AI Borrowing Slowed Sharply After June’s Record Peak
Matthew Leo · Published October 23, 2026 · Business & Policy
The flow of new debt into AI infrastructure slowed sharply in September after a record-setting first half of 2026.
AI-related companies raised US$23 billion in debt during September, less than half August’s total and far below the US$113-billion peak in June, according to Morgan Stanley data reported by the Financial Times on October 10. The data includes public bonds and private placements.
Using those reported figures, September issuance was about 80 per cent below the June peak: a US$90-billion difference divided by US$113 billion. That calculation describes monthly deal volume. It does not prove that AI demand or construction spending fell by the same amount.
The market is slowing after an enormous year
Morgan Stanley estimated that companies had raised US$466 billion of AI-linked debt in 2026 through the period covered by the report, up from US$101 billion during the same period in 2025. In the US investment-grade market, the FT said AI-related bond issuance stopped in September after blue-chip technology groups borrowed about US$306 billion from January through August.
Those figures come from Morgan Stanley’s dataset as reported by the FT; the underlying transaction list is not public in the report, so Mapletechie could not independently reproduce the totals.
The slowdown is not the same as a credit shutdown. Morgan Stanley attributed much of the decline to the volume already raised earlier in the year. Investors are also asking harder questions about returns, political opposition to data-centre construction and how much exposure they already carry.
Bond buyers were already demanding more compensation
The change has been visible in pricing. Reuters reported on September 22 that spreads on AI-related issuers were around 115 basis points, compared with 78 basis points for the broader investment-grade market, based on Goldman Sachs and ICE BofA data.
Reuters interviewed portfolio managers who described selectivity rather than panic. The concerns were supply, uncertain returns on capital and concentration limits, not an immediate expectation that major hyperscalers would default.
That distinction matters. A borrower can still obtain financing and pay more for it. Over time, a higher borrowing cost changes which projects are viable, especially when a data centre needs years of power and customer commitments before producing a return.
Canada feels the financing conditions indirectly
Most of the debt in the reported totals was not issued by Canadian companies. The consequences can still reach Canada through cloud prices, construction decisions, pension portfolios and the availability of foreign capital for domestic projects.
Mapletechie’s analysis of DayOne’s IPO filing showed how power costs, depreciation and interest expense rise alongside new capacity. Our report on Broadcom’s reported financing for Anthropic also showed how a technology supplier can become a lender to its customer. September’s pullback connects those stories: the financing structure is becoming as important as the chips.
For Canadian institutions buying the debt, a useful exposure review should combine bonds issued by a technology parent, project vehicles and financing structures backed by the same customer. Looking at each security separately can hide concentration on one company’s spending plan.
What to watch next
- New issue concessions: wider spreads show how much extra yield buyers require.
- Project delays: postponements can reveal a gap between announced capacity and financeable capacity.
- Power and customer contracts: long commitments matter more when debt is expensive.
- Supplier financing: loans from chip or equipment vendors can support demand while concentrating risk.
- Canadian pension exposure: funds may hold the same AI buildout through public debt, private credit, infrastructure and equity.
September did not end the AI infrastructure cycle. It showed that lenders are no longer treating every part of it as equally easy to finance.
Tags: AI debt, data centres, credit markets, AI infrastructure, Canada