The pace of spending in the AI arms race is now surpassing the ability of tech giants to generate their own profits. Over the past week, five hyperscale technology companies—Amazon, Google, Meta, Microsoft, and Oracle, with a combined market capitalization exceeding $12 trillion—released their first-quarter earnings reports. Without exception, each raised their capital expenditure plans. In response, Morgan Stanley's chief cross-asset strategist Andrew Sheets updated his forecast, projecting the combined capital expenditure of these five giants to reach $800 billion by 2026, and further climbing to over $1.1 trillion in 2027. Sheets wrote:

We forecast hyperscale tech capital expenditure of approximately $800 billion in 2026, which is nearly double the 2025 spend and triple the 2024 level. Next year, my colleagues estimate that U.S. hyperscale capex could reach $1.1 trillion.

While the numbers are staggering, they raise a critical question: where will the money come from? The answer is debt. For many years, these tech giants accumulated substantial free cash flow through their "asset-light" business models. However, the situation has now quietly reversed. Free cash flow data from Amazon and Meta shows that both companies have approached or fallen into negative territory.

What does this mean? Simply put: their own funds are insufficient, forcing them to borrow. This is especially true as these companies must also maintain stock buybacks and dividend payments. Consequently, nearly all new capital expenditure must be supported by issuing debt. Morgan Stanley anticipates that 2026 will be the busiest year on record for the U.S. investment-grade (IG) bond market:

Total issuance is projected to be approximately $2.25 trillion, a 25% year-over-year increase. Net supply is expected to be around $1 trillion, a 57% year-over-year increase.

Notably, the tech sector has already contributed 18% of the U.S. investment-grade bond supply this year—the highest proportion the industry has ever recorded, and double the level from the same period in 2025.

Morgan Stanley summarizes the core driver behind this trend in four words: "AI Capex Driving Supply." Debt Market Shows Signs of Strain The market is taking notice. Following a $300 billion surge in AI-related debt issuance, investors are beginning to show fatigue. A clear example: Meta issued up to $25 billion in investment-grade bonds last week, with peak order books reaching around $96 billion. While this figure seems substantial, it represents a noticeable contraction compared to the $125 billion in demand attracted by a $30 billion bond from the same issuer last October. More telling details have emerged:

An issuer linked to SoftBank Group was forced to increase the bond's yield to complete the financing due to insufficient demand. Investors are beginning to demand stronger protective covenants—including a "backstop" from Alphabet, Google's parent company, guaranteeing data center rental payments even if a tenant defaults. Some investors have outright rejected certain deals. One investor told Bloomberg they passed on a $14 billion bond for an Oracle data center in Michigan, partly because the bond contained call provisions unfavorable to creditors.

Robert Tipp, Global Head of Bonds at PGIM Fixed Income, stated:

Ultimately, these companies are selling a lot of debt, and they are going to have to pay up to borrow. The market is facing a wall of worry after corporate spreads tightened dramatically to historic lows.

John Servidea, Global Co-Head of Investment Grade Debt Capital Markets at J.P. Morgan, added:

We are seeing what different investors value in these financings, how they assess risk and return. We've seen these deals get done with pretty healthy demand, but as supply increases, we expect terms and structures to continue to evolve.

Banks Are "Reaching a Breaking Point" The fatigue in the bond market is just the tip of the iceberg. Deeper pressures are building within the banking system. According to a May 3rd report by the Financial Times, major lenders including JPMorgan Chase, Morgan Stanley, and Sumitomo Mitsui Banking Corporation (SMBC) are actively seeking ways to offload data center-related debt risks to a broader range of investors to free up space on their balance sheets. Matthew Moniot, Co-Head of Credit Risk Sharing at Man Group, stated bluntly:

The scale we're talking about... is far beyond anything we've ever contemplated. The banks will be overwhelmed very quickly.

A specific case highlights the severity of the problem: Banks including JPMorgan and Mitsubishi UFJ Financial Group (MUFG) spent over six months trying to distribute $38 billion in construction debt related to Oracle's data center projects in Texas and Wisconsin. The result—insufficient demand forced some banks to sell the loans at a discount to non-bank lenders. $38 billion, for a single project, unsold after six months. This situation stems from hard constraints on internal risk limits at banks—once exposure to a single borrower or industry reaches its cap, banks cannot finance new projects. Moniot said:

If I were a chief risk officer at a bank, and bankers were coming to me asking for multi-billion dollar credit approvals for individual projects, I would ask them how they plan to distribute that risk.

