NextFin News - Microsoft's latest earnings did not pop the AI bubble. They showed how to keep it inflated - and how the biggest technology companies are quietly refinancing it onto the balance sheets of bond investors.
The software giant emerged from its fiscal fourth-quarter results in late July as a rare AI winner, one of the few hyperscalers that has managed to turn a wall of artificial-intelligence spending into accelerating revenue growth. That demonstration, combined with a wave of equity and debt offerings from the megacap cohort and a pullback in Treasury yields from their late-September peaks, has restored investor confidence in the very stocks most associated with the AI trade. The implication is uncomfortable for bears and bulls alike: the bubble is more likely to grow bigger before it pops than to deflate quietly.
Here is the tension the market has to resolve. Microsoft's shares gained 37.5% from July through September, their best quarter since 1998, adding roughly $1 trillion in market value. Yet the company is simultaneously cutting the headline capital-expenditure number through an accounting change, while Oracle prepares to raise up to $50 billion and the broader group leans on a bond market that is still pricing the 10-year Treasury yield above 5%. The AI boom is no longer running on optimism alone. It is running on contracted backlogs, depreciation schedules, and the willingness of fixed-income investors to underwrite it.
The Monetization That Changed the Story
The first question is whether the AI spending is producing anything. For much of 2025 and early 2026, the answer from Wall Street was a qualified maybe: revenue was growing, but not fast enough to justify capital budgets that were doubling. Microsoft's fiscal fourth quarter flipped that framing. Revenue came in at $90 billion, up 18% year over year, while Azure and other cloud services grew 43%, accelerating from 40% in the prior quarter and beating the 39% to 40% guidance the company had issued. Annual Azure revenue surpassed $100 billion for the first time. Azure-only revenue grew 41%.
More important than the growth rate is what sits behind it. Commercial remaining performance obligation - contracted cloud revenue not yet recognized - rose to $678 billion, up 8% from the previous quarter, driven by commitments from clients other than AI model developers. Paid seats for Microsoft 365 Copilot crossed 30 million, more than doubling the net addition from the prior quarter, while GitHub Copilot revenue accelerated more than 60% quarter over quarter after moving to consumption-based pricing. Microsoft's fiscal 2026 revenue from commercial arrangements with OpenAI was reported at $24.1 billion, and OpenAI owed the company about $6 billion as of June 30. The partnership is no longer a speculative bet; it is a disclosed revenue stream large enough to move a $90 billion quarter.
The OpenAI relationship remains the anchor of the longer-dated backlog. Microsoft said OpenAI has committed to an incremental $250 billion of Azure services, with revenue-sharing, exclusive intellectual-property rights, and API exclusivity running until artificial general intelligence is reached or through 2030, whichever comes first. Model and product IP rights extend to 2032. Chief executive Satya Nadella called the expanded agreement "the next chapter in what is one of the most successful partnerships and investments our industry has ever seen."
"We are building a planet-scale cloud and AI factory, maximizing tokens per dollar per watt," Nadella said on the earnings call, framing the spending as an efficiency race rather than a blind land grab.
The efficiency framing is the bridge between capex and margins. In the fourth quarter, roughly two-thirds of Microsoft's capital spending went to short-lived assets centered on CPUs and GPUs, with the remainder directed to long-lived assets such as data-center land and buildings. That distinction is the hinge on which the entire AI-investment debate turns. Chips depreciate fast and can be redeployed; a data-center shell is a 25-year bet on demand that may not exist in year ten. The market rewarded the message: shares climbed about 9% in after-hours trading on July 29, the strongest reaction after an earnings report in months, because revenue growth was finally catching up with investment.
The Accounting Change That Made Capex Look Smaller
There is, however, a second story inside the numbers, and it is less flattering. Starting in fiscal 2027, Microsoft is extending the estimated useful life of its data centers and office buildings from 15 years to 25 years. The change, announced by chief financial officer Amy Hood, reclassifies a portion of the company's data-center leases from finance leases - which appear on the balance sheet as capital expenditures - to operating leases, which flow through the income statement as operating expenses. The cash leaving the building is identical. Only the line item changes.
The effect is visible in the headline capex figure. Following the earnings release, Microsoft's calendar 2026 investment outlook was lowered from approximately $190 billion to approximately $175 billion - a $15 billion reduction that did not cancel a single construction contract or defer a single server order. The company stated plainly that, excluding the effect of the useful-life change, its calendar 2026 investment forecast remains unchanged. Analysts widely expect fiscal 2027 capital expenditure to grow 20% to 30%, landing near $220 billion, even as the reported number benefits from the reclassification. First-quarter fiscal 2027 capex alone is projected to exceed $50 billion.
Hood was careful to describe the change as a timing adjustment, tying it to "our operating history and expected use of these assets," and said it would have only a minimal benefit to fiscal 2027 operating income. Morningstar, asked to assess the move, called it "not controversial." That may be true as a matter of accounting. But it is also a reminder that the AI investment cycle is being presented to investors through a lens that softens the near-term optics while leaving the economic exposure intact.
