How AI Spending Became the Stock Market’s Biggest Problem

For most of this AI boom, the rule on Wall Street was simple: the more a tech giant spent on AI, the more its stock was rewarded. Last week the rule stopped working. Alphabet fell about 12% in one day, its worst session on record, after reporting negative free cash flow for a second quarter in a row. Amazon dropped 7% after earnings, S&P Global cut Meta’s credit outlook to negative, and on Friday the Nasdaq lost 2.9%. Behind all of it is one number: about $740 billion, what Amazon, Alphabet, Microsoft and Meta plan to spend on AI infrastructure in 2026, while the cash from their normal businesses no longer covers the bill.

Here is my view in one sentence. AI spending is not a problem because it is big. It is a problem because it has started to outrun the cash that pays for it, and when that happens the pain does not land on the companies doing the spending. It lands on the companies that depend on it. This is an opinion piece built on the numbers below, not a forecast.

What Is AI Spending?

When people say “AI spending” they mostly mean capital expenditure, or capex: the money big tech companies put into things that last for years. That means data centers, the chips that fill them (Nvidia’s GPUs, Google’s own TPUs), servers, networking and the electricity to run it all. The biggest buyers are the so-called hyperscalers: Amazon, Microsoft, Alphabet (Google) and Meta, with Oracle and a newer group of “neocloud” companies such as CoreWeave close behind.

Here is why it matters for profits. Capex behaves differently from a salary or an ad campaign: the cash leaves immediately, but the cost reaches reported profit slowly, spread over the years the equipment is used. Accountants call this depreciation. That is why the damage can stay hidden for a while. Alphabet’s depreciation was already up about 49% in the third quarter, according to 404K Research, and it will keep rising as more of the new equipment goes into service. We saw the same pattern in our Google stock analysis.

So why is Big Tech spending so much? Mostly because they believe demand for AI will keep growing, and none of them wants to run short of computing power when it does. The money goes into two jobs: training AI models, which takes enormous amounts of computing power, and running them for customers every day, which is what the cloud businesses sell. The spending is easy to justify while cloud revenue grows fast, and it is growing fast: Azure rose 48% last quarter and AWS 39%. The catch is that the equipment has to be paid for first and earns its money later. When spending grows faster than the cash coming in, companies cover the gap from their savings and, increasingly, by borrowing. That is the part investors have started to worry about, and it is what the rest of this article looks at.

The AI Spending Figures: How Big Is the Problem?

Big. Together the four hyperscalers plan to spend roughly $720 billion to $745 billion on capex this year, and TechCrunch put the total above $740 billion. Alphabet’s plan of up to $205 billion is more than double the $91.4 billion it spent in 2025, per Yield Theory’s tracker. And 2027 is meant to be bigger: Alphabet and Amazon both said so on their calls, and S&P Global projects Meta’s capex rising from about $140 billion this year to more than $225 billion in 2027.

Capex definitions differ by company, and Microsoft’s figure is a calendar-year estimate, so treat the combined total as an order of magnitude. Sources: TechCrunch, Fortune, 404K Research, Yield Theory, company earnings.

This is the number that bothers me. In the second quarter the four companies earned more cash than they spent. In the third quarter they spent about $200 billion on capex against roughly $175 billion of operating cash flow, which leaves free cash flow of about negative $26 billion, based on 404K Research’s tally. The spending is no longer funded by the business. The gap is funded by someone else.

To be fair, demand is not imaginary. Azure grew 48% last quarter, AWS grew 39%, and Microsoft’s contracted backlog stands at $677 billion, up 73%. The question is not whether people use AI. It is whether the returns arrive fast enough to justify the bill.

Can Rising AI Spending Create a Market Crash?

Spending on its own cannot. A company investing cash it earned is just investing. The danger starts when three things stack up: spending exceeds cash flow, the gap is filled with borrowed money, and the payoff comes later than the market assumes.

The second part is already visible. The five biggest spenders, the four above plus Oracle, raised about $326 billion of debt in 2026, roughly triple last year, according to S&P Global Market Intelligence figures cited by TechCrunch. Microsoft sold $25 billion of bonds. And on October 30 S&P Global said Meta’s leverage could pass 1.0x as soon as 2027, from about 0.1x in 2025, mostly because of leased data center capacity, as Fortune reported. The same week, the Financial Times said OpenAI cut the target of its latest funding round to $20 billion from $30 billion.

