Comedy or Tragedy? ⭐

Every quarter, I write an analysis piece about Microsoft’s earnings, and the focus has largely been on AI for obvious reasons in recent years. But as the AI infrastructure spending continues rising sharply, I keep waiting for some sign that we’ve finally hit peak spending or that the business will miraculously start making sense because customer spending rises to match.

Neither has happened, and I’m positive the latter is impossible. This reminds me of the 1998 Saturday Night Live skit Change Bank in which representatives of that fictitious bank explain how its business model, which involves making change for large bills without charging fees, makes sense. “Our customers ask us, ‘How do you make money doing this?’,” one representative says. “The answer is simple. Volume.”

This seems to be the accounting principle that Microsoft employs with AI. Which, unlike the SNL skit, isn’t funny at all. The underlying impetus appears to be more about dramatically outspending any potential newcomers to this market to ensure that when the dust settles, Microsoft will remain standing, and dominant, alongside Amazon and Google, with their traditional positions in place. In short, this is a familiar and well-worn strategy for any dominant company, and it’s only the scale that makes it so fascinating.

When I wrote my most recent analysis of Microsoft’s quarterly earnings, the only obvious takeaway given the information Microsoft had provided was that it had found a new accounting trick to obscure the true level of its spending: In addition to paying for hardware components like chips and servers over double the period of time that it had done before, Microsoft is now extending the useful lives of its datacenters and office buildings from 15 to 25 years. And it is shifting that spending away from the capital expenditures (CAPEX) category that’s used to account for AI infrastructure spending and moving it into an operational leases category. As with the previous change, it is doing this to make its spending–which will rise from $40 billion in the reported quarter to $50 billion in the current quarter–look smaller. That is, it will report the $50 billion number, but the real spending is much higher, perhaps $60 billion or more.

Microsoft does this for the same reason any company highlights the positives and downplays the negatives in its financial reports: This is a marketing exercise with the highest imaginable stakes, and its stock price and market capitalization (market cap), and thus its market power, rely on a favorable reception with Wall Street. The analysts who should be protecting our retirement funds and the nation’s economy are more interested in excitement and good news than they are in being responsible. And Microsoft is only too happy to provide the happy noises they expect. In an era of historic overspending with no return on investment in sight, these changes will help in that regard.

Reporting on Microsoft’s business has only become more difficult over the years because of a steady reduction of transparency. To be fair, Microsoft has always used accounting tricks to present the business in the best possible light. It’s just that the shenanigans we see now are bolder and involve so much more money.

Consider the Microsoft earnings report from the same quarter 15 years ago, or Microsoft’s fourth quarter of fiscal 2011, as reported on July 21, 2011. In that quarter, Microsoft earned a net income of $6.17 billion on revenues of $17.37 billion, gains of 4 percent and 8 percent, respectively, year-over-year (YOY). Compare those figures to the net income of $35.8 billion it earned on revenues of $90 billion in the quarter ending June 30, 2026, which were gains of 31 percent and 18 percent.

A few things stand out if you know your history.

First, Steve Ballmer was the CEO of Microsoft at this time, and during his run from 2000 to 2014, Microsoft’s stock price flat-lined at about $30 the entire time. But its market cap–and thus its market power–declined from about $509 billion to $318 billion. Ballmer and Microsoft had lost the interest of Wall Street. This is why he had to go.

Today, Satya Nadella is the CEO of Microsoft, but more importantly, I think, Amy Hood is the CFO, a title she’s held since May 2013. During her tenure overseeing Microsoft’s financials, the company’s stock price rose from $29 to almost $500, with an all-time high of $528 last October. Microsoft’s market cap in May 2013 was $292 billion. Today, it’s about $3.7 trillion. It’s almost 13 times higher.

This chart shows Microsoft’s market cap over time with Ballmer’s beginning and ending dates as CEO.

In that quarter 15 years ago, Microsoft announced that it had sold over 400 million Windows 7 licenses cumulatively since that product’s release in late 2009, almost two years early. And that is one of my clear memories of this period: Microsoft somehow managed to sell almost exactly 20 million Windows licenses every month, like clockwork, because it artificially smoothed out those sales by placing any monthly license sales above 20 million into unearned revenue, an accounting method for deferring earnings. It did this because revenue spikes are always interspersed with downward spikes, and Wall Street does not like that kind of volatility. But in a market in which PC sales were only growing 2 percent overall, Windows license sales held steady. Then, as now, Microsoft was trying to play the game as best it could. The difference is that it plays the game better under Amy Hood. Much better.

