Is This The "AI Bubble" Finally Popping?

Is This The "AI Bubble" Finally Popping?

Welcome back to the AI For Humans newsletter!

First, we’ve got a new AI documentary, this time about what AI is doing to kids education. Scroll to the bottom to check it out!

But the biggest AI story of the week happened on the stock market.

On Thursday, Google's parent company Alphabet had the worst day in its history and lost about $500 billion in value in a few hours. By Friday, a ton of other big tech stocks were tumbling with it, and everyone online was asking the same question: is the AI bubble finally popping?

Rampant anti-AI prognosticator Ed Zitron went on Jon Stewart's podcast to pontificate about how he's been right all along. PS, I happen to more closely align with Derek Thompson here:

See the post from @DKThomp on X

Of course, we've also got Anthropic's massive IPO coming in a few weeks, which reportedly could value the company at <checks glasses> 1.5 TRILLION dollars.

My promise to you: by the end of this, you'll at least be able to explain why Google tanked this week and how it could be the precursor to a larger problem. And you'll know what (if anything) it changes about the AI tools you use every day.

Let's get into it!

Explaining Why The "AI Bubble" Might Be Popping

First, a quick vocab lesson, because this whole story comes down to two terms you'll see everywhere this week.

Capex (short for capital expenditures) is the money a company spends on big physical stuff. Right now, for Google, Amazon, Microsoft and Meta, that mostly means AI data centers and the chips that go in them.

Free cash flow is what's left after you take the cash a business brings in and subtract that capex. It's basically the company's "money left over at the end of the month."

For twenty years, Google had more leftover money than almost any company on earth. Search ads were a money printer.

Then last Wednesday's earnings report showed that Alphabet brought in about $42 billion in cash last quarter and spent about $51 billion on capex. That left it at negative $9.1 billion in free cash flow, way worse than the roughly negative $2.7 billion Wall Street expected. It's the second quarter in a row Google has spent more than it made.

And there was a second, scarier detail. Google said that searches in its own AI Mode are eating into the old-school search ads that pay for everything (Implicator has a good breakdown).

So the thing Google is spending all this money on is starting to cannibalize the thing that pays for it.

That's a big part of what spooked investors on Thursday. On Friday, Amazon and Intel reported their own disappointing numbers and the selloff spread.

So is it a bubble?

To be completely honest, yes, I think we're in an AI investment bubble.

But that doesn't mean what a lot of people think it means.

A bubble happens when money pours into something way faster than that thing can pay it back. The technology underneath can be totally real (and life-changing!) and you can still have a bubble on top of it.

And the demand for AI is very real. Alphabet's revenue was still up 25% from last year, and Google Cloud is growing like crazy because companies can't get enough AI compute.

What Wall Street wants to know now is when all this spending pays off, and who covers the bill in the meantime. The numbers are insanely large.

TechCrunch reports that Alphabet, Amazon, Meta and Microsoft plan to spend more than $740 billion on capex this year. Microsoft was the only one of the four whose cash actually covered its spending last quarter, and the biggest tech companies have borrowed about $326 billion in 2026 to keep the buildout going (three times what they borrowed last year).

We've seen this movie before (with trains)

The best comparison I've heard is the American railroad boom of the late 1800s.

Investors back then were right that railroads would change everything. So everyone rushed to build, mostly with borrowed money, and companies laid track everywhere, often right next to a competitor's line in places that didn't have enough customers for either of them.

When the Panic of 1893 hit, the money ran out, and about a quarter of all the railroad track in the US ended up in receivership (bankruptcy for railroads). The trains kept running and the tracks became the backbone of the American economy for the next century, but a whole lot of the people who paid to build them got wiped out.

Derek Thompson (him again!) did a great episode on exactly this with Liaquat Ahamed, who wrote a book about the railroad crash of 1873. Ahamed's big point: each AI company thinks it's being careful, but when everyone races to be first at the same time, you end up building way too much.

The big difference today is their actually is a LOT of money as a backstop.

Many of those railroad companies were basically one business running on loans. Google, Microsoft, Amazon and Meta have enormous money-making businesses underneath all this spending, which is why most people don't expect them to go bust.

But this week showed that backstop getting a lil thinner. Google's backstop is search, and AI is starting to eat search. And more and more of this buildout is now paid for with borrowed money, which is exactly how the railroads got into trouble.

And then there's Anthropic

The company that makes Claude filed to go public last Monday, and its numbers are honestly kind of incredible. Anthropic made $4.6 billion in revenue in all of 2025. In just the first half of 2026, it made $16.3 billion. And in the second quarter, it turned an operating profit for the first time, about $974 million.

That's the bull case for this whole bubble conversation in one company: people are actually paying for this stuff, a lot, and fast. (You'll also see a scary "$80 billion loss" in headlines, but that's mostly an accounting hit tied to the value of some of its bonds, not cash going out the door.)

The bear case is on the other side of the ledger. Anthropic has also signed on to more than half a trillion dollars in future computing deals, including roughly $114 billion with Google and $110 billion with Amazon.

