Author: Rita

Trendy Guide

This year, global AI-related debt issuance reached 336 billion dollars in the first seven months. The total for the year is expected to be 580 billion, more than doubling from last year.

Morgan Stanley visited a round of investors and found that money is still being poured into AI data centers. However, the approach has changed; previously, investments were made blindly, but now everyone is asking three questions: Will there be overbuilding? Will technology become obsolete? Can the supply chain keep up?

The credit spreads of high-quality, large-scale cloud providers have widened by 24 basis points this year. In contrast, the overall investment-grade market has actually narrowed by 2 basis points. The gap is clear. The market's message is very straightforward: AI bonds can be purchased, but they must be selected carefully; the era of blindly rushing in is over.

After 336 billion, the money is still pouring in

According to Morgan Stanley's statistics, global AI-related debt issuance reached 336 billion dollars in the first seven months of this year, with an annual forecast of 580 billion. The investment-grade market accounts for the bulk. The largest cloud providers, specifically Google, Amazon, Meta, Microsoft, and Oracle, contributed 132 billion. The remaining 90 billion came from semiconductor and data center construction.

The five major companies are expected to increase capital expenditure by another 54% by 2027, rising from 126.7 billion to 141.4 billion. The issuance volume in the second half of the year will not slow down; capital expenditure is accelerating, and financing demand will only grow larger with no respite visible on the supply side.

The credit spread speaks, choose quality, do not buy blindly

This year, the overall credit spread in the investment-grade market narrowed by 2 basis points. High-quality, large-scale cloud providers, on the other hand, widened by 24 basis points, completely diverging. Oracle was downgraded by S&P in June, and the curve for 10-year to 30-year bonds has further steepened. Morgan Stanley maintains its underweight judgment on the tech sector, as supply continues to be under pressure, and credit quality has begun to differentiate.

What do investors prefer? Short-term bonds with amortization protection come first. Hut 8's bonds have construction guarantees and full amortization, while the developer's credit history is mediocre, but the structural protection is solid. QTS's Microsoft lease bonds, however, have top-tier tenant quality but lack an amortization mechanism. Moody's estimates that 50% to 65% of the principal will still be unpaid at maturity. Construction risks can be hedged by structural enhancements, but asset risks are trickier; even if the tenant is Microsoft, it cannot cover all risks.

Financing channels are diversifying, not just borrowing in dollars anymore

The financing mix of large-scale cloud providers has changed. Previously reliant on dollar-denominated debt, they now utilize multiple currencies along with equity and loans, adopting a multi-faceted approach. Google derives 61% of its debt financing from non-dollar currencies, including euros, pounds, Canadian dollars, Swiss francs, and yen. Amazon issued 17.5 billion dollars in delayed draw term loans. Oracle and Google have also announced equity financing plans.

Morgan Stanley determines that equity financing signifies a message, while debt financing channels remain in place. The investment cycle is both large and lengthy, so companies must maintain financing flexibility. The five major companies are to increase capital expenditure by 54% by 2027, while Broadcom has a large private placement of 35 billion dollars. In the second half of the year, supply pressure will only increase, which is unavoidable.

The high-yield market, construction risk is the focus, and there will be delivery pressure in the second half

In the high-yield market and the investment-grade market, what investors are concerned about differs. Construction risks, refinancing paths, and whether they can upgrade ratings post-completion are their primary concerns. Morgan Stanley predicts that high-yield data center debt issuance will reach 50 billion, and leveraged loans will be around 15 billion, a number that investors broadly accept.

The structure is also changing. Loan cost ratios are climbing, amortization is starting later than before, and the proportions are lower. Termination rights have been extended from 6 months to over a year, loosening cash flow waterfalls considerably. The cost per megawatt is increasing, and timelines for delivery are stretching longer.

A key number for the second half of the year is that over 740 megawatts of critical IT capacity is expected to come online. Reports from FTI Consulting and industry feedback suggest that the likelihood of construction delays is high. The timeframe has been compressed to 12 to 18 months, the power systems are complex, cooling interfaces are risky, tenants frequently change designs, supply chains are tight, and labor is still lacking. 35% to 45% of equipment purchases are not under the control of construction partners, leaving developers with a significant amount of execution risk.

Among 13 high-yield data center debts, the first redemption dates are all after the estimated final delivery dates. The ability to refinance and secure lower-cost funds hinges entirely on one question: Can they be completed on time?

Trendy Perspective

The special aspect of this report is that it jumps over the hype surrounding AI itself and directly points to the stratification occurring in the AI debt market. The credit spreads of large-scale cloud providers are widening, and the entire investment-grade market has not followed suit. This suggests that investors do not need to sell off other sectors to make room for AI; they are directly selecting within the AI sector.

The comparison between QTS and Hut 8 is particularly interesting. Construction guarantees and full amortization can fill in gaps in the developers' credit quality. However, for assets without amortization, even with Microsoft as a tenant, refinancing risks still hang overhead. When the lease expires in 2028, what can be done with that data center? This is the real question that needs to be answered. Chip generations change every two years, and the thresholds for liquid cooling and power density keep rising. Asset risks will shift from theoretical deduction to the real exam paper presented to investors.

Disclaimer

This article is a整理 and interpretation of Morgan Stanley's research report dated July 13, 2026 by Trend Research. The ratings, target prices, profit forecasts, and related judgments cited in the text are solely the views of the analysts at that brokerage and do not represent the views of Trend Research or constitute any investment advice.

The market carries risks; decisions must be made independently. This article should not be considered as a basis for buying or selling any securities.