The change "affects only the timing of future depreciation," Hood said, signaling that the company expects its data centers to remain productive for a decade longer than it previously assumed.
The skeptical read is straightforward: extending a data center's depreciation life from 15 to 25 years is a bet that the assets inside will remain economically useful for a quarter-century. Microsoft depreciates its CPUs and GPUs over six years. A building may last 25 years; the silicon inside it will be replaced at least three times over that span. The accounting change reduces reported capex, but it does not reduce the risk that today's AI hardware becomes tomorrow's stranded asset - and it does nothing to ease the burden of the operating-lease payments that now carry the equipment cost off the capex line.
The $570 Billion Refinancing Machine
If Microsoft is showing how to monetize AI, the rest of the megacap cohort is showing how to finance it. Oracle announced a calendar 2026 plan to raise between $45 billion and $50 billion through a combination of debt and equity sales to build additional cloud infrastructure - potentially the largest high-grade U.S. offering since Meta Platforms raised $30 billion in late 2025. Roughly half of Oracle's funding is expected to come from equity-linked and common equity issuances, including a new at-the-market program of up to $20 billion and mandatory convertible preferred securities carrying a 6.25% to 7.25% dividend.
Oracle is not an outlier. It is the leading edge of a cohort-wide shift from self-funding to external financing. Alphabet, Amazon, Meta, Microsoft, and Oracle together issued approximately $121 billion of bonds in 2025, compared with roughly $40 billion in 2020 - a threefold increase in five years. Morgan Stanley expects global AI-related debt issuance to reach around $570 billion in 2026, with about $236 billion already priced by the end of May, roughly four times the pace seen a year earlier. U.S. technology companies had issued $157 billion of bonds through late September, a 70% increase from the prior year.
This is the mechanism by which the bubble grows rather than pops. When AI spending was funded out of operating cash flow, every dollar of capex was a direct hit to free cash flow, and every missed monetization target showed up immediately in the stock. Now the spending is being intermediated through capital markets. Equity offerings dilute existing holders gradually. Debt pushes the reckoning into the future. Special-purpose vehicles and sale-leaseback structures move assets off the balance sheet entirely. The economics do not change - the data centers still have to earn their keep - but the visibility of the risk does.
And the timing of this refinancing wave is not accidental. It is being underwritten by a bond market that, after a violent late-September selloff, has given back part of its yield advance. The 10-year Treasury yield eased to 5.20% on October 14, 2026, down from an intraday high of 5.34% in late September - the highest level since 2002. In the third quarter, the 10-year note ended at 4.16%, down from 4.26% at the start of July, while the 30-year bond dropped from above 5.00% to 4.73%. For a megacap issuing 20- or 30-year infrastructure debt, a 100-basis-point move in long rates is the difference between a manageable financing cost and a project that never clears the return hurdle.
Second-Order Thinking: Who Actually Bears the AI Risk?
The conventional read of these developments is that megacap AI stocks are becoming safer: revenue is contracted, financing is available, and yields are cooperating. The second-order read is different. The risk is not disappearing; it is migrating from equity holders to fixed-income investors, and that migration is what allows the equity multiple to stay elevated.
Consider the chain. First order: Microsoft shows AI monetization, so its stock rises. Second order: the demonstration gives bond investors the confidence to buy Oracle's, Meta's, and Alphabet's AI debt at yields that keep the projects viable. Third order: because the debt market is absorbing the infrastructure cost, equity investors no longer need to see immediate free-cash-flow payback to hold the stocks - they only need to believe the contracted revenue will arrive before the debt matures. The bubble is no longer sustained by belief in AGI. It is sustained by the term structure of corporate credit.
This is why the "bubble will get bigger before it pops" framing is more than rhetoric. A bubble funded by equity optimism pops when the story breaks. A bubble funded by long-duration debt pops when refinancing breaks - and that event can be delayed for years by falling rates, covenant-lite structures, and the willingness of insurers and pension funds to reach for yield. The megacaps have effectively converted an equity-risk problem into a credit-risk problem, and credit risk is slower to reprice.
There is a precedent for this pattern, and it is not comforting. In the late 1990s, the fiber-optic buildout was financed by a wave of high-yield debt and sale-leaseback structures that kept the equity story alive long after the underlying demand failed to materialize. The telecom bubble did not deflate; it transferred its losses to bondholders and only then collapsed. Today's AI infrastructure financing has the same shape: the assets are real, the demand is projected, and the bill is being passed to investors who are being paid to look the other way.
