Despite AI hype, Wellington finds the risk/return for data center construction loans unappealing. Spreads have compressed by 200 bps while leverage has surged to 90% LTV. The firm is cautious about the terminal value and exit risk if a primary tenant leaves, making most current deals a "pass."

The primary risk for investment-grade AI debt is not weak company fundamentals, but rather massive supply overwhelming investor demand. In contrast, the high-yield market's main concern is construction risk, including project delays and cost overruns on new data centers, representing a shift to asset-level analysis.

Despite market hype, Madison avoids asset classes like office and data centers. They view them as binary investments—success or failure—with high capital costs and low liquidity. They prefer residential assets where underwriting mistakes are less catastrophic, protecting investors from the potential for major principal loss.

Lenders financing data centers are increasingly divorced from the underlying economics. They don't care about GPU utilization; they look through the facility to the prime credit of the hyperscaler tenant with a long-term lease. This financialization fuels overbuilding, regardless of actual AI demand.

To navigate the AI boom, Stonepeak assesses data center risk with a two-axis matrix: customer creditworthiness (e.g., Google vs. OpenAI) and location desirability (e.g., Northern Virginia vs. a remote farm). This framework clearly distinguishes between a safe, long-term contract with a tech giant in a prime market and a speculative bet on a cash-burning startup in an unproven location.

Investors are treating AI-related debt differently based on risk. Broad, unsecured bonds from hyperscalers are widening due to long-term investment risk. In contrast, securities backed by existing, cash-flowing data centers (ABS and CMBS) are more stable because the assets are already operational and leased.

Investors are financing the AI data center build-out not based on AI's potential profitability, but on the triple-A credit ratings of hyperscalers. These tech giants are contractually locked in, making projects a reliable, low-risk return for financiers regardless of AI's ultimate success.

Financier Blue Owl Capital takes on risky equity positions in massive AI data centers by applying a real estate model. It mitigates risk by structuring deals to receive regular payments on equity and locking tenants like Microsoft into 15-year leases that are extremely difficult to exit.

Evaluating data center investments is like analyzing net lease real estate. With a tenant like a MAG-7 company, the investment is primarily a bet on the counterparty's creditworthiness, not the long-term value or potential obsolescence of the physical data center itself.