The Real AI Story Isn't the Models. It's the $700 Billion Bill.

 I'm David Bellairian, founder and CEO of Glencore Associates. When I wrote here last week about layoffs at profitable companies, the thread running under all of it was Oracle redirecting billions of dollars of payroll into AI data centers. I want to pull on that thread properly, because the actual number behind it is bigger than almost anyone outside the industry realizes, and the honest question about it isn't being asked enough.


The size of the number

Amazon, Alphabet, Microsoft, Meta, and Oracle are on track to spend somewhere between 660 and 700 billion dollars combined on capital expenditure in 2026. Amazon alone is guiding to around 200 billion. Alphabet is in the 175 to 185 billion range. Meta is targeting 115 to 135 billion. Microsoft is tracking toward 120 billion or more. Oracle, the smallest of the five, is still at roughly 50 billion. Depending on which estimate you use, this is a 36 to 62 percent increase over 2025, which was itself already a record year.

Roughly three-quarters of that spending, somewhere around 450 billion dollars, is going directly into AI infrastructure: chips, servers, data centers, and the power and networking to run them. Every one of these companies has said the same thing on their earnings calls. They aren't demand-constrained. They're supply-constrained. They cannot build fast enough to meet the compute requests they're already getting.


How they're actually paying for it

This is the part I don't think gets enough attention. These companies now spend somewhere between 45 and 57 percent of revenue on capital expenditure, a ratio that used to belong to utilities and heavy industry, not software companies. Capex has started to exceed what their own operations generate in cash, which means the hyperscalers, some of the most cash-rich companies in history, are turning to debt markets to finance the rest. Analysts at Morgan Stanley and JPMorgan have estimated the technology sector may need to issue something on the order of 1.5 trillion dollars in new debt over the next few years to keep this buildout going.

That's a real shift. A balance sheet built on decades of cash generation is being levered up to fund a bet on a technology that's still, by most measures, in its early commercial innings.


The math that has to close eventually

Here's the number that made me actually sit up. AI-related services generated something like 25 billion dollars in revenue in 2025, against more than 250 billion dollars of infrastructure spending that same year. That's roughly ten cents of revenue for every dollar of capex. Goldman Sachs has estimated that for this spending to earn a normal return, the industry would need something on the order of a trillion dollars a year in profit, against a consensus estimate closer to 450 billion. That gap doesn't have to close this year or even next year. But it has to close eventually, or a lot of this capital was misallocated.

I want to be careful here, because the bear case on AI capex has been wrong before, repeatedly, at every stage of this cycle. Demand for compute keeps showing up faster than anyone models it. But wrong before isn't the same as wrong forever, and a sector spending more of its revenue on fixed assets than most utilities do is a sector where the downside case, if it ever arrives, arrives at scale.


Where I actually put money in a supply-constrained bet

When an entire industry says it's supply-constrained rather than demand-constrained, the more durable trade usually isn't guessing which model or which chatbot wins. It's the physical bottleneck underneath all of it, power, land, and the specialized real estate that can actually house this equipment. That's closer to how I think about it at Glencore: I'd rather own a piece of the constraint than try to pick the winner competing for it. Real estate is one of five asset classes in the portfolio I run for exactly that reason, it doesn't require betting correctly on which AI company's product wins the year.

None of this means the buildout is wrong. It might be exactly right. But 700 billion dollars a year, increasingly financed with debt, against 25 billion dollars of revenue, is the kind of gap that deserves more attention than it's getting in between the product announcements.


I'm the founder and CEO of Glencore Associates, a Los Angeles private investment and asset management firm, and founder of AIDiscover, an AI-powered marketing and brand discovery agency. I write about investing, AI implementation, and company building on Substack and Medium.


This is my own read of public data, not investment advice.

Comments

Post a Comment

Popular posts from this blog

The Layoffs Aren't About the Balance Sheet Anymore

 I'm David Bellairian, founder and CEO of Glencore Associates and founder of AIDiscover. I read a lot of layoff announcements this year, partly because I allocate capital for a living and partly because AIDiscover works with companies going through exactly this kind of restructuring. The pattern in 2026 doesn't look like the layoffs I remember from prior downturns, and I think most coverage of it is asking the wrong question. The scale of it Tech layoffs topped 245,000 in 2025 and are already running past 180,000 for 2026, on pace to come in higher than last year. Amazon alone cut 16,000 corporate roles in January, on top of 14,000 the previous October. Oracle cut as many as 30,000, roughly a fifth of its workforce, days after reporting strong quarterly earnings. Uber cut 3,300 jobs as part of a restructuring the CEO described as removing management layers, not cutting costs to survive. Cisco cut 4,000. HPE announced 2,500 over eighteen months. Here's the detail that actual...
Read more

Are We in the Late Stage of the Big Debt Cycle? What the Fed and Leading Economists Are Actually Saying

 I'm David Bellairian, founder and CEO of Glencore Associates. I allocate capital for a living, and lately the question I get most from other operators and investors isn't about any single stock. It's about the debt. Is the United States actually approaching a crisis, or is this the same recurring panic that's been wrong for fifteen years running? I went back through what the Fed itself is saying, what the government's own watchdog projects, what Ray Dalio's debt-cycle research shows, and what the leading skeptic on the other side argues. Here's what I found. The numbers as of 2026 Total federal debt has climbed past $38 trillion. Debt held by the public, the portion that matters most for interest costs, sits above $31 trillion. The Government Accountability Office, Congress's own fiscal watchdog, projects annual deficits averaging over $2 trillion every year through 2036, with publicly held debt reaching roughly 123 percent of GDP by then. Net interest ...
Read more

David Bellairian on Holding a Law Degree Without Practicing Law

Image
Three years of law books, and a career that went somewhere else. Photo via Unsplash. I’m David Bellairian. I run Glencore Associates , a private investment and asset management firm in Los Angeles, and I founded AIDiscover , an AI marketing agency. I also hold a Juris Doctor degree, and I have never practiced law. I never sat for the bar. People find that combination odd enough to ask about, so this essay is the answer I give, written down once. The short version I went to law school, finished the degree, and then chose not to pursue the bar exam, because by the time I finished I knew that my passion was management and marketing, not litigation or legal practice. That isn’t a story about failing at something. It’s a story about figuring out what I was actually for, and then being honest enough to act on it. I want to be exact here, because precision is the whole point of the degree. I am not an attorney. I am not licensed to practice law in California or anywhere else. I do not give le...
Read more