Morgan Stanley Research Report Analysis: AI Bond Market Surges, But Investors Are Becoming Selective on Quality
TechFlow Selected TechFlow Selected
Morgan Stanley Research Report Analysis: AI Bond Market Surges, But Investors Are Becoming Selective on Quality
AI bonds can be bought, but you need to be selective; the era of blindly rushing in is over.
Navigating Web3 tides with focused insightsAuthor: Rita
TechFlow Guide
Global AI-related debt issuance this year reached $336 billion in the first 7 months alone. The full-year forecast is $580 billion, more than double last year.
Morgan Stanley visited a range of investors and found money is still pouring into AI data centers. But the approach has changed. Previously it was blind investment; now everyone is asking three questions: Will there be overbuilding? Will technology become obsolete? Can the supply chain keep up?
Credit spreads for high-quality hyperscale cloud providers have widened by 24 basis points this year. The entire investment-grade market, however, narrowed by 2 basis points. The gap is evident. The market's message is clear: AI bonds are buyable, but must be selected carefully; the era of blind rushing is over.
After $336 Billion, Money Is Still Pouring In
Global AI-related debt issuance statistics from Morgan Stanley show that in the first 7 months of this year, it has already reached $336 billion, with a full-year forecast of $580 billion. The investment-grade market accounts for the majority. Hyperscale cloud providers—Google, Amazon, Meta, Microsoft, plus Oracle—contributed $132 billion. The remaining $90 billion comes from semiconductors and data center construction.
Capital expenditure for the five major companies is set to rise another 54% in 2027, jumping from $126.7 billion to $141.4 billion. Issuance volume will not slow down in the second half; capital expenditure is still accelerating, financing demand will only grow larger, and there is no window for breathing room on the supply side.
Credit Spreads Are Speaking, Pick Quality, Do Not Buy Blindly
Overall credit spreads in the investment-grade market narrowed by 2 basis points this year. For high-quality hyperscale cloud providers, however, they widened by 24 basis points, a complete divergence. Oracle was downgraded by S&P in June, and the 10-year to 30-year curve steepened further. Morgan Stanley maintains its underweight judgment on the technology sector; supply pressure remains high, and credit quality has begun to diverge.
What do investors prefer? Short-term maturities, prioritized with amortization protection. Hut 8's bonds have construction guarantees plus full amortization; the developer's own credit record is average, but the structural protection is solid enough. As for QTS's Microsoft lease bonds, the tenant quality is top-tier, but there is no amortization mechanism. Moody's estimates that 50% to 65% of the principal will remain unpaid at maturity. Construction risk can be hedged through structural enhancement, but asset risk is difficult; even if the tenant is Microsoft, it cannot cover it.
Financing Channels Are Diversifying, Not Just Borrowing USD Anymore
The financing mix of hyperscale cloud providers has changed. Previously it relied entirely on USD debt; now it involves multi-currency plus equity plus loans, walking on several legs. 61% of Google's debt financing comes from non-USD currencies, including Euro, GBP, CAD, CHF, and JPY. Amazon issued $17.5 billion in delayed draw term loans. Oracle and Google have also announced equity financing plans.
Morgan Stanley judges that equity financing releases a signal meaning, while debt financing channels continue to be maintained. The investment cycle is large and long; companies must maintain financing flexibility. Capital expenditure for the five major companies will rise another 54% in 2027, and Broadcom has another $35 billion large-scale private placement. Supply pressure in the second half will only be heavier; it is unavoidable.
High-Yield Market, Construction Risk Is the Focus, Delivery Pressure Remains in Second Half
In the high-yield market and investment-grade market, investors care about different things. Construction risk, refinancing paths, whether ratings can be upgraded after completion—these are what they are watching. Morgan Stanley predicts high-yield data center bond issuance of $50 billion and leveraged loans of $15 billion; investors basically accept these numbers.
Structures are also changing. Loan-to-cost ratios have gone up, amortization starts later than before, and the proportion is also lower. Termination rights have extended from 6 months to over a year, and cash flow waterfalls have loosened considerably. Cost per megawatt is rising, and delivery timelines are stretching longer.
There is a key number for the second half: over 740 MW of critical IT capacity is expected to come online. Reports from FTI Consulting and industry feedback both say the possibility of construction delays is high. Construction periods are compressed to 12 to 18 months, power systems are complex, cooling interface risks are high, tenants change designs frequently, supply chains are bottlenecked, and labor is scarce. 35% to 45% of equipment procurement is not in the hands of construction partners; developers themselves are bearing a large amount of execution risk.
For 13 high-yield data center bonds, the first call dates are all after the estimated final delivery dates. Whether refinancing is possible, whether lower-cost funds can be obtained, all depends on one thing: whether construction can be completed on time.
TechFlow Perspective
The most special thing about this report is that it skips the AI hype itself and points directly to the fact that the AI bond market is stratifying. Credit spreads for hyperscale cloud providers are widening, while the entire investment-grade market has not followed. This indicates that investors do not need to sell other sectors to make room for AI; they are filtering directly within AI.
The comparison between QTS and Hut 8 is particularly interesting. Construction guarantees plus full amortization can make up for the developer's own credit shortcomings. But for assets without amortization, even if Microsoft comes as a tenant, refinancing risk still hangs overhead. When the lease expires in 2028, what can that data center still be used for? This is the question that truly needs to be answered. Chip generations change every two years, and the thresholds for liquid cooling and power density are rising higher and higher. Asset risk will move from theoretical deduction to a real exam paper placed in front of investors.
Disclaimer
This article is a compilation and interpretation by TechFlow Research of Morgan Stanley's research report dated July 13, 2026. Ratings, target prices, earnings forecasts, and related judgments cited in the text are the views of the broker's analysts, represent only their affiliated institution's position, do not represent the views of TechFlow Research, and do not constitute any investment advice.
The market carries risks; decisions must be independent. This article should not be used as a basis for buying or selling any securities.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
@chaoxiangooo