To address this, banks are exploring "Significant Risk Transfer" (SRT) instruments—slicing up highly concentrated single data center loans and selling the riskiest portions off their books to investors like private credit funds and insurance companies. David Lucking of law firm Linklaters noted:

Banks will typically still retain a portion of the exposure. SRT investors want to know the bank still has some skin in the game.

Frank Benhamou, Portfolio Manager at Cheyne Capital, pointed out that data center SRTs differ fundamentally from traditional products:

There are a limited number of operators, extreme concentration, and significant construction risk. You naturally demand a higher return for that.

Goldman Sachs Warns: IG Bond Market Is "Equitizing" This wave of AI debt is also altering the structure of the entire investment-grade bond market. Amanda Lynam, a Goldman Sachs IG bond strategist, noted that U.S. investment-grade issuance since the start of 2026 has had its strongest opening on record—reaching $794 billion by April 20th, which, when annualized, aligns closely with Morgan Stanley's full-year forecast of $2.25 trillion. However, the structural change is more noteworthy. Jeffrey Papai, a Goldman Sachs IG bond trader, wrote in a recent report that among the 660 issuers who sold investment-grade bonds over the past year, just 11 accounted for roughly 25% of the duration-adjusted supply. Among them, four hyperscale tech companies (Meta, Amazon, Oracle, Google) plus four large data center financings collectively accounted for nearly 20% of the total duration-weighted issuance. A comparison illustrates how extreme this is: Oracle is now the largest single issuer in the investment-grade index on a risk-adjusted basis; Meta has jumped from the 51st largest issuer in the IG index to the 8th largest in less than a year. Last week's $25 billion Meta bond offering, or the largest single data center financing transaction, had a duration-weighted size nearly equivalent to the entire outstanding bond volume of Boeing, and even exceeded the total outstanding bonds of Ford or General Motors.

Consequently, Goldman Sachs issued a warning:

"We are now dealing with a market increasingly concentrated in AI build-out, similar to the equity market, but in a more negatively convex way—because fixed income has no upside."

In other words: equity bets on AI can yield profits if stocks rise; bond bets on AI earn interest, but if problems arise, the losses are real. Seeking Funding Globally as Domestic Capacity Falters Facing capacity limits in the U.S. investment-grade bond market—where a single issuer's share typically does not exceed 2-3%—tech giants have begun seeking financing in global markets. Goldman Sachs data shows a significant increase in issuance of euro, pound, and Swiss franc-denominated bonds by hyperscale companies since 2024. As this article was finalized, Alphabet, Google's parent company, launched a euro-denominated bond sale of at least €9 billion, alongside a Canadian dollar bond offering—both setting new records in their respective markets.

Meta has taken a different path: establishing off-balance-sheet Special Purpose Vehicles to distribute debt pressure. Following last year's $27 billion "Project Beignet" with Blue Owl to finance a Louisiana data center, Meta is now collaborating with Morgan Stanley and J.P. Morgan on a $13 billion "Project Sopaipilla" to fund its data center in El Paso, Texas. The essence of this structure is to disperse debt among as many parties as possible. Morgan Stanley: When Will the AI Bubble Burst? Watch for Four Signals As the entire AI super-cycle becomes increasingly reliant on the smooth functioning of the debt market, Morgan Stanley listed four potential warning signals that could trigger a spike in credit spreads and cause the AI "house of cards" to collapse:

Debt growth outpacing earnings growth. The leveraged finance market growing faster than the high-quality credit market. Mergers and acquisitions activity exceeding long-term trend levels. An acceleration in private equity-backed deals coupled with a decline in equity contribution.

Another直观 market signal is worth noting: on the same day U.S. stocks closed at a record high and several "Magnificent Seven" tech stocks surged, the credit default swap spreads for Meta hit a record high and have been widening daily. Record-high stock prices alongside record-high CDS spreads—the simultaneous occurrence of these two events is itself a signal worth pondering.

Morgan Stanley ultimately offered a concise but weighty conclusion:

The credit market is financing the AI build-out.

The implication is clear: once the credit market closes its doors, the AI super-cycle will come to an end.