Since I wrote that recent analysis piece, Bloomberg published a report detailing what it found in Microsoft’s 10-K report, an annual requirement imposed by the U.S. Securities and Exchanges Commission (SEC). Microsoft’s fiscal year ended June 30, so this report covered the same fiscal year as the quarter wrote about. 10-K reports are meant to include a detailed accounting of a business’s financial performance and business operations, which is precisely what Microsoft wants to avoid in this era of overspending. Given this, the 10-K report is especially interesting now because Microsoft is not at all transparent quarter-to-quarter.

The take-aways are as follows:

• Microsoft reported $24.1 billion in revenues (not profits) from OpenAI during the fiscal year ending June 30.
• Microsoft in March had claimed to be on track to record $37 billion in AI revenues over a one year period (meaning a “run rate” of $37 billion, a figure based on a quarterly high of $9.25 billion just multiplied out by four), but that has no bearing on a fiscal year. It’s just a 12-month period that started one quarter ago because that number was notably good.
• Over 50 percent of Microsoft’s “actual AI sales” (revenues) and most likely “about 70 percent” of those sales came from OpenAI, which highlights how dependent Microsoft remains on that company. Doing the math, this places Microsoft’s total annual AI sales somewhere between $41 billion and $48 billion.
• Microsoft spent $145.3 billion on capital expenditures (CAPEX) related to its AI infrastructure buildout in that fiscal year.
• So Microsoft “lost”–the company would likely say “invested”–about $100 billion on AI in the fiscal year. But that doesn’t tell the full story, of course. The money that Microsoft and OpenAI exchanged was mostly invented out of thin air and never existed to begin with. Microsoft “earns” (reports) revenues from OpenAI for the latter company’s use of Azure infrastructure, but Microsoft also directly invested $13 billion in OpenAI at some point and indirectly invested in OpenAI by spending money over multiple years on infrastructure for which OpenAI is by far the biggest customer. And the two companies have a revenue-sharing agreement that’s not public. In short,
• OpenAI is paying Microsoft with Microsoft’s money. And Microsoft is recording that as revenues. Worse, that’s most of Microsoft’s AI revenue growth, as much as 70 percent of it.

Professional AI hater Ed Zitron is a bit rough for my tastes, but he notes that Microsoft’s total capex spending on AI infrastructure since the beginning of 2022 is over $260 billion, which at least provides a better picture of the true hole Microsoft has dug for itself. $260 billion out in four and a half years, $24.1 billion in from OpenAI this past year, or $40 to $48 billion all told. (And, no, there were no previous OpenAI revenue disclosures. This is the first time Microsoft has ever disclosed revenues from OpenAI. So much for financial transparency.)

“We are now four years into the AI bubble, and Microsoft has little to show for it other than one very large and very unsustainable company that requires near-infinite resources to keep paying its cloud compute bills,” Zitron concludes, adding that analyst estimates have OpenAI and Anthropic making up over 70 percent of all AI revenues across Microsoft, Google, and Amazon.

For the love of God.

Looking just at Microsoft and just at the past fiscal year, when the software giant spent $145.3 billion on AI (but really much more because of all the accounting tricks) and maybe earned $40 to $48 billion in revenues (an average of $44 billion), one might say that, in a best case scenario that we know to be a lie, Microsoft earned $1 for every $3.30 it spent on AI. But we all understand it’s much worse than that.

In some ways, then, Microsoft is nothing like Change Bank from that SNL skit. Like that pretend company, it is relying on volume to make this business make sense. But Change Bank lost money by exchanging the same amounts of money with customers. This real company is losing money, vast sums of money, by exchanging equally vast sums of money with a company that is fraudulent enough to be called a pyramid scheme.

So I am reminded of another revered comedy bit from the 1980s, this one from the movie Spaceballs. In which the villain, Dark Helmet, watching a videocassette of the movie he’s in, sees the future. “When will then be now?” he asks. Which is like asking Microsoft when all the money it’s flushing down the toilet for–sorry, investing in–AI will ever pay off.

The answer, I suspect, is the same. “Soon.”