AI-related capital expenditures are projected to reach $740 billion in 2026, with a year-on-year growth of approximately 70%. This surge is consuming the operating cash flows of major U.S. tech companies, and except for Microsoft, other companies may see their free cash flow turn negative. AI-related debt accounts for about 14% of the U.S. investment-grade bond market, with funds shifting from equity premium to bond market absorption. As pressures on cash flow and financing intensify, the market’s tolerance for errors is diminishing. Any slowdown in the pace of return realization could lead to risks spreading to broader asset classes.
When the AI-related capital expenditures of Silicon Valley giants swell to nearly the scale of their annual cash flow, the market's concern is no longer about 'whether it is worthwhile' but rather 'whether they can sustain it.'
According to the latest publicly available data, $Alphabet-A (GOOGL.US)$ 、 $Amazon (AMZN.US)$ 、 $Microsoft (MSFT.US)$ 、 $Meta Platforms (META.US)$ the combined capital expenditure guidance for four hyperscale cloud vendors in 2026 is approximately USD 650 billion.
If additional factors are included, $Oracle (ORCL.US)$ followed by $CoreWeave (CRWV.US)$ the total scale rises to USD 740 billion.
These figures are not only higher than market expectations but also deviate by multiples. What does $740 billion mean?
$740 billion represents an approximate 70% year-over-year increase compared to 2025.
It is double the market consensus forecast at the end of 2025 (approximately 35% growth in capital expenditure).
$740 billion is close to the total annual operating cash flow of the entire hyperscale cloud vendor ecosystem.
More alarmingly, Goldman Sachs analyst Shreeti Kapa pointed out that if this level is reached, the intensity of such spending will approach the peak of the dot-com bubble in the 1990s, which accounted for 1.4% of GDP. Although still below the intensity seen during the Industrial Revolution, it is rare in modern technological history.
The well-known financial blog ZeroHedge analyzed and wrote:
"These numbers are so large that we immediately joked that after using all free cash flow to cover capital expenditures, the Magnificent Seven will simply be unable to afford any stock buybacks in 2026 (or even longer)."
What truly shook the market was not that a single company 'spent more money,' but rather that the capital expenditures of the entire hyperscale cloud vendor ecosystem spiraled out of control simultaneously. This is not a routine increase in capital expenditure, but rather a structural leap.
Cash flow is being 'devoured' by AI.
There isn't enough money left for stock buybacks—what once seemed like a joke is now becoming a reality.
According to Goldman Sachs estimates, if capital expenditures reach US$700 billion by 2026, this figure will be nearly equivalent to the total operating cash flow of hyperscale cloud vendors.
Bank of America concluded in a more detailed model:
Microsoft's solitude: By 2026, Microsoft is projected to be the only company whose operating cash flow can still cover its capital expenditures.
Meta’s turning point: Meta has hinted that it may shift from being 'net debt neutral' to having 'positive net debt' at some point.
The rest: Even if share repurchases are completely halted, free cash flow will be depleted.
Bank of America wrote in its report:
“Excluding Microsoft, even if stock repurchases are suspended or slowed down in the fourth quarter, the remaining companies’ cash flow surpluses will significantly diminish.”
This implies that if capital expenditures continue to rise, cash balances will decline rapidly, making debt financing unavoidable. And this will become a major issue.
AI evolves into a debt bubble: related debt accounts for approximately 14% of the U.S. investment-grade bond market.
As internal cash flows are insufficient to cover expenditures, tech giants have been forced to enter the bond market on a large scale.
Several months ago, ZeroHedge warned: 'AI is now also a debt bubble, quietly surpassing all banks to become the largest sector in the market.'
Bloomberg wrote in its latest edition of Credit Weekly:
Large technology companies are preparing to spend far more on artificial intelligence than investors had previously anticipated, and regardless of outcomes, fund managers are increasingly concerned about the potential impact on credit markets.
In the week leading up to February 11, 2026, the market witnessed an unprecedented frenzy:
Oracle: issued a record-breaking $25 billion bond. Despite its stock price plummeting due to negative cash flow and soaring default risks, the bond issuance attracted $129 billion in subscription orders.
Google: One week after Oracle’s issuance, Google followed suit by completing a $20 billion dollar-denominated bond issuance (initially planned at $15 billion). This marked its largest bond issuance ever, with subscription orders exceeding $100 billion. Google even plans to issue a rare 100-year bond—a first attempt by a tech company since the internet bubble of the 1990s.
Why issue so much debt? Because the money from advertising and cloud services alone isn’t enough.
Estimates suggest that global data centers will require approximately $2.9 trillion in capital expenditure before 2028 (a figure that continues to rise). However, companies' internal operating cash flows can only cover half of this capital expenditure.
How will the remaining $1.5 trillion gap be filled? There is only one answer: debt.
This includes corporate bonds, asset-backed securities (ABS/CMBS), private credit, and even sovereign debt.
By the end of 2025, AI-related investment-grade debt will account for 14% of the U.S. IG market, becoming the largest single thematic segment, surpassing even the banking sector in scale.
Morgan Stanley forecasts that investment-grade bond issuance in the technology, media, and telecommunications sectors could reach $2.25 trillion in 2026, setting a new historical record.
Cracks have begun to appear in the bond market.
