NEW YORK — PIMCO strategist Lotfi Karoui outlined a split outlook for artificial-intelligence debt on Nov. 16, 2026: slower data-centre construction could support some existing bonds even as it makes life harder for highly leveraged infrastructure companies. The AI bond spreads debate comes as investors assess how construction delays, future borrowing and computing demand interact.
The distinction is important. A bond can become relatively more attractive because fewer competing securities reach the market, without the company or project behind every AI loan becoming financially stronger.
AI bond spreads: Why a construction slowdown could help
PIMCO’s published comparison showed that selected investment-grade bonds linked to data centres leased by major technology companies traded at an option-adjusted spread about 146 basis points wider than comparable debt issued by their technology-company tenants as of Nov. 10. The difference had approached 180 basis points in late September.
A basis point is one-hundredth of a percentage point; 146 basis points equal 1.46 percentage points. This is a difference between two groups of credit spreads, not the total yield promised by a particular bond.
PIMCO argues that delays to new projects could reduce the supply of newly issued data-centre bonds. When fewer comparable securities arrive, existing issues may command better prices, narrowing their spread relative to the tenants’ senior debt.
This is a conditional market assessment, not evidence that those spreads have already narrowed since Nov. 10 or that every data-centre bond will gain value.
Why a shortage of computing capacity changes the equation
Large technology companies use data centres to deliver cloud services and AI applications. If construction slows while customers continue demanding computing power, already operating facilities could remain heavily used.
PIMCO’s analysis identifies diversified technology companies as better positioned to handle delays because they generally have several revenue streams and financing options. They may also be able to redirect spending across projects rather than depend on a single facility opening on schedule.
Analysis: Scarce computing capacity can support revenue for facilities already in service. But an incomplete project cannot automatically benefit from that scarcity: its owner may still be paying for land, equipment and financing while awaiting permission, electricity connections or completion.
That tension is central to deciding whether a bond’s spread adequately compensates investors for the risks attached to its cash flows.
Strong tenants and heavily indebted operators face different risks
The key dividing line is not simply whether an investment involves artificial intelligence. It is who owes the money, who pays for the computing capacity and what happens when construction or customer revenue is delayed.
Bonds backed by leases from diversified, financially established technology tenants may have protections that make scheduled payments more resilient. Even then, investors must examine the documents: not all leases transfer the same obligations, and a strong tenant does not eliminate structural or refinancing risks.
Specialist infrastructure operators with greater leverage can be more exposed. PIMCO points to continuing interest expenses, lease obligations and commitments to purchase graphics processors even when expected revenues move further into the future.
WWC has examined a related issue in its coverage of tighter financing conditions for leveraged AI projects. The comparison concerns the cost and availability of finance, not a claim that all specialist cloud businesses face the same outcome.
The investment pipeline remains enormous, but forecasts are not guarantees
PIMCO’s Nov. 16 report shows consensus capital-spending estimates of approximately $1.17 trillion for 2027 and $1.30 trillion for 2028 across Amazon, Alphabet, Meta, Microsoft and Oracle. Those are aggregated market forecasts published in the firm’s research, not spending already completed or binding budgets approved by each company.
The firm expects some planned construction and expenditure to be postponed rather than permanently eliminated. Its central scenario therefore involves a more extended investment timetable, not the disappearance of future computing demand.
In an earlier Sept. 28 analysis, PIMCO estimated that AI-related issuers had accounted for almost one-quarter of US-dollar investment-grade issuance by nonfinancial companies in 2026 to that point. It found little evidence that the surge had directly forced unrelated corporate borrowers out of the market.
This helps distinguish two issues: new debt can place pressure on prices in its own sector without automatically creating a credit shortage for every borrower.
WWC’s explanation of US corporate-bond yields and risk premiums provides the relevant distinction: a bond’s headline yield, its credit spread and the return an investor ultimately earns are not identical.
What this could mean for AI financing
Analysis: A reduced pipeline of new bonds may support the relative value of existing debt, but less construction can also defer the service revenue required to repay infrastructure loans. Investors need to assess both forces at once.
For Philippine readers evaluating overseas funds or global technology exposure, concentration also matters. A portfolio with several issuers may still depend on the same group of technology tenants, power infrastructure providers and computing customers.
That issue connects with WWC’s examination of interconnected AI investment and supplier relationships. Such relationships are not inherently improper, but they can leave seemingly separate securities exposed to similar assumptions about demand.
No published forecast establishes that slower construction will produce a guaranteed profit for investors. Credit quality, the price paid and the terms protecting bondholders still matter.
What happens next
The next evidence will come from actual debt issuance, changes in comparable bond spreads, project-completion schedules and companies’ disclosures about computing commitments and cash generation. PIMCO’s November outlook should be tested against those developments rather than treated as an outcome already delivered.
A slowdown that mainly reduces new borrowing could benefit holders of certain existing securities. A slowdown that also weakens tenant payments or delays revenue could undermine that benefit. The balance will differ across transactions.
FAQ
What are AI bond spreads?
They measure the additional yield investors require on AI-related debt compared with a benchmark or comparable security. The precise benchmark matters when interpreting any quoted figure.
Did PIMCO say AI bonds will definitely rally?
No. It identified a possible narrowing in selected data-centre bond spreads if construction delays constrain future issuance, alongside risks for leveraged operators.
What does 146 basis points mean?
It equals 1.46 percentage points. In PIMCO’s Nov. 10 comparison, it was the spread difference between selected data-centre bonds and matched tenant debt, not a guaranteed investor return.
Why could slower construction help existing data-centre debt?
Fewer new bonds may reduce competing supply, while operating data centres may benefit if computing capacity remains scarce. Those effects depend on demand and deal protections.
Are leveraged AI infrastructure companies equally protected?
No. Interest, equipment and lease commitments can persist when projects are delayed. Their financial position depends on cash flow, debt terms and future customer demand.
