Microsoft shares staged a late-June comeback, clawing back losses from a turbulent week that saw tech stocks battered by renewed interest rate anxiety. The rally put a green finish on what had been shaping up as one of the month’s ugliest stretches for the software giant, but the relief may prove short-lived. Beneath the surface, two enormous forces are tugging at the company’s valuation: a multibillion-dollar artificial intelligence infrastructure buildout measured in gigawatts, and a fast-approaching Windows 10 end-of-support deadline that could shake up a core cash cow.
Investors who listened closely to Microsoft’s recent earnings calls already know the playbook: the company is plowing capital into AI-ready data centers at a speed and scale the industry has never seen. The numbers are staggering. Analysts project fiscal 2025 capital expenditures north of $55 billion, more than double the run rate of just two years ago. Most of that will flow into servers, networking gear, and — most critically — the electricity to power them. Microsoft is now the world’s largest corporate purchaser of renewable energy, with a portfolio of power purchase agreements (PPAs) exceeding 20 gigawatts. It has inked deals to buy output from nuclear plants, signed contracts for offshore wind in Europe, and is pioneering 24/7 clean-energy matching at its data centers. The phrase “gigawatt data center” is no longer hyperbole; it is the operational reality for the campuses under construction outside Chicago, on the banks of the Thames, and in rural Iowa.
That appetite for watts has enormous implications for the company’s margins. Each new AI-optimized data hall consumes as much electricity as a small town, and the cost of that power — even when hedged via long-term PPAs — is rising as tech giants elbow each other out of the queue for grid interconnections. In Northern Virginia, the planet’s busiest data center market, a new facility can wait three years just for a grid connection. Microsoft is fighting for priority access alongside Amazon, Google, and Meta, and the competition is driving up land prices, labor costs, and the price of backup diesel generators. The upshot: the payback period on an AI cluster has stretched out to five years or longer, depending on utilization rates.
Despite those headwinds, the AI narrative has been a powerful propellant for Microsoft’s market capitalization. The exclusive partnership with OpenAI, the deep integration of Copilots into Windows, Office, and GitHub, and the Azure cloud platform’s booming inferencing business have convinced many that the AI revolution will be a Microsoft-led affair. Revenue from Azure AI services grew more than 40% last quarter, and Copilot for Microsoft 365 now counts over 60% of the Fortune 500 as paying customers. The stock trades at a generous 33 times forward earnings, a multiple that assumes years of sustained double-digit growth.
But there’s a catch. As the AI dream requires ever more capital, a more pedestrian line item is creeping into the risk columns of analyst spreadsheets: Windows 10 end of support. On October 14, 2025, Microsoft will stop issuing free security patches for the operating system that still powers roughly 65% of the world’s desktop PCs, according to StatCounter. That is roughly 900 million active machines. Millions of those belong to enterprises and government agencies that cannot simply stop using the software overnight; they must either upgrade to Windows 11 or pay for Extended Security Updates (ESU).
The ESU program, first introduced for Windows 7, allows customers to purchase monthly security patches for up to three years after support ends. For Windows 10, Microsoft has priced the first year at $61 per device, doubling each subsequent year. The sticker shock is real: a large bank with 100,000 endpoints could be looking at a bill exceeding $6 million in year one, climbing to $24 million by year three. The alternative — a fleet-wide migration to Windows 11 — carries its own costs in hardware refreshes, application compatibility testing, and user disruption. Many IT departments are caught between a rock and a hard place.
Investors have historically celebrated the ESU program as a high-margin, low-effort revenue stream that compels customers to pay for what was once free. But the Windows 10 cliff arrives in a different computing landscape than the Windows 7 expiration of 2020. The rise of cloud-native working, the increasing viability of Chromebooks in the enterprise, and the sticky adoption of Apple Silicon in office environments mean that defection is a genuine risk. A corporate license buyer who is already unhappy about the ESU pricing may finally port a line-of-business application to the web and hand employees a Chromebook Plus. Microsoft’s Windows OEM revenue, which still contributes over $10 billion annually, would suffer a direct hit if even a fraction of enterprises take that route.
