AI Debt Issuance Falls to $23 Billion in September as Lenders Reassess Returns

AI Debt Issuance Slows Sharply as Investors Reassess Infrastructure Spending

Global borrowing linked to artificial intelligence infrastructure fell sharply in September, marking a shift in the financing environment behind one of the technology industry's largest investment cycles.

AI-related debt issuance declined to approximately $23 billion in September 2026, nearly half the previous month's volume and around 80% below the $113 billion peak recorded in June, according to Morgan Stanley data cited by the Financial Times. Companies raised an estimated $466 billion in AI-linked debt during 2026 through September, compared with $101 billion over the corresponding period in 2025. <Cite refs={["turn303934news18","turn303934search1"]}/>

The figures highlight the scale of financing required to build the infrastructure supporting AI development and deployment. They also indicate that investors are paying closer attention to how companies plan to turn substantial capital expenditure into sustainable revenue and cash flow.

The slowdown, however, should not automatically be interpreted as a collapse in AI investment. Morgan Stanley attributed much of the decline to companies raising substantial funding earlier in the year. Concerns about returns, rising borrowing costs and resistance to new data-centre construction are adding to the pressure.

A sharp reversal after record borrowing

June's $113 billion issuance peak illustrates how quickly financing activity accelerated as technology companies and infrastructure developers sought capital for AI expansion. By September, monthly issuance had fallen to $23 billion.

The year-to-date figure remains substantial. At $466 billion, AI-linked borrowing through September was more than four times the comparable period's $101 billion in 2025.

The figures cover AI-related financing across public bonds and private placements, according to the reporting. They should not be confused with total technology-sector debt or with capital expenditure alone.

The US investment-grade bond market experienced an especially notable pause: no AI-related bonds were issued in that market during September, following approximately $306 billion in borrowing by major technology companies from January through August. This is a narrower measure than the global AI-related issuance total. <Cite refs={["turn303934search1","turn303934news28"]}/>

The distinction matters because a quiet month in one segment of the market does not establish that every source of financing has closed. Companies may use different currencies, lending arrangements, private placements or other funding structures depending on their credit profiles and project requirements.

Why AI infrastructure demands so much capital

AI development depends on physical infrastructure as well as software. Large-scale computing requires advanced processors, servers, networking equipment, data-centre buildings, reliable electricity, cooling systems and supporting grid infrastructure.

These projects often require major spending before they generate meaningful revenue. Developers must secure sites, arrange power, install equipment and attract customers while managing construction schedules and operating expenses.

Debt financing can help companies spread these costs over time, but it also creates repayment obligations. If demand grows more slowly than expected, facilities take longer to become operational or electricity costs rise, the cash generated by a project may fall short of expectations.

Investors are therefore looking beyond announcements of new capacity. They want greater visibility into customer contracts, project completion, equipment utilisation and the expected timing of cash flows.

Major AI financing deals face greater scrutiny

The financing requirements are visible in several large transactions reported during 2026.

The Financial Times reported on a financing package of approximately $60 billion connected to Anthropic's lease of Google's advanced chips, with Broadcom providing guarantees for part of the structure. The package includes different debt tranches with different levels of security and credit exposure. <Cite refs={["turn303934news22"]}/>

Separately, The Times reported that SpaceX was pursuing approximately $40 billion in financing to acquire Nvidia AI chips. The proposal illustrates the scale of funding being considered for computing infrastructure, but a reported financing plan should not be treated as a completed transaction. <Cite refs={["turn303934news24"]}/>

Such deals show why lenders are examining more than the projected growth of AI. They must evaluate who carries the debt, what assets or guarantees support it, how repayment will be funded and what happens if the underlying technology or business model fails to deliver the expected returns.

Different financing structures can also distribute risk unevenly between borrowers, lenders and investors. A large headline figure does not necessarily mean that every participant faces the same exposure.

Suppliers face both opportunities and risks

The AI infrastructure expansion continues to create potential opportunities for suppliers of electrical equipment, power-grid components, liquid-cooling systems, cybersecurity products and networking hardware.

However, these businesses may also be affected if developers delay construction, reduce planned capacity or renegotiate financing. A slowdown in project approvals can push back equipment orders and create uncertainty around supplier revenue.

The wider issue is whether AI-generated income can justify the enormous cost of building and operating the infrastructure. Strong demand for computing does not guarantee that every project will earn an attractive return, particularly when borrowing costs, competition and operating expenses remain significant.

What happens next?

The next phase of AI financing will depend on whether the September slowdown proves temporary or develops into a longer period of more selective lending.

Investors will be watching new issuance volumes, borrowing costs, credit spreads, financing terms and the progress of major data-centre projects. They will also look for evidence that AI services can generate enough revenue and cash flow to support the industry's expanding infrastructure commitments.

A recovery in borrowing would suggest that companies can continue securing capital, although lenders may still demand stronger protections and clearer evidence of project viability. A prolonged slowdown accompanied by wider credit spreads, cancelled projects or delayed construction would point to deeper financing pressure.

For now, the data indicate a significant reduction in new AI-related borrowing after an exceptionally active first half of the year. The central question is not simply how much capital the industry can raise, but whether the investments it finances can deliver returns that justify their cost.

The next development to watch is whether AI financing rebounds in the final quarter of 2026—and whether that borrowing supports projects with credible paths to revenue and profitability.