They Always Rebuild the Park

Happy Sunday.

Quick question before we start. Where were you in December 1996?

Some of you were running desks. Some of you were building the companies you run now. And I'd bet every one of you remembers the two words that dropped that month, because they followed the market around for years afterward. Irrational exuberance. The most powerful financial voice on earth stood at a podium and politely asked whether asset prices had lost their minds.

You know what happened next, because you lived it. The Nasdaq tripled. It ran for three more years after the question was asked. The smartest man in the room was wrong, and wrong, and wrong, until one day he was right, and by September 2001 every dollar of that run was gone.

Alan Greenspan died a few weeks ago at 100. His wife's statement said he had irrational exuberance for baseball and for jazz. He was in on it to the end.

I've been chewing on that whole sequence ever since. Because the AI bubble call is now well over a year old, and it comes from serious people. Jim Chanos, the man who exposed Enron, calls this the platinum tier of fraud. The investor who called 2008 is short Nvidia and the chip sector, targeting a 30 to 40 percent correction by March 2027. The Federal Reserve just ranked AI the third biggest threat to financial stability in the country, behind only geopolitics and an oil shock.

And the market keeps not caring. Same as it didn't care in 1997. Or 1998. Or 1999.

So today I'm going to try something different. I'm not going to argue the bubble callers are wrong. Most of their math is right, and I'll show you the ugliest of it with no sugar on top. I'm going to argue they're answering a small question that happens to be sitting right next to a much bigger one. The word bubble describes what happens to prices over three years. It says nothing about what happens to the world over the next hundred.

And this buildout, whatever your portfolio thinks of it this quarter, is doing the second thing.

Every bubble is a Jurassic Park movie

A technician named Dave Lundgren has the best frame for this I've heard all year, and it's built on dinosaurs.

There are eight Jurassic Park movies now. Same story every time. Same structure, same beats, same ending. And we still reach into our pockets and pay to see it, because in the moment it feels different. Different characters, different dinosaurs, different island. New paint, same canvas.

Bubbles work exactly like that. Humans have been telling one or two stories for three thousand years, and every mania maps onto one of them. The internet bubble had its characters, its catalyst, its monster. So did housing. So did the work-from-home run after Covid. And inside every single one of them, people said the same four words. This time is different. It always feels different, because it's usually the first one you've personally lived through. So we laugh at the people who did silly things in 1929, and then we turn around and say yes, but AI is going to change the world, and there will be data centers in space.

Here's the part I love. Lundgren isn't a bear. He's long the leadership. His point is that the canvas and the paint are two different things, and you should never confuse which one you're trading. He buys and sells stocks, not narratives. As long as market structure holds, he stays long, gladly, because bubbles are where fortunes get made. The old Fidelity line is that there's plenty of money on the left side of a bubble. Semiconductors went up 600 percent from the 1998 bottom to the 2000 peak. You just have to know, every single day, why you own what you own, and it can never be because of the movie.

So let's grant the bears the canvas. It's the same movie, again.

Here's the paint.

The scoreboard says bubble

Start with the picture everyone's passing around.

Semiconductors are 19.7 percent of the entire S&P 500. One industry, one fifth of the American stock market. Highest weighting ever recorded, above even the dot-com peak. Everything else American public companies do, banking, medicine, oil, retail, food, entertainment, has been squeezed into the other 80 percent.

Zoom out a layer and it gets tighter. Seven companies are 34 percent of the index, about 22.7 trillion dollars, and they throw off roughly 70 percent of its economic profit. And because roughly 40 cents of every auto-enrolled 401(k) dollar flows into those same names through default target-date funds, this isn't a trade some people chose. It's a bet nearly every American worker is making, whether they know it or not.

Now lay the price action against every mania of the last 40 years.

Three years in, up roughly 250 to 300 percent, tracking the same arc as dot-com and Japan and housing at the same point in their lives. Every one of those lines eventually turned. Historians will tell you each cycle had different plumbing underneath, and that's true. The shape didn't care.

