Morgan Stanley favors stronger cloud-company bonds as AI debt risks divide borrowers

Morgan Stanley is recommending stronger cloud-company bonds rather than treating debt financing the AI buildout as one trade. In a November 10 note covered by CNBC on November 11, strategist Fernanda Lima favored high-quality hyperscalers—the largest cloud operators—over data-center and neocloud borrowers facing greater risks around funding and getting projects built.

The same building boom, different financial cushions

The recommendation comes after a flood of AI-related borrowing unsettled bond investors this year. Yields rose for many borrowers, while the cost of protection against default increased. Lima’s response is selective: she sees a better balance of potential reward and risk in high-quality hyperscaler bonds, not an equally attractive opportunity across AI infrastructure debt.

The dividing line is financial resilience. Investment-grade hyperscalers have strong balance sheets, while lower-rated companies tend to carry more leverage. Lima expects financing terms to diverge further between stronger and weaker borrowers. For data-center bonds and neoclouds, she anticipates greater swings in credit pricing because of execution and funding risks.

For HQ hyperscalers, strong balance sheets and earnings cushions provide the capacity to absorb higher borrowing costs, while attractive prospective returns sustain the incentive to invest

Fernanda Lima, Morgan Stanley strategist, in the November 10 note quoted by CNBC

Recent borrowing offers a market test

Lima expects credit markets to remain open to hyperscalers, which have already demonstrated a willingness to pay more to borrow. Microsoft, Meta Platforms and Oracle issued new U.S. bonds after September 30. Alphabet tapped Europe, and some of Meta’s bonds also went to European investors. Together, the issuances totaled $67 billion across the two markets.

Most of those borrowers subsequently saw their bond spreads narrow from the levels at which the deals were priced, Lima said. Narrowing spreads generally accompany higher bond prices. Oracle was the exception: its spreads widened. That distinction keeps the bank’s preference from being a claim that every large cloud borrower is enjoying the same reception.

Her valuation argument also depends on the earlier selloff. Lima said hyperscaler spreads remained near their widest levels of the year and at levels consistent with A/BBB credit ratings. She described the resulting risk-reward balance as the best it had been all year, even after the improvement following most recent deals.

A quiet year-end, then a larger spending bill

Morgan Stanley expects little, if any, further AI-company bond issuance during the remainder of 2026. Its spending outlook points in a different direction for the following years: the bank forecasts $1.66 trillion in AI-company capital expenditures in 2027, up 75% from 2026, followed by $1.77 trillion in 2028. Those are forecasts, not committed spending totals.

Project schedules remain a complication. Lima warned that local data-center restrictions could delay construction. She nevertheless considers hyperscalers better equipped to handle uncertainty over securing power and starting new facilities. Her preference therefore extends beyond their ability to service debt: it includes their capacity to keep pursuing construction plans when the route to new capacity becomes less predictable.

Bonds may not be the only funding route. Lima said she would not be surprised by more transactions, including equity issuances, to bridge the gap between investment and monetization. Her recommendation favors borrowers able to absorb that waiting period; it does not assume the gap between spending and earning money from AI has already closed.