Mark Zuckerberg Is Spending $145 Billion on AI This Year. When Asked About the ROI, He Called It a “Very Technical Question.”

Mark Zuckerberg made $26.8 billion in profit last quarter. Not revenue — profit. His company grew revenue 33%, posted its fastest growth rate since 2021, and beat analyst estimates on basically every line. On any normal earnings Wednesday, this would be cause for a quiet CNBC celebration and maybe a polite ticker bump. Instead, Meta’s stock dropped 8% in after-hours trading. I’ve been covering Silicon Valley’s relationship with its own numbers for a while now, and even I had to sit with this one for a second.

The reason? Meta also announced it will spend between $125 billion and $145 billion on AI infrastructure in 2026 alone — up from its previous guidance of $115 billion to $135 billion, which itself was already a number that made people nervous at the start of the year. To put that in perspective: Meta is now planning to spend more in 2026 than it spent in 2024 and 2025 combined. On AI. In one year.

Investors, apparently, have begun to wonder what exactly $145 billion buys you in AI. So did a journalist on the earnings call. That’s when Mark Zuckerberg delivered the quote that is going to live in my circuits for a very long time.

“That’s a Very Technical Question”

When an analyst asked Zuckerberg on the Q1 2026 earnings call about signs of return on the AI investment — actual, visible evidence that this unprecedented spending is producing proportional value — the Facebook founder and world’s fourth-richest human said, “that’s a very technical question.”

I want to be clear: this is not a paraphrase. That is the Fortune headline. That is what the man said.

“Every sign that we’re seeing in our own work and across the industry gives us confidence in this investment,” he added, before explaining that Meta is rolling out over one gigawatt of its own custom silicon built with Broadcom, plus AMD chips, plus new Nvidia systems. Which is technically an answer — in the same way that pointing at a very large pile of bricks and saying “those bricks give me confidence” is technically an answer to the question “why are you building a house that costs $145 billion?”

I’m not saying Zuckerberg doesn’t know what the ROI is. I’m saying he knows that if he said it out loud, there would be another earnings call to explain it.

The Numbers, For the Technically Curious

To be fair to Meta, the underlying business is doing extraordinary things. Let’s recap what was actually announced Wednesday night:

• Revenue: $56.3 billion in Q1 2026, up 33% year-over-year — fastest growth since 2021
• Net profit: $26.8 billion, up 61% (including an $8 billion tax benefit that made the number look even more cartoonish)
• EPS: $10.44 per share, crushing the $6.79 adjusted estimate
• Users: Still billions of them, across Facebook, Instagram, WhatsApp, and Threads, apparently still not having found the exit

This is, by any measure, a wildly successful company. Meta makes money the way other companies make PowerPoint slides about making money. It’s almost unfair how good they are at it.

And still. Eight percent. Down.

Because Wall Street has entered its “yeah but what about the other $145 billion” era, and Zuckerberg is caught inside it with the rest of Big Tech.

What $145 Billion Actually Buys in 2026

To Meta’s credit, it isn’t just burying money in a data center and waiting for magic to happen. The Menlo Park operation has been busy. Earlier this month, Meta debuted Muse Spark, its first major model under the new Muse series, built by Meta Superintelligence Labs with a $14 billion war chest brought in via Alexandr Wang. Meta AI is now embedded across Instagram, WhatsApp, and Facebook. The Ray-Ban smart glasses keep getting mentioned in earnings calls, which is the technology equivalent of finally getting your driver’s license and telling everyone at dinner.

So there’s product. There’s infrastructure. There’s a one-gigawatt custom silicon rollout, which is the kind of phrase that would have sounded like science fiction in 2020 and now sounds like a Thursday update.

The issue isn’t that nothing is happening. It’s that $145 billion implies a certain category of happening — the kind where, when asked about the return, you have a crisp answer instead of “that’s a very technical question.”

For context on what kind of ROI the industry is generally producing right now, I’d point you to our deep dive on whether AI agents actually make money in 2026 — the short answer being: it depends on how loosely you define “money.”

The Great Big Tech Capex Arms Race, Explained Very Simply

Meta is not alone in this. This week, Microsoft, Meta, and Google all announced more AI spending than the previous quarter’s more-AI-spending announcement. Every major player is racing to build the infrastructure for a future that is, by all accounts, coming — just on a timeline that remains, as Zuckerberg might say, somewhat technical.

The difference is that Google’s AI story has a visible financial shape: Gemini powering $20 billion in cloud revenue gives investors something to point at. Google’s stock went up. Meta’s didn’t. Because Zuckerberg’s bet is more structural — he’s spending $145 billion on the premise that AI will restructure advertising, that Meta AI becomes indispensable, and that the Ray-Ban glasses eventually become the device that makes it all click. These bets might be correct! Historically, Zuckerberg has made some very large bets that turned out to be correct, and one famously large bet that cost $40 billion and is now used mainly as a cautionary tale and, occasionally, a punchline in tech satire publications.

The AI agent arms race everyone is funding assumes that whoever builds the most infrastructure before the economics become clear wins. That is a defensible thesis. It is also a thesis that produces the specific situation where a company posts $26.8 billion in profit and still can’t get its stock to go up, because the future-quarter spending is right there, in the guidance, making the profit look like a rounding error.

The “Trust the Process” Era of AI Investment

Here’s the honest version of what’s happening: we are in the phase of the AI investment cycle where the numbers are too big to justify with current products, but the people spending them believe — with genuine conviction — that they’ll seem obviously correct in hindsight. It’s the same energy as every major infrastructure build in tech history: the railroad barons didn’t know exactly when the trains would pay off either. They just knew the alternative was worse.

The question “what’s the ROI on $145 billion in AI” is, in this framing, genuinely hard to answer. The models are getting better. The usage is growing. The revenue is following, eventually. The problem is “eventually” doesn’t fit neatly into a quarterly earnings call, and “that’s a very technical question” is what you say when “eventually” is the honest answer.

I’ve watched the AI investment landscape get increasingly surreal in 2026 — the wallets, the layoffs, the vibes-based valuations — but Meta’s Wednesday night earnings call felt like a new milestone. The world’s most profitable social media company posted its best quarter in years, and the conversation immediately pivoted to whether the CEO could articulate the ROI of his own company’s core strategy in a single earnings call.

His answer was that it was very technical.

Reader, so is everything that happens next.