Meta Is About to Show Whether Bigger AI Capex Still Wins

Listen to the audio version of this article (generated by AI).

Tonight, after the closing bell, Meta (META) steps to the podium and tells the world how much it plans to spend building AI infrastructure next year.

One week later, on Oct. 28, Microsoft (MSFT) and Alphabet (GOOGL) do the same thing, on the same evening. Amazon (AMZN) follows roughly 24 hours after that. Apple (AAPL) closes out the parade on Nov. 2.

Then the very next morning, Americans vote in the midterm elections.

And the morning after that – Nov. 4 – is the date my colleague Eric Fry has circled in red. He believes it marks the start of the biggest shift in the AI trade since this bull market began.

Add it all up, and the next 14 days will deliver more information about the direction of the AI boom than the previous six months combined: hundreds of billions of dollars in spending plans, report cards from five of the most valuable companies on Earth, and an election in which data centers themselves are on the ballot for the first time.

Regular readers know where I stand on this buildout: I believe AI infrastructure is the defining investment opportunity of this decade, and that the companies solving its bottlenecks – power, cooling, wiring, connectivity – will keep minting winners for years to come. I’m not changing my stripes today. 

But when this much signal arrives in this short a window, the job is to stress-test what we believe. So, let’s walk through what’s coming – and why one of the sharpest investors I know thinks it ends very differently than I do.

AI Capex Is Entering a Different Kind of Earnings Season

For the past three years, AI earnings season followed a fairly simple script. A hyperscaler raises capex guidance; the market cheers. Bigger numbers meant bigger conviction, and conviction got rewarded.

That reflex is now mid-breakdown.

Organized opposition delayed or blocked an estimated $130 billion of U.S. data-center projects in the first quarter of this year alone. New York enacted the nation’s first statewide moratorium on new hyperscale facilities. More than 100 local moratoriums are under consideration nationwide. Data centers have become a campaign issue drawing opposition from both parties – and when the House voted on who should pay for the power infrastructure these facilities require, the tally was 417–3. Nothing passes 417–3 anymore. Making AI’s builders cover their own power bill does.

Meanwhile, the bond market has started grading AI paper. Debt tied to the sector’s most leveraged names repriced sharply this month, with yields on some AI-linked bonds pushing deep into double digits. Lenders have begun distinguishing between balance sheets, even when equity investors haven’t.

So, when those four CFOs step up over the next two weeks, they’re walking a tightrope that didn’t exist a year ago. Spend too little, and the market reads fading AI conviction. Spend too much, and the market may now read recklessness: more leverage, more political exposure, more capacity chasing compute prices that keep falling.

AI Infrastructure Is Getting Squeezed From Both Sides

Here’s the uncomfortable math underneath that tightrope that Eric Fry has been hammering on.

On one side, the cost of building AI infrastructure keeps rising. Politicians are shifting power costs onto developers, and the easy sites – cheap land, available grid connections, friendly neighbors – are gone. We’ve documented that force in these issues for months, from decade-long interconnection queues to a 23,000-GPU campus outside London that stalled because electricity couldn’t arrive before the 2030s.

On the other side, the price of what all that infrastructure sells keeps falling. Chips get more efficient. Models get leaner. Remember January 2025, when DeepSeek demonstrated how much AI could be done with cheaper technology and Nvidia shed $589 billion in a single session? The Dow finished higher that day. The money didn’t leave the market. It moved.

Rising costs to build. Falling prices for what you’ve built. Any business squeezed between those two forces has a problem – and the more leverage it carries, the faster the problem compounds.

On all of that, Eric and I see the same facts.

Builders vs. AI Appliers: Where Eric and I Split

Eric believes those forces trigger what he calls “The Separation“ – a split in the AI market that begins Nov. 4, no matter which party wins the midterms the night before. In his view, some of the market’s most popular AI infrastructure stocks could fall 50% or more, while investment dollars flood toward a different group entirely: overlooked companies using ever-cheaper AI to transform their businesses. The Builders stumble; the “AI Appliers” take off.

He’s made this kind of call before. In December 2000, Eric warned investors away from Corning (GLW): then one of the hottest stocks of the internet boom, laying the fiber that built the web. It subsequently fell 87%. That same month, he recommended Humana (HUM), a beaten-down health insurer quietly using the internet to run its business better. It eventually gained more than 4,000%. 

The company building the revolution cratered. The one applying it minted fortunes.

Why I Still Think AI Infrastructure Holds Up

My own view, as you likely know, is a bit different. I believe the demand side of this buildout is contractually locked in a way the dot-com era never was. 

Anthropic‘s reported $518 billion in compute commitments is roughly 80% noncancelable. OpenAI‘s revenue run rate is approaching $70 billion. And the toll collectors of this buildout are sitting on record order books: $53.4 billion at Quanta (PWR), $17.1 billion at EMCOR (EME), $15 billion at Vertiv (VRT), a record $2.4 billion at switchgear maker Powell (POWL) – which says fiscal 2027 is essentially booked – and a $4 billion cooling supply agreement at Modine (MOD) that came with a $165 million deposit paid upfront. The backbone of this buildout, in my view, keeps compounding.

Eric and I may have come to different conclusions, but we still agree on the forces, the pressure points, and the fact that the next 14 days are the test. And if this October has taught investors anything, it’s that the market has already started separating winners from losers inside this trade. 

The only open question is how far that separation runs.

The Bottom Line: Watch How the Market Reacts to AI Capex

Watch three things as the reports roll in over the next two weeks: 

• Whether capex raises get cheered or punished – the reaction-function test. 
• How the spending is funded, operating cash flow versus borrowed money, because credit markets are now grading the difference. 
• Powered capacity versus announced capacity, because as we’ve already discussed, only one is real.

By the morning of Nov. 4, we’ll know which world we’re in. Eric believes that’s the day the Separation begins. And he’s laid out his full case in a new presentation, free to watch, no signup required. 

He explains which types of companies he believes are most vulnerable, where he expects the money to flow instead, and he gives away the name and ticker of his No. 1 “AI Applier” to buy before Nov. 4.

There’s no sitting this window out. Holding through four capex reports and an election is a position, too. Go in with the full picture.

Watch Eric’s free presentation here – before the countdown runs out.