· AI
Microsoft, Amazon, Alphabet, and Meta Spent About 203 Billion on AI Last Quarter. Three of Them Burned Cash to Do It.
This is not a post about whether AI spending is a bubble. It's about what today's AI prices are actually funded by, and why that is not something to plan your business around.
Four companies reported earnings between October 21 and October 29 this year, and together they spent about 203 billion dollars on AI infrastructure in a single quarter. I added it up from their own filings: Meta's 34.91 billion dollars, Microsoft's 39.37 billion, Alphabet's 51.36 billion, and Amazon's 77.26 billion, per their Q3 2026 earnings releases filed with the SEC. That is more than double what the same four companies spent in the same quarter a year ago.
Three of the four paid for part of that spending by burning cash. Amazon's trailing twelve month free cash flow swung from a 14.8 billion dollar inflow a year ago to a 42.4 billion dollar outflow, which the company's own release attributes mainly to its spending on AI infrastructure. Alphabet's quarterly free cash flow, operating cash flow minus capital expenditures by its own definition, came in at negative 9.1 billion dollars, down from a positive number a year earlier. Meta's quarterly free cash flow was negative 2.1 billion dollars, against positive 10.6 billion a year ago. Microsoft is the outlier. By the plain definition, its reported operating cash flow of 58.0 billion dollars still cleared its 39.4 billion dollar capex bill by about 18.6 billion dollars.
Here is the honest case for all of this spending, and it is a real one. Every dollar of it is chasing demand that is already showing up in the numbers, not demand these companies are hoping shows up later. Amazon's own release reports AWS segment sales up 39 percent year over year to 46.0 billion dollars. Microsoft's Azure cloud revenue hit 33.2 billion dollars, up more than 48 percent, and CFO Amy Hood told reporters after the release that "we still have demand that exceeds supply." That is not a company begging investors to be patient. That is a company telling you it would sell more cloud capacity today if it had the servers built. None of this capex is funded by a government grant or a committee's five year plan. It is these companies' own operating cash and their own shareholders' money, placed on a bet about where their customers' workloads are going. A free market take on that is: let them. If the bet is wrong, the market will reprice the stock, which is already happening, not rescue the balance sheet.
That case is real, but it only covers why the spending makes sense to the people doing it. It does not tell you whether the prices you are paying for AI today will still be the prices you are paying for AI next year, and that is the part that matters if you are a business trying to budget, not a hyperscaler trying to win a platform race. When operating cash flow stops covering the build, the gap gets filled by debt, leases, and other people's capital, and that changes who is actually exposed when growth slows. Mary Jander's analysis for Futuriom found Oracle's capital expenditures running at 37 percent of revenue in fiscal 2025 and 66 percent in the first half of fiscal 2026, after it pledged 300 billion dollars in AI infrastructure to OpenAI; her analysis, published in April, noted Oracle's stock had fallen more than 57 percent since that pledge. That is a cautionary tale about financing structure, not proof that Microsoft or Amazon are in trouble. They are not. But it shows the AI buildout is not one uniform story. Some of it is the most profitable companies on earth reinvesting their own cash into real, contracted demand. Some of it is leveraged bets on demand that has not arrived yet. From the outside, both show up in the same capex headline.
None of this tells a small or mid-size business whether an AI bubble will pop. I don't know, and anyone who tells you they do is selling something. What it does tell you is that the frontier AI prices you are paying right now sit on top of a financing structure that three of the four biggest players are currently stretching to maintain, in a single quarter, during what every one of them describes as strong demand. That is not the kind of foundation you build a five year vendor commitment on. It is a reason to keep your workloads portable: route the routine 90 percent of what you run through whichever model is cheapest this quarter, keep a real option to move if a vendor reprices or a smaller model catches up, and treat today's frontier rate card as a snapshot, not a floor.
That is the same discipline I push clients toward through MojoAI: build the cost and quality baseline before you sign the renewal, not after the rate changes on you. If your AI budget assumes today's prices are the permanent price of doing business, this earnings season is a good week to revisit that assumption. Get in touch if you want a second set of eyes on it.
Sources
References used in this article. Links also appear alongside the relevant claims.
Let's make it happen.
You don't need a polished brief. A couple of lines about where you want your company to go is plenty, and we'll come back with what we'd tackle first.
We scope the work around your goals and timeline before quoting anything, so you know exactly what you're getting.