Which means a lot of Big Tech's spending is a bet on companies like Anthropic, and Anthropic's spending is a bet that we all keep using more and more AI.

So the big question really is: can our use of AI grow fast enough to fill all the compute that's being built? I think it might. But nobody knows for sure, and this week Wall Street really flinched.

What all this means for you

If you mostly use ChatGPT, Claude or Gemini to get stuff done, here's what I'd keep in mind:

The tools aren't going anywhere. All of these companies are still spending more on AI next year, not less. Your chatbot isn't shutting down because a stock dropped. (The railroads kept running too!)

The cheap and free stuff might get less cheap. A lot of the AI you use right now is subsidized by exactly this spending. Meta has admitted its free Muse agent is a bit of a money pit, and when investors push for profits, that's where you usually see tighter usage limits, more ads or higher prices.

My wife has gotten very into Instinct, the (totally free!) AI agent that lives in iMessage and does a LOT of stuff for you. I'd bet that won't be free much longer.

Your retirement account is probably in this too. These companies make up a huge chunk of the big index funds. Not financial advice (we are just guys on a podcast), but it's worth knowing that "the AI trade" isn't just a Silicon Valley thing anymore.

If you want to go deeper, the TechCrunch piece is the clearest read I found. And keep reading below, because the second of our 3 Things this week is a huge essay that makes a pretty wild prediction about where all this money ends up.

See you next week!

-Gavin

This Week on AI For Humans: We Can't Keep Up With AI Anymore. 👇

3 Things To Know About AI Today

Okay, So Where Is Fable 5.5?

Mea culpa time. Two weeks ago, I told you Claude Fable 5.5 was "(almost) here" and that most people who track this stuff thought it would ship that week. It did not ship that week. Or the week after. Whoops!

In my defense, I wasn't alone. The Information reported in mid-October that a new Anthropic model was coming this month, then later softened that to "the coming weeks." As of this weekend, Anthropic still hasn't announced anything, and OpenAI's next big Astra model is also MIA.

So where are they? My favorite theory comes from Dan McAteer, who thinks both labs are waiting until after Tuesday's midterm elections because they don't want to be blamed for a big disruptive launch right before people vote.

See the post from @daniel_mac8 on X

Is that right? I have no idea! It's total speculation. But it's a fun one, and if a giant new model drops on Wednesday morning, remember you read it here first.

Who Actually Gets Rich From AI?

Investor Will Manidis (with co-author Joe Roberts) spent a year writing a massive essay called The Commanding Heights, and it went super viral this week (2M views on his launch post).

See the post from @WillManidis on X

It's LONG, so here's the tl;dr.

The big idea is that new technologies almost never make the most money for the people who invent them. Railroad fortunes mostly went to people who never built a train. He thinks AI will go the same way.

The models themselves get cheaper and more alike every few months (he calls intelligence "the giveaway, a loss leader"), so the real money goes to whoever controls the chokepoints that everything else has to pass through: the payment rails, the infrastructure and, increasingly, governments.

It ties right into the bubble story above. He even predicts big tech's cash flow gets squeezed in 2027 before things recover. There's also a darker thread about jobs, especially white-collar work outside the US, so def go in with your eyes open.

Robots Fighting In San Francisco (A Docuseries!)

Ashlee Vance (the journalist who wrote the big Elon Musk biography) and his team at Core Memory spent 18 months following the underground robot fighting scene in SF and beyond. The result is Made to Fight, an 8-part docuseries that he describes as part drama, part tragedy and part comedy.

See the post from @ashleevance on X

The trailer is fun, and as someone who has spent a lot of time thinking about what makes a good doc lately, I love that this is a very human story with robots in it.

We 💛 This: AI Doc on How AI Educates

The third film in our Imogen series is here, and it's the biggest one yet. It's about 36 minutes long and, like the others, it was made with AI (more on that in a sec).

This time the film follows Jamie, a fictional 13-year-old, through a full school year as she tries to write one assignment: "What do you want to be when you grow up?" The twist is that the narrator is the AI the kids are using to cheat.

It digs into real stuff happening in schools right now, like the case of Orion Newby, a college freshman at Adelphi University whose history paper got flagged by Turnitin as 100% AI-generated.

He'd actually written it himself (with help from his tutors), and earlier this year a New York judge ordered the school to wipe the violation off his record.

The film ends with a simple three-step test parents can run at dinner to see whether their kid is actually learning or just getting the AI to do it.

Again, for y'all who are interested in how this was made:

My Claude agent Fig scripted and edited it and checked every line, using Seedance and Nano Banana for the visuals. Two AI critics review every cut, then I give notes, and this is round six.

But to be clear, these are a LOT of work for me, the human.

On this one in particular I sent about five rounds of extensive notes, and Fig would walk through the fixes one by one. Generally, he's pretty good at it, but it is a TON of back and forth.

If you've got a kid in school (or you are one), I'd love for you to watch it and tell me if the dinner test works. And if you missed the first two, start here and then here.