Morgan Stanley tells investors to buy hyperscaler bonds as AI debt fears spread

CoreWeave's bond yield topped 16% earlier this month — and Morgan Stanley thinks that panic is creating an entry point in the debt of the strongest AI spenders. The Wall Street firm is telling clients to stick with high-quality hyperscaler bonds as investors grow wary of the debt funding the artificial intelligence build-out.

The call matters because AI-related supply has flooded the bond market this year, pushing yields up and widening credit default swap spreads across the sector. Credit default swaps are essentially insurance against default, and wider spreads signal growing fear that borrowers won't pay. The divide between winners and losers is now the central question for anyone financing the AI infrastructure boom.

"We see an improving risk-reward for HQ hyperscalers and expect more spread volatility for data center bonds and neoclouds given execution and funding risk," Morgan Stanley strategist Fernanda Lima said in a note Tuesday.

Why is AI debt suddenly under pressure?

The numbers tell the story of a market that repriced fast. Many AI-linked names saw yields jump and five-year credit default swap spreads widen as supply flooded in. Bond yields move inversely to prices, so rising yields mean falling prices — investors selling, or demanding a bigger premium to hold the risk.

Lower-grade borrowers took the hardest hits. CoreWeave, the neocloud company, became what the market now treats as the poster child for AI debt worries after its bond yield topped 16% earlier this month. That is a level associated with serious distress for investment-grade observers, though CoreWeave carries a junk rating.

The company then delivered a partial rebuttal. CoreWeave reported a revenue beat for its third quarter and raised its full-year guidance after the bell Tuesday. Its shares jumped more than 25% on Wednesday. The U.S. bond market was closed that day in observance of Veterans Day, so equity traders delivered the verdict alone.

Morgan Stanley's read is that this bifurcation will deepen. Hyperscalers carry investment-grade ratings and strong balance sheets. Junk-rated companies tend to be more leveraged. As borrowing costs climb, that gap in financing terms between higher-quality and lower-quality companies should widen — a structural shift in how the AI build-out gets funded.

Who is still borrowing, and how much?

The credit market remains open for hyperscalers, Lima said. In fact, they have already shown a willingness to pay more to issue bonds.

The issuance list since Sept. 30 reads like a who's-who of Big Tech:

• Microsoft has issued new bonds in the U.S.
• Meta Platforms has issued in the U.S., with some of its bonds also placed in Europe.
• Oracle has issued new bonds in the U.S.
• Alphabet went to the European market for its latest issuance.

The total came to $67 billion across both markets since that date, Lima noted. That figure alone captures the scale of the AI capital machine — and why investors started asking hard questions about who ultimately pays for it.

The market's response, according to Lima, has been mostly forgiving. Spreads have somewhat tightened on most names since the recent deals were priced. When spreads tighten, it usually means the bond price has gone up. Only Oracle saw spreads widen.

"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," Lima wrote.

That last point is the core of the bull case. These companies aren't borrowing to survive; they're borrowing because the expected returns on AI infrastructure still clear their internal bars, even at higher funding costs.

What comes after the election?

The rest of the year should be quiet on the issuance front, with little activity, if any, from AI companies, Lima said. 2027 is where she expects a different story.

Morgan Stanley anticipates $1.66 trillion in capital expenditures from AI names next year — up 75% from 2026 — and $1.77 trillion in 2028. Those figures frame the stakes: the AI build-out is not a one-year spending spike but a multi-year commitment that will keep testing credit markets, power grids and local politics.

Politics is the newest variable. With Democrats sweeping Congress and voters favoring local data center restrictions, construction will likely be pushed back, Lima said, as project timing shifts. Data center development already depends on securing power and permits; a hostile or simply cautious political environment adds delay and cost.

Lima believes the hyperscalers will manage this post-election world better than their smaller rivals.

"Even with risks emanating for future capacity build-out post elections, we are of the view that HQ hyperscalers will remain well capitalized to pursue their capex ambitions, and are better prepared to navigate further uncertainty to secure power and start new data center construction," she said.

She also flagged a funding gap worth watching. Lima said she wouldn't be surprised to see more transactions announced — like recent equity issuances — to help bridge the distance between investments and monetization. In plain terms: the money coming in from AI products still lags the money going out to build them, and companies will keep tapping whichever market is cheapest to fill the difference.

Is this the trade of the year in AI credit?

Lima's bottom line is unusually direct for a strategist note. "All-in-all, with hyperscalers trading not far from YTD wides, and at levels consistent with A/BBB ratings, risk-reward is as good as it's been all year," Lima said.

That is a market-timing claim, not just a sector view. Hyperscaler bonds are trading near their widest levels of the year — prices near their lows — while the underlying credits still sit at A/BBB-caliber ratings. For bond investors, wide spreads on strong balance sheets are the classic setup for attractive returns if nothing breaks.

The catch is what "nothing breaks" requires. Execution risk sits with the data center builders and neoclouds, not the hyperscalers, in Morgan Stanley's framing — but a prolonged monetization gap, higher rates or political delays to construction would eventually reach even the strongest credits. The $1.77 trillion projected for 2028 assumes the money keeps moving.

For now, Morgan Stanley's position is clear: own the debt of the companies that can absorb higher borrowing costs, avoid the leveraged tail of the AI trade, and treat CoreWeave's 16% yield as a warning about the second tier rather than a verdict on the whole build-out.