The behavior underneath the prices looks like the third act, too. Jeff Huge, a strategist who's been tracking this cycle closely, calls it the speculation generation. People aren't buying companies anymore. They're trading tickers with leverage. Volume in S&P 500 call options more than doubled in twelve months, from 1.2 trillion to 2.6 trillion dollars. Quick translation, a call option is a bet on a stock going up where you put down a little and win or lose a multiple of it. So that's not just record speculation. It's record speculation with the leverage baked in. Institutional allocators are sitting at their biggest overweight to stocks on record. The top 10 names are 41 percent of the index, which is the exact reading from March 2000. Price to sales, one of the only ratios you can't massage with accounting, just set an all-time record. The dividend yield on the S&P hit its lowest level ever.

Today’s Sponsor

Apple’s Starlink Update Sparks Huge Earning Opportunity

Apple just secretly added Starlink satellite support to iPhones through iOS 18.3.

One of the biggest potential winners? Mode Mobile.

Mode’s EarnPhone already reaches 490M+ users that have earned over $1B, and that’s before global satellite coverage. With SpaceX eliminating "dead zones," Mode's earning technology can now reach billions more in unbanked and rural populations worldwide.

Their global expansion is perfectly timed, and investors like you still have a chance to invest in their pre-IPO offering at $0.52/share.

With their recent 32,481% revenue growth and newly reserved Nasdaq ticker, Mode is one step closer to a potential IPO.

Tap into a $1T opportunity — invest now at just $0.52/share and get up to 20% bonus.

^{Please read the offering circular and related risks at invest.modemobile.com. This is a paid advertisement for Mode Mobile’s Regulation A+ Offering.}

^{Mode Mobile recently received their ticker reservation with Nasdaq ($MODE), indicating an intent to IPO in the next 24 months. An intent to IPO is no guarantee that an actual IPO will occur.}

^{The Deloitte rankings are based on submitted applications and public company database research, with winners selected based on their fiscal-year revenue growth percentage over a three-year period.}

The honest ugly

Here's the part the bulls skip, because I promised you the ugliest of it.

OpenAI just offered the United States government a 5 percent stake in the company. The pitch is giving every citizen a share of AI's profits. The problem is that AI has no profits, so the arithmetic works out to giving every citizen a share of the losses. Prediction markets put the odds of the government actually taking a stake in OpenAI this year under 30 percent, with Anthropic facing the same skepticism.

Days before that, Meta let slip that it plans to sell excess compute. Two words that were never supposed to sit next to each other in this story. A company that spent two years telling markets it couldn't build capacity fast enough is now a landlord looking for tenants.

There's a wrinkle inside the earnings, too. Huge points out that several megacap tech names hold big stakes in private AI labs, and under equity-method accounting, when the paper valuation of a stake doubles, the gain flows straight into reported earnings. Call it the Berkshire effect. Your portfolio marks up, your earnings look heroic, and none of it has anything to do with selling more of anything. By his math, strip those markups out of one recent quarter and some of the blowout tech beats land right on top of ordinary analyst estimates. A slice of the earnings everyone is celebrating is the boom grading its own homework.

The financing is the part that actually keeps me up. Estimates put the American AI datacenter buildout at 11 trillion dollars between 2024 and 2029, with roughly 7 trillion of it financed through debt. A growing slice sits in special purpose vehicles, off-balance-sheet structures designed to keep the borrowing away from the parent company's credit rating, absorbed by a private credit market that gets a fraction of the scrutiny banks do. Sequoia's David Cahn tracks the gap between what's spent on AI compute and what AI actually earns at roughly 600 billion dollars a year, and widening. MIT found that 95 percent of enterprise AI pilots produced no measurable return. Ilya Sutskever, the man who proved scaling worked, says the age of scaling is over. Open-source models reach 90 percent of frontier performance within months, at a fraction of the price.

Every word of that is real. If you stopped reading here, you'd sell everything.

Don't stop reading here.

The bubble isn't where you're looking

Because this is where 2026 stops rhyming with 2000, and it's the strangest thing happening in markets right now.

