Three years ago, Microsoft closed its largest-ever acquisition: the videogame producer Activision Blizzard, for $75.4 billion. The deal was meant to turbocharge Microsoft's gaming division, Xbox, into a subscription powerhouse akin to Netflix and Spotify. But last week, the company cut 3,200 Xbox jobs and said it is selling or spinning off four of its game studios. A fifth is under review.
In an internal memo later published by Xbox, Asha Sharma, the executive running the division, plainly told her staff that "our business today is not healthy." She disclosed that, in a typical year, gaming-studio investments lost 64 cents for every dollar put in.
I am a computer scientist who spent nearly seven years working at Microsoft. Most recently, I helped build the AI and speech systems behind Copilot and Teams. I left the company in 2024, and today I'm a Microsoft shareholder. I'm writing this because Xbox's decline is a small preview of the issue every Microsoft shareholder should consider: Will the company's AI infrastructure spending, which is running at multiples the size of the Activision deal every year, earn back its cost of capital? Or will it end up looking like Xbox in a few years?
The Activision story initially looked promising. In fiscal 2024, Microsoft's gaming revenue jumped 39% to $21.5 billion. But that jump was almost entirely the Activision acquisition landing on the books. Strip out the deal, and Xbox's underlying business has been flat or shrinking since 2022.
Gaming revenue fell roughly 7% year over year in Microsoft's fiscal third quarter, which ended in March. Game Pass, the subscription service the Activision deal was supposed to supercharge, was projected in an internal Microsoft presentation (and later disclosed in a 2023 trial over the acquisition brought by the Federal Trade Commission) to reach 77 million subscribers by fiscal 2026. The Wall Street Journal recently reported it has about 30 million.
Sharma's memo also put a number on something Microsoft doesn't normally disclose, which is that Xbox's operating margin is roughly 3%. Do the arithmetic: A 3% margin on fiscal-2025 gaming revenue of $23.5 billion implies roughly $700 million of operating income. Against the deal's $75.4 billion cost, that is a return on invested capital under 1% versus a cost of capital for Microsoft of roughly 10%. That is value destruction at magnitude. (These are back-of-envelope estimates built on Sharma's disclosed margin, since Microsoft doesn't break out Xbox's audited financials. Regardless, the gap is too large for precision to change the conclusion.)
Now on to Microsoft's bigger bet. The instinct it showed in acquiring Activision -- spend big now, trust that it pays off later -- is running through its AI business at a scale that dwarfs the Activision deal into a rounding error.
The most obvious way to see this is in Microsoft's capital spending, including its capital leases, as a share of revenue: about 14% in fiscal 2022, 15% in 2023, 23% in 2024, 30% in 2025 and, per the company's forward guidance, roughly 50% for 2026. In just four years, Microsoft has gone from spending one revenue dollar in seven on infrastructure to roughly one in two. The bet is that the spending itself will create a new tier of growth large enough to justify it. But nobody, including Microsoft, can currently prove it.
Microsoft isn't alone. This spring, Alphabet's Google, Meta Platforms, OpenAI, and Anthropic all announced nearly identical AI strategies. They will embed armies of human engineers inside their enterprise customers' own offices to implement their AI products. When every major AI company converges on the same expensive, labor-intensive approach almost simultaneously, that isn't evidence that this business idea will work. It is four companies making the same bet because none of them have found a cheaper way to make AI deliver measurable value inside their enterprise customers yet.
Microsoft's key vulnerability is that, unlike its competitors, it doesn't own a frontier general-purpose model. Its OpenAI partnership has dwindled also in value. OpenAI now runs on rival clouds and sells directly to enterprises. If value concentrates in a handful of model providers, a risk CEO Satya Nadella has described as a possibility, Microsoft will become a distribution layer atop models it doesn't control.
Microsoft argues that distribution, enterprise trust, and integration depth are the real moat in the AI race, and the fact that even the model owners are racing to build Microsoft-style enterprise sales machines is genuine evidence for that view. Both points may be true, but neither is provable today.
The market has noticed. Microsoft's stock has fallen by double digits over the past two years while its peers rallied.
Warren Buffett might call this a "too hard" problem, one that is too complex to reasonably assess. The AI business isn't necessarily bad; in Microsoft's latest fiscal quarter, it reported substantial growth in its Cloud division, which has a commercial backlog of $627 billion. But no analyst can tell you whether in five years the company's $190 billion in annual AI capex will look like the best capital-allocation decision in its history or like a bad Activision bet on steroids.
Microsoft's layoffs are somewhat encouraging. Cutting a failed bet on Xbox rather than doubling down is judicious fiscal discipline. The question investors should ask now is if Microsoft can make that same call on AI infrastructure after having gone all in.
Guest commentaries like this one are written by authors outside the Barron's newsroom. They reflect the perspective and opinions of the authors. Submit feedback and commentary pitches to ideas@barrons.com .
Naveen Parihar is an independent investor and AI researcher with a PhD in computer engineering. He has worked on speech-recognition and AI systems for 24 years, including at Nuance Communications and Microsoft. He owns 150 shares of Microsoft common stock.
This content was created by Barron's, which is operated by Dow Jones & Co. Barron's is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
July 17, 2026 13:23 ET (17:23 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments