By August 1, 2026, will at least two of Alphabet, Microsoft, Amazon, and Meta guide 2026 AI-related capex below current Wall Street consensus estimates?

The $600 Billion Question

Imagine walking into your boss’s office, saying, “I’d like $600 to $645 billion for a vibes-based infrastructure project,” and getting a yes. That’s essentially what Alphabet, Microsoft, Amazon, and Meta did when they sold Wall Street on the AI data center binge.

The market nodded along for a year. Then OpenAI reportedly missed internal growth targets, Meta flashed a $115–$135 billion capex plan, and suddenly the room went quiet. The question now isn’t whether AI is important. It’s whether investors will keep signing blank checks for a gold rush that’s starting to look suspiciously like deferred earnings season.

Here’s my call: by August 1, 2026, at least two of these four will guide their 2026 AI-related capex below the Street’s expectations at the time they speak. The megaspending era isn’t over. But the “whatever it costs” chapter is.

From Infinite Horizon to Earnings Season

For the past 18 months, AI has lived in that magical accounting zone called "the long term," where margins don’t matter and cash flow is a tomorrow problem. The justification was simple: build the compute, and the revenue will come.

Now the bill is showing up where it always does—cloud and ads. Google Cloud is supposed to be growing around 50% year over year to justify its buildout. Azure and AWS are expected to tack AI onto every contract like a resort fee. Meta is promising that all this silicon will somehow turn Reels users into AI customers.

The trouble: those lines are not exploding fast enough to make a $600+ billion 2026 capex party look painless. BCA’s Noah Weisberger put a clock on it: investors might give these companies a few more quarters, but not years, to convert AI spending into visible cash. When your sector is 17% of the S&P 500, the market’s patience is less Zen meditation and more shot clock.

Meta: Poster Child for Capex Hangovers

Meta is the canary, except the canary is wearing a $135 billion price tag and yelling about custom chips.

Mark Zuckerberg has laid out an eye-watering AI capex range and pitched it as the cost of owning the stack: proprietary MTIA chips, data centers everywhere, and a more closed, monetizable Llama 4 for serious customers. Investors responded by… not exactly celebrating. Meta stock has lagged Nvidia and Amazon despite being part of the same AI fairy tale.

When your capex guidance becomes the headline risk around your stock, you’ve hit the edge of the grace period. Meta is already trimming fat elsewhere—layoffs and cost cuts—to make room for AI. The next logical step is to stage or soften the capex curve: “We still believe in the opportunity, but we’ll pace investments with demand.” Translation: downshift 2026 versus what the Street currently has penciled in.

Meta is a prime candidate to blink first, not because it’s bearish on AI, but because its investors are visibly allergic to another Reality Labs-style bonfire.

The Others Don’t Have Infinite Credit Either

Microsoft, Alphabet, and Amazon have had smoother rides, but they’re not immune. Options markets are pricing in chunky earnings moves because nobody’s sure how far they’ll push the capex story before someone on the buy side throws a chair.

The underlying pressures are the same:

• AI demand looks less vertical than advertised. OpenAI, the unofficial AI index fund, reportedly missed its own growth targets. If the poster child of paid AI usage is wobbling, it’s harder to argue every GPU barn will be fully booked at premium rates.
• Cloud is the scoreboard. Azure, Google Cloud, and AWS are where AI ROI is supposed to show up. If growth there looks merely good instead of euphoric, the case for “just keep doubling capex” weakens quickly.
• Custom chips and partnerships cut both ways. MTIA, TPUs, Trainium, Inferentia: all are sold as ways to stretch each dollar of capex. Great. But if you can do more with less, guess what the CFO is going to do with the “less” part.

Add in power constraints, data center siting fights, and antitrust scrutiny, and it becomes not just financially awkward but physically difficult to keep ramping capex along the straight line the Street has drawn to 2026.

I don’t expect Microsoft to suddenly pretend it’s a dividend stock. It has the cleanest AI story and the most obvious monetization through Copilot and Azure. If anyone keeps pushing spend, it’s Satya Nadella. But Alphabet and Amazon have more reason to let consensus run hot, then pleasantly “surprise” by easing the slope once they’ve proven the basic model works.

How the Climb Turns Into a Taper

The pivot, when it comes, won’t be some dramatic “never mind, actually AI is over” moment. It’ll sound like responsible adulthood.

Management will talk about “normalizing” the build cycle, “bringing capex into line with revenue,” and “driving operating leverage as AI matures.” Guidance will come in as a range that conveniently sits below where analysts had modeled AI infra spend. Capex-slashers will be praised for discipline; the ones who keep pressing the accelerator will be forced to show receipts.

That pressure is already baked into the tone of questions analysts are asking. They’re less “how big can this be?” and more “show me the line item.” In a market this concentrated, the fastest way to blow up everyone’s year is to admit you’ve been buying GPUs on spec.

Could the Street be too pessimistic? Sure. If AI adoption re-accelerates—enterprise contracts close faster, usage spikes, pricing firms up—then capex stays heavier for longer. But that scenario has to fight reality: estimates are starting from a $600–$645 billion bar. The upside fantasy is already priced in.

The Forecast: Two Will Blink

My base case: sometime between now and August 1, 2026, at least two of Alphabet, Microsoft, Amazon, and Meta route their 2026 AI-related capex below whatever number Wall Street is dreaming up at that moment. It might be Meta plus one cloud peer. It might be all four in a macro scare. The common thread is simple: investor patience will be rationed, not infinite.

The blank-check phase of AI is ending. The new regime is: bring cash, show margins, and kindly explain why your tenth data center in Ohio won’t be a very expensive GPU museum.

In 2023, the market paid these companies to build the future. By 2026, it will be charging late fees.