The best earnings in semiconductor history are getting cheaper by the day. Nvidia printed 81.6 billion dollars of revenue last quarter, up 85 percent from a year ago, and the stock has gone nowhere all year. It trades around 21 times forward earnings. That's in line with the S&P 500, it's a fraction of its own five-year average of 72 times, and it's the cheapest the stock has been since 2019. Micron grew revenue 346 percent and the stock fell 25 percent from its high. It trades near 7 times forward earnings. Samsung and SK Hynix broke records and sit in bear markets.

Quick translation. A forward P/E is what you pay today for a dollar of next year's expected profit. In March 2000, tech traded at 55 times. Today it's around 22. Chip stocks specifically trade at 18 times, cheaper than the market they supposedly inflated.

In a classic price bubble, multiples run out ahead of earnings. Right now the exact opposite is happening. Earnings are doubling and multiples are compressing. The market is actively de-rating its own winners while they report the best numbers in their history.

Here's the hole in that, though. A low multiple on a fake E isn't cheap. If part of that profit is paper markups and circular vendor financing, 21 times earnings is a mirage, not a bargain.

Fair enough. But notice what that does to the argument. It moves the fight off the price and onto the quality of the number underneath it. Which is exactly where the froth actually went.

Not into prices. Into spreadsheets. Wall Street just raised its long-term earnings growth forecast for the S&P 500 to 25.5 percent a year, the highest ever recorded. Analysts now assume chipmakers will keep 50.3 cents of every revenue dollar as profit, forever, more than triple the market's margin. Ed Yardeni put it plainly in a note last week, questioning whether analysts are being rational when they treat a historically boom-and-bust industry as a permanent secular grower.

If there's a bubble here, it isn't in the prices. It's in the forecasts. And that's the dangerous kind, because prices correct in public. Forecasts just get revised, quietly, on a Tuesday.

Bubbles end. Supercycles don't.

Here's the frame I actually believe, and it requires holding two things at once.

A bubble is a statement about prices. A supercycle is a statement about the production function, the actual machinery of how an economy turns inputs into output. They can occupy the same three years and have completely different endings. They usually do.

Electricity had a bubble. Railroads had several. The internet had the most famous one of all, and I made this point a couple weeks back with Cisco. At its March 2000 peak, Cisco traded at roughly 50 times revenue. Not earnings. Revenue. Baked into that price was total certainty that Cisco would change the world. Two things happened from there. Cisco changed the world, and the stock fell 90 percent while it was doing it. The narrative was completely right and the stock was completely wrong, at the same time, for years.

The crash wiped out the people who funded the internet. It did not wipe out the internet. Amazon and Google built most of their empires in the decade after it, on infrastructure the dead companies had already paid for.

Now look at the longest chart anyone has.

A dollar put into American stocks in 1870 is worth over 35,000 real dollars today. On that chart, the Great Depression is a bump. The technology wins. The line goes back up.

But hold onto how long that took. It matters more than it sounds like right now.

The country is paying a deferred bill

Now the deeper reason I can't sign the death certificate, and it has nothing to do with stock prices. It's about what the money is actually buying.

[CHART 5: US energy consumption versus real GDP since 2004]

American energy consumption has been essentially flat since 2004 while GDP grew by roughly 60 percent, adjusted for inflation. For twenty years, growth came from software, from code you could copy for free, and corporate America spent its cash flow on buybacks instead of concrete. We financially engineered instead of building. The grid, the transformers, the transmission lines, the generation capacity, all of it aged in place while the index kept setting records.

The 11 trillion dollar figure looks insane compressed into five years. Spread it across the two decades of buildout we skipped, and it starts to look less like a mania and more like a deferred mortgage payment coming due all at once.

That's why I think the money keeps flowing even through the defaults that are absolutely coming. When the stress hits, and it will, the balance sheet that shows up won't be a bank's. Washington is already circling stakes in these companies. The buildout has been quietly reclassified from a trade into infrastructure, and countries do not let their infrastructure fail. They nationalize the pain and keep pouring.

