Key Takeaways
- The AI boom is generating tremendous physical infrastructure needs. Data centers, power production, and chips must be built to support AI’s growth, creating long‑lasting investment opportunities.
- Many of the biggest areas of opportunity are in bonds and loans. Large, established technology companies are increasingly borrowing to fund infrastructure spending, opening a door for fixed income investors.
- Focus on well-structured deals with claims on hard assets. By investing in essential infrastructure through secured financings, one can seek steady returns while mitigating risk – even as technology continues to evolve.
Figure 1: Hyperscalers’ capital spending is expected to rise further
The key question is how to commit capital to long-duration projects while technology and business conditions evolve. In our view, rather than trying to pick AI winners, the answer is to focus on the infrastructure layer itself – one layer below the AI applications – through enforceable collateral, control over key contracts, and protections that help ensure repayment even if things don’t go as planned.
Stewardship also matters in how these projects are developed. The strongest data center investments are built in partnership with local utilities and communities. Done well, they can add needed power and grid infrastructure, distribute fixed costs, support local economic development, and avoid worsening affordability by ensuring that communities share in the benefits of the buildout. We look for projects where the data center contributes to the broader infrastructure needs of the region, rather than simply adding power demand to an already constrained system.
Counterparty quality matters as well. Large global hyperscalers – the biggest tech companies and cloud service providers – typically have diversified revenues, strong balance sheets, and long-term strategic flexibility, even as their capital spending continues to rise. By contrast, tenants tied solely to a single AI application and still operating at a loss present very different risks.
What a well-structured deal looks like
Of the $5 trillion in potential capital spending through 2030, we estimate that upwards of 40% may need to be financed through debt markets (see Figure 2). This anticipated wave of new borrowing makes it especially important to be selective and to structure deals carefully (for more, see our 22 May publication, “AI Credit Expansion: Assessing the Micro and Macro Risks”).
Figure 2: Debt markets will play a significant role in AI infrastructure funding needs
In this context, secured financing means investors have direct claims on the physical assets, lease revenues, and contracts that generate cash flow – not just a general promise to repay. In GPU financings, that extends to liens on the chips themselves and the accounts that collect revenue. Key contracts, such as data center leases and power purchase agreements, should also be pledged and locked into the collateral package.
Equally important are cash flow controls. If revenue falls short or debt coverage ratios drop below agreed thresholds, a well-designed structure can redirect cash into reserve accounts to protect lenders – with debt paid down more quickly if the shortfall persists.
When it comes to deal pricing, credit spreads should reflect not just today’s risk, but also the supply outlook in a sector where longer-term debt issuance is rapidly ramping up. In our view, investors should seek to maximize liquidity, recognizing that bespoke, private-credit-style safeguards do not have to come at the expense of tradability. Structures such as Rule 144A private placement bonds can help preserve flexibility and accountability for issuers, while allowing large investors known as qualified institutional buyers (QIBs) to trade the securities in a deep and liquid market (for more, see our March 2026 commentary, “Spreads May Be Converging Across Public and Private Markets, But Liquidity Is Not”).
Areas of caution
In a sector awash with enthusiasm, some of the most important decisions involve knowing what to avoid.
Speculative builds without contracted cash flows. Repayment that depends on future demand rather than signed, enforceable contracts introduces risks that may not be adequately compensated. Investors should look for deals backed by committed agreements with creditworthy counterparties. We prefer structures where repayment is driven by committed contracts with clear remedies and lender control over the associated cash flows.
Maturity and repayment risk. A 20-year hyperscaler lease financed by a five-year debt maturity creates risks if market conditions make refinancing a challenge. Given the volume of five-year maturities being issued across the sector, a refinancing wall of potentially hundreds of billions of dollars could emerge in 2030. Structures that pay down debt over time from lease cash flows – rather than relying on refinancing – are more resilient. The same principle applies to residual value: It is nearly impossible to predict what data centers or GPUs will be worth in five to 10 years, so a conservative approach assumes zero residual value, with debt paid down entirely from contracted cash flows. Our bias is for liabilities that amortize from contracted cash flows and do not depend on favorable refinancing conditions down the road.
Lack of transparency into underlying contracts. Lease termination provisions, construction delay clauses, force majeure clauses, power interruption remedies, and casualty-event mechanics can all materially reduce contractual cash flow or reduce payments below what is needed for debt service. Investors who cannot conduct diligence on the underlying contracts face risks that are hard to fully assess. If we cannot diligence the actual contract mechanics, we treat the risk as unpriced and step back.
Discipline over FOMO
The expansion of AI infrastructure is a multiyear investment cycle that will generate deal flow across asset types, geographies, and capital structures. The sheer volume of expected financing needs is an advantage for disciplined investors: There is no need to chase any single transaction.
We believe the most durable approach involves focusing on the infrastructure layer rather than making bets on which AI applications will ultimately prevail. In a sector defined by rapid change, investing one layer below the technology itself – with real assets, strong collateral, and clear cash flow controls – is how debt investors can participate without sacrificing the discipline that protects capital when cycles turn.
