failures, versus roughly 94% and 96% for the other two providers; recovery from each disruption in about 9 minutes versus 13 to 14. The metric the company pushes is 'goodput' — the share of paid compute that turns into completed work instead of being lost to errors or overhead. m:…7896
A credit question, not just an equity one
My read: the funding mix of the AI buildout is changing faster than the narrative. Debt moves from 15% to 40% of capex while credit markets already price tech stress at an 8-point discount — and the vendors selling the compute are publishing TCO frameworks to justify the spend. The buildout is being repriced as a credit question, not just an equity one. m:…5642 m:…3575
What is still unresolved
- Standard Chartered's debt-share figures are the bank's estimates and projections, not realized financing data.
- PIMCO's note is sector-level analysis; individual software issuers' credit paths will diverge sharply at the 2028–2029 wall.
- Crusoe's ROI, goodput, and recovery figures are company-reported from its own study, not independently replicated.
- Whether AI lab revenue growth is genuinely moderating or the Q2 strength was the outlier is not settled by one bank's reading.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.