That's not a clean story. It might be a deeply unfair one, private upside and socialized downside, and if the bailout comes we'll all write angry letters about it.

But it tells you the direction of travel. The pouring doesn't stop.

What's actually being built

For all of recorded history, a nation's supply of intelligence was demographics plus luck plus 25 years. You needed the birth rate, then the schools, then a generation of patience, to produce one cohort of sharp young engineers and analysts and doctors. Smart young people were the scarcest strategic resource on earth, and countries fought over them with visas and universities and salaries.

That constraint is dissolving in front of us. What this buildout manufactures is cognition as a utility.

Every country, aging or young, rich or modest, is about to be equipped with the equivalent of millions of brilliant, tireless, endlessly patient young minds. They don't emigrate. They don't age out. They don't revolt. They don't ask for a corner office. Japan's demographic collapse becomes a procurement problem. The Gulf states can buy the workforce they never birthed. A talented kid in Lagos or Manila gets a research team that only Goldman could afford five years ago.

That is not a chatbot story. That's a rearrangement of what makes a nation powerful.

And it's why the next ten years matter so far beyond any drawdown. The rails being laid right now, the chips, the grids, the model standards, the alliances over who gets compute and who doesn't, will harden the same way the railroad gauges of the 1870s did. The same way Bretton Woods did in the 1940s. The same way internet protocols did in the 1990s. Nobody voted on TCP/IP. It just got built first, and then thirty years of the global economy ran on it.

We're in the constitution-writing decade of the intelligence economy. Constitutions outlive every crash that happens while they're being drafted.

So what

Usual reminder, I'm not a financial advisor and none of this is advice. You know the drill.

"The technology survives" is cold comfort if your horizon is ten years and not a hundred. Japan is the proof, and it's sitting right there in Chart 2. They saved the system. They nationalized the pain. The technology kept compounding, Japanese companies kept building world-class things, and the Nikkei took 34 years to get back to even. The rails and your returns are not the same asset. You can be completely right about the century and completely wrecked by 2030.

Which is what that chart is really saying. Since 2013, American stocks returned 11.8 percent a year after inflation. From 1928 to 1948, it was 0.6 percent. From 1966 to 1982, real returns were roughly zero for sixteen years straight. From the 2000 peak to 2013, roughly zero again. The regime you and I grew up in is the outlier, not the norm. We're not good investors. We're spoiled ones, and the market doesn't owe us the last decade twice.

So hold both. The financial structure has real rot in it. The debt is real, the circular revenue is real, and a repricing is closer to inevitable than not.

And understand that this isn't one market anymore. Lundgren describes two robust markets running at the same time, a real bull market in the AI leadership and a real bear market in everything sitting in AI's crosshairs. They're telling you two different stories about the same technology. When money does rotate out of the leaders, remember how small everything else has gotten. Huge describes the recent shifts into real estate and staples as pouring buckets of money into thimbles. The defensive sectors are now so tiny relative to tech that a trickle of rotation makes them look like rockets. A thimble overflowing is not a new bull market.

But don't let the market's verdict on 2027 earnings stand in for history's verdict on the technology. Those are two different courtrooms, with two different judges and two very different calendars. The people who confused them in 2001 sold the internet at the bottom.

Greenspan asked how anyone knows when exuberance has gone too far. Thirty years later, the honest answer is still that we don't. Not in real time. Not ever. The dot-com crash was the loudest financial event of its decade, and from a hundred years out it will barely be a footnote to the thing it briefly interrupted.

One last thing about those dinosaur movies. Lundgren is right that it's the same story every time. The fences fail, the dinosaurs get out, the people run.

But watch what happens after the credits, in every sequel, without exception.

They rebuild the park. Bigger.

Stay curious 😎

- John

Today’s Sponsor

Introducing The First Agentic CRM

Get revenue agents, workflows, and automations across every stage of your motion. Access customer data in real time through Attio's web app, MCP, API, and SDK.

Then Ask Attio anything about your business and get instant answers.

It's the CRM that runs the work behind every win.

Try Attio Now