The Counter-Thesis: The Financing Is Getting More Expensive, and the Bubble May Be Smaller Than It Looks
The strongest argument against the "bigger bubble" view comes in two parts, and both are backed by mainstream institutions. First, the debt market is not as accommodating as the headline issuance numbers suggest. Bond investors are demanding higher yields on AI-linked debt than they did even nine months ago. Meta's latest data-center financing, backed by a 20-year lease commitment beginning in 2028, priced at a higher yield than its 2025 record deal - a sign that lenders are starting to price AI fatigue into the cost of capital. When Alphabet's Google issued an aggressive spending forecast in recent months, it triggered a selloff across tech stocks and pushed down prices of AI-adjacent bonds from Microsoft and Amazon.
Second, the bubble may be materially smaller than the panic implies. Goldman Sachs Research revised its 2026 AI capex estimate higher, from $465 billion to $527 billion, yet noted that AI capital expenditure still accounts for only 0.8% of GDP - nearly half the 1.5% of GDP reached during comparable technology booms over the past 150 years. JPMorgan Asset Management put sell-side 2026 capex estimates for just five U.S. companies at $697 billion, up $173 billion since the start of the year, while arguing that it remains too early to tell whether the spending will deliver a sufficient return on investment. The bears' own numbers, in other words, describe a boom that is large in absolute dollars but small relative to the economy that is absorbing it.
The most direct bearish call comes from Capital Economics. John Higgins, the firm's chief economic adviser for financial markets, wrote in September that "there are plenty of signs that we are now in the late stages of a bubble in AI," with the firm's "best guess" that the bubble will begin to burst in 2027. That timeline gives the megacaps roughly three more quarters of easy refinancing before the bill comes due - which is precisely the window in which the bubble can grow biggest.
The falsifying signal for the "bubble grows bigger" view is specific and observable. If the 10-year Treasury yield holds above 5.5% for a month while spreads on AI-related corporate bonds widen by more than 50 basis points, the cheap-refinancing thesis breaks, and the megacap cohort will be forced to choose between cutting capex and defending the stock price. A second signal would be a downward revision to Microsoft's $250 billion OpenAI commitment or Azure growth guidance falling back below 40%. Either would show that the contracted-revenue backbone of the trade is softer than presented.
What Comes Next: Three Horizons, Three Scenarios
Short term (sentiment and liquidity): The path of least resistance is higher. Earnings season has confirmed that AI revenue is arriving, yields have pulled back from their peaks, and the megacaps have the market's attention. Microsoft is due to report fiscal first-quarter 2027 results in late October, with analysts expecting revenue near $90.7 billion, earnings per share around $4.72, and Azure growth of roughly 45% in constant currency. A clean print keeps the rally intact through year-end.
Medium term (fundamentals): The base case is continued growth with margin pressure. Azure's 45% growth rate is strong, but the capital intensity required to sustain it means free cash flow will lag reported earnings. Microsoft Cloud gross margin in the fourth quarter was 65%, down from 67% two quarters earlier, as the sales mix shifted toward Azure and AI infrastructure investment continued. The accounting change helps reported capex, but cash outflows keep climbing - full-year fiscal 2026 cash capital expenditure rose to roughly $115.9 billion, while operating cash flow grew to about $182.9 billion. Winners will be the companies that can show contracted revenue covering at least two years of committed spend.
Long term (structural): This is where the cyclical and structural forces separate. The cyclical leg - falling yields, easy refinancing, earnings momentum - will revert when rates rise or growth slows. The structural leg - the permanent shift of enterprise computing to cloud AI platforms - will not. Microsoft's $678 billion backlog and the OpenAI exclusivity through 2030 are structural assets regardless of where the 10-year yield trades. The bubble, in other words, has a cyclical price and a structural floor.
The scenarios split on financing conditions. In the base case, yields stabilize in the low-5% range, AI debt issuance hits the projected $570 billion for 2026, and megacap stocks grind higher on contracted revenue visibility. In the upside case, a faster-than-expected decline in long rates unlocks another leg of multiple expansion, and the AI infrastructure buildout is refinanced at costs that make even marginal projects viable. In the downside case, long yields break above 5.5%, AI bond spreads widen, and the cohort is forced to cut capex - at which point the suppliers, led by Nvidia, feel the pain before the hyperscalers do.
For now, the tape reflects the base case with a bullish lean. Microsoft trades near $535, up from a 52-week low of $349, with a market capitalization of about $3.97 trillion. Nvidia, whose fate is tied to the hyperscalers' appetite, trades above $220 with a market value of roughly $5.3 trillion. But the session was not kind to the whole cohort: Oracle fell about 9%, Meta dropped 3.6%, Amazon slid 2%, and Alphabet lost 1.7% - a reminder that the financing trade is idiosyncratic, and that every new equity offering is a test of how much dilution investors will absorb.
The megacap AI trade has entered its most dangerous phase, which is also its most profitable. It is no longer a story about whether artificial intelligence works. It is a story about how long the credit market will fund the buildout before demanding proof that the returns exist. Microsoft's earnings suggest the answer, for now, is: longer than the bears want to admit.
The bubble is not being disproven. It is being underwritten - and a bubble with a willing creditor is the hardest kind to bet against.
Explore more exclusive insights at nextfin.ai.
