When hyperscalers borrow

AI-related borrowing is creating a significant credit opportunity, but bond investors should prioritise project execution, financial resilience and contractual protection over the technological promise of hyperscalers.

• Hyperscalers issued approximately $182 billion of high-grade bonds by August 2026, versus $93 billion during all of 2025, while they represented about 4.8% of the Bloomberg US Corporate Investment Grade Index in July 2026.
• Specialist data-centre developers are also borrowing in high yield to fund buildings, power connections and cooling systems; direct data-centre exposure reached roughly 3.4% of the US corporate high-yield index by August 2026.
• The investment cycle supports utilities, electrical equipment, engineering and construction, but the credit case depends on durable cash flows, contracted demand and sufficient spread compensation for leverage and execution risk.
• Physical delivery is the central risk: capacity under construction in major North American data-centre markets equalled almost 70% of existing inventory in the first half of 2026, while power connections, permits, equipment, inflation and supply chains can delay revenues and increase funding needs.
• Oracle’s September force-majeure notice concerning Project Jupiter in New Mexico, caused by permitting problems affecting a gas-pipeline component, illustrates why investors must identify who bears cost overruns, delays and incomplete infrastructure; leases, joint ventures and project vehicles can also obscure the true level of financial commitments.

Takeaway: The opportunity lies in selective credit exposure rather than indiscriminate financing of AI growth: investors should favour issuers and bonds with resilient balance sheets, secured power, credible completion schedules, strong tenants, seniority, guarantees and manageable refinancing needs, while recognising that even investment-grade customers do not automatically make the underlying borrower investment grade.