This is the chain Michael Burry has been warning about, and we walked through it step by step in Burry and the AI Bears Are Betting on a Crash. Here’s My Take. His bet is on semiconductors and the Nasdaq-100, which is exactly where the spending ends up. For the wider picture of who else is sounding the alarm, see our roundup of market crash warnings, and our earlier look at what happens to global markets if the AI trade unwinds.

Does that make a crash likely? No. Earnings are still strong and the biggest spenders still have enormous businesses behind them. What it does make the market is fragile: it now depends on a handful of companies continuing to spend money they no longer fully have.

Not So Fast: Big Tech Knows That

The hyperscalers are not blind to any of this, and last week gave a live demonstration of what they do about it. On October 28 Microsoft reported capex of $45.2 billion, below the $50 billion it had projected, and blamed delays in third-party capacity rather than a pullback. Microsoft’s shares rose about 2.4% after hours. CoreWeave fell 4.4% and Nebius 5.1% the same evening, and two days later CoreWeave sank 11% and Nebius 9%. A slightly smaller check from one customer was enough.

On the same call, Satya Nadella said, without naming anyone, that a business built on a few customers buying lots of capacity is not one he would call a real business, and that Microsoft would sell to large customers only when the pricing is right. Whether that is strategy or positioning, the message is clear. The buyers know they can slow down.

That is the part I think the market is underpricing. If the returns disappoint, Big Tech has options: order less, lean on its own chips, lease less, stretch the build-out. It still owns cloud businesses growing 39% to 48%. That flexibility is a safety valve for them and a threat to everyone downstream.

So Who Pays the Price?

My bet is the companies whose revenue is someone else’s capex. Two groups stand out, and a third sits one step further down the line.

1. Chipmakers and suppliers

Nvidia reported revenue of $96.2 billion for the quarter that ended July 26, up 106%, with $89.0 billion from data centers, and guided to $108 billion for the next one. Ahead of that report, analysts had pencilled in about $43.6 billion of quarterly sales to the hyperscalers, close to half of the total. When four budgets are half of your sales, a small change in those budgets matters a lot. On Friday Nvidia fell 4%, Broadcom more than 5% and Qualcomm 11%. Nvidia reports again on November 17, and I will be listening for any hint about 2027 spending.

2. The neoclouds, with CoreWeave as the example

CoreWeave is the cleanest example, and I want to be fair to it: it is not a bad business. Its revenue more than doubled in the second quarter and its backlog is about $104 billion. But backlog is not cash, and the cash side is stretched. As of June 30 it carried about $35.6 billion of debt plus $16.3 billion of lease obligations, it guided to $860 to $940 million of interest expense for the third quarter, and three customers made up 72% of its second-quarter revenue. In mid-October its shares fell about 15% in a day and, as we noted in our Burry piece, credit-default swap spreads on its debt moved above 1,000 basis points. Nebius has a similar profile, with three customers making up 24%, 21% and 14% of its second-quarter revenue.

Put simply, these companies borrow heavily to build capacity for a few big customers, and those customers just showed they can lower their orders.

3. Smaller AI companies on the same funding cycle

Smaller AI startups depend on the same flow of money, from investors, from lenders and from the cloud companies that rent them computing power. When OpenAI has to cut its fundraising target, it is a warning that the supply of capital is not unlimited. The bigger the company, the more cushion it has. The smaller ones will feel any tightening first.

So What Now? Watch the Suppliers, Not Just the Spenders

AI spending became the stock market’s biggest problem not because it is large, but because it is now financed by debt and by the hope that returns arrive on time. Big Tech can survive a slowdown, and it knows it. The companies that sell chips, rent out computing power or borrow to build for a few customers cannot.

I am not telling anyone to sell. I am saying it is worth knowing whether the AI stocks you own are the spenders or the suppliers, because in a slowdown they will not behave the same way.

The next tests are Nvidia’s report on November 17, CoreWeave’s next report, and the 2027 capex guidance in January. If capex plans keep rising and free cash flow stops getting worse, the bears will have to wait. If either breaks, look first at the companies on the receiving end of the spending, not at the ones writing the checks.

This article is opinion and analysis for informational purposes only and is not financial or investment advice. AllinAllSpace is not a licensed financial advisor. Investing in stocks involves a high risk of loss, and forecasts may not come true. Figures are from company reports and the sources linked above, and are accurate as of November 2, 2026; capex and free cash flow definitions differ by company. The view in the final section is our editorial opinion, not a recommendation.