The interplay between the AI spend and the Windows 10 revenue risk is what makes this moment so delicate for Microsoft. The company is essentially asking shareholders to fund a generational infrastructure bet while simultaneously managing a product-led revenue cliff that could land in the same fiscal year. If AI monetization accelerates on schedule — meaning Copilot seats, Azure OpenAI Service, and the projected $100 billion AI run rate by 2028 materialize — the Windows 10 ESU shock will be a rounding error. But if enterprise AI adoption follows the more typical S-curve of enterprise software, the double whammy of squeezed margins and plateauing legacy revenue could compress the multiple that Wall Street has bestowed.
Analysts at several major banks have begun flagging the tension. In a recent note, Morgan Stanley pointed out that the “AI capex super-cycle” is creating a free-cash-flow trough that Microsoft has not experienced since 2015, when the company was digesting the Nokia write-down and transitioning to the cloud. Back then, the Windows franchise was strong enough to cushion the fall. Today, Windows is a mature business with single-digit growth at best, facing an exogenous shock in the form of the ESU deadline. The buffer is thinner.
There are, of course, bullish counter-arguments. First, Windows 10 ESU revenue could be a significant high-margin tail that offsets near-term AI capex pressure. If half of eligible enterprise devices opt for ESU in year one, that could generate $2–3 billion in incremental revenue at nearly 90% margin, dropping straight to the bottom line. Second, the AI infrastructure being built is already generating return: Azure’s growth rate has re-accelerated from the low 20s to 30%+ partly because enterprises are training and inferring models on Microsoft’s custom Maia accelerators. Third, the massive power deals, while costly, also give Microsoft a competitive moat. A new entrant simply cannot acquire 2 GW of reliable, sustainable power in most geographies without a decade of relationship-building with utilities. Microsoft’s PPAs are as strategic as its patents.
Still, for a company that has prided itself on predictable, compounding growth, the stock’s recent rally masks an unusual degree of execution risk. Shareholders attending the July fiscal-year earnings call will be listening intently for two numbers: the pace of AI revenue growth relative to capital spending, and any quantitative update on the Windows 10 migration pipeline. Microsoft’s CFO Amy Hood has been characteristically disciplined, refusing to provide long-term capex guidance beyond the current fiscal year. But the consensus expectation — $55 billion in FY25, rising to $65 billion in FY26 — is now baked into models. Any deviation from that AI investment trajectory would be seen as a sign that the opportunity is either smaller than advertised or that the returns are not yet coming.
On the Windows 10 side, the company’s messaging has been mixed. The aggressive Copilot+ PC push, built around Qualcomm Snapdragon X and Intel Lunar Lake chips, is designed to make Windows 11 attractive enough to spur voluntary upgrades. But the hardware refresh cycle remains sluggish; IDC reports that PC shipments will grow only 2% this year. If customers decide that their six-year-old Windows 10 machine is “good enough” and just pay the ESU fee, Microsoft gets per-device revenue but misses out on the broader ecosystem lock-in that a new PC creates. The company wants upgrades, not ESUs, because upgrades reinforce Microsoft 365 adoption, OneDrive storage upsells, and the entire Copilot value proposition.
That tension — between short-term ESU cash and long-term platform vitality — is being debated inside Redmond. Some insiders argue that ESU pricing should be aggressive to force migration; others counter that gouging enterprise customers at a time when they are also being pitched expensive AI software could backfire. The final decision appears to have landed on a middle ground: ESU pricing that is high enough to be profitable but low enough that most large accounts will pay it while they plan a move to Windows 11 or Windows 365.
For investors, the next few quarters will clarify which narrative is winning. If cloud growth accelerates and Copilot bookings soar, the AI gigawatts will look prescient and the Windows 10 cliff will be an afterthought. If not, the stock may find that bearish arguments — centered on capex intensity, rising power costs, and licensing headwinds — gain traction. The late-June rally bought time, but time is exactly what the Windows 10 ESU clock doesn’t offer.