Although demand remains strong at present, cracks have started to emerge.
Bloomberg data shows that last week, the spread on U.S. investment-grade corporate bonds widened by approximately 2 basis points. Oracle's newly issued $25 billion bond underperformed significantly against Treasuries in the secondary market. Moreover, when Oracle announced its stock offering to raise funds, market anxiety spiked, causing its stock price to plummet.
Alexander Morris, CEO of F/m Investments, stated:
“The current investment boom in artificial intelligence has indeed attracted many buyers, but the upside potential is limited, and the margin for error is minimal. No asset class is immune to depreciation.”
The current equilibrium is extremely fragile. The market is in a state of 'autopilot,' with the bond market remaining open as long as the AI narrative continues to hold. However, any shock similar to the 'DeepSeek Moment' anticipated in January 2025, or disruptions caused by technological iteration weakening the moats of industry giants, could lead to an abrupt closure of the bond market.
Chain Reaction Between the Software Industry and Private Credit
AI is not only depleting the cash flows of tech giants but also undermining the valuation logic of the traditional software industry, planting the largest potential risk in the credit market.
Bloomberg noted that as AI tools increasingly penetrate professional service sectors, investors have begun reassessing the growth prospects of the entire software industry.
AI Efficiency Gains = Decline in Software Demand: With companies like Anthropic introducing AI tools tailored for professional services, investors are growing concerned that AI may render many SaaS (Software as a Service) products obsolete. If AI can write code and generate reports, why would businesses continue purchasing expensive software licenses?
Selloff in Software Company Bonds: Leveraged loan prices for software companies have fallen by approximately 4% this year.
Private Credit Under Threat: This represents the most perilous link in the chain.
According to Barclays' analysis, the software sector constitutes the largest risk exposure for Business Development Companies (BDCs, or publicly traded private credit funds), accounting for roughly 20% of their investment portfolios. The bank highlighted in its report:
"Software is the largest industry exposure within BDC portfolios, representing approximately 20%, making it particularly sensitive to recent declines in software equity and credit valuations."
While AI giants burn through cash to build infrastructure, they are inadvertently creating a technology that could kill their downstream clients—software companies. Should these companies default due to their products being replaced by AI, the private credit market holding substantial amounts of their debt will collapse first, triggering a domino effect.
Prisoner's Dilemma: Knowing it’s a bubble, why invest anyway?
In the face of Goldman Sachs’ critique of 'too much investment for too little return,' why do the CEOs of Google, Microsoft, and Amazon still choose to 'move full steam ahead'?
The answer lies in the concept of 'Nash Equilibrium' from game theory.
For these giants, this represents a classic binary strategic choice:
If you don’t invest: You face permanent loss of market share. AI infrastructure operates under a 'winner-takes-all' dynamic. If you fall behind now, you will never catch up. Just as IBM missed out on cloud computing, the result is strategic obsolescence.
If you over-invest: Financial statements suffer, profit margins shrink, and returns are delayed due to overcapacity. But at least you’re still in the game, still at the table.
Prisoner's Dilemma: If your competitor invests and you don’t, you lose customers; if you invest and your competitor doesn’t, you win the market. Thus, the rational strategy is always to invest.
As Goldman Sachs analyzed, this dynamic creates a Nash Equilibrium, where even if short-term returns are squeezed, continuous capital expenditure remains rational at the individual level.
This is why, even when facing the risk of transitioning from 'net cash' to 'net debt,' and even with the burden of hundreds of billions of dollars in debt, these giants will not slow down. For them, a decline in market capitalization (due to worsening financial conditions) is tolerable, but disappearing from the scene (due to falling behind technologically) is unacceptable.
Outcome Analysis: Trillions in profits or irreversible ruin?
The ultimate outcome of all this hinges on a central question: Return on Investment (ROI).
Goldman Sachs analyst Shreeti Kapa has run the numbers:
Over the past decade, large technology giants have typically generated profits that are 2 to 3 times their capital expenditures. Given projected average annual capital expenditures of $500 billion to $600 billion for 2025-2027, these companies would need to achieve an annual profit run rate exceeding $1 trillion to maintain the returns investors have become accustomed to.
However, current market consensus estimates for 2026 profits stand at only $450 billion.
This represents a significant gap. Even the most optimistic strategists would struggle to explain how monthly subscription fees of $30 and occasional enterprise contracts could double the profits of these giants in the short term.
Goldman Sachs has outlined two possible scenarios:
Bull Case (Cloud 2.0): The adoption trajectory of AI follows that of cloud computing. Amazon AWS achieved breakeven within three years and reached a 30% operating margin within a decade. If AI can replicate this path, the massive investments being made today will yield extraordinary returns. The current $1.5 trillion backlog in cloud orders supports this narrative.
Bear Case (Global Crossing Redux): History shows that pioneers of major technological innovations often fail (e.g., Global Crossing during the fiber-optic era). While today's tech giants are more financially robust, the scale of current spending and intensifying competition suggest that not all of them will generate sufficient long-term profits to reward today’s investors.
Before this high-stakes gamble is resolved, the bond market’s “vigilantes” may wake up first. If they decide to stop funding this feast, this round of AI-driven prosperity, fueled by debt, could come to an abrupt and dramatic end.
Editor/KOKO
