Listen to the audio version of this article (generated by AI).
On the morning of Sept. 18, 1873, the most trusted banker in America closed his doors.
Jay Cooke was a household name at the time. He was the man whose firm had financed the Union’s war effort by selling bonds to ordinary families across the North. If any institution in the country was safe, it was Jay Cooke & Co.
When it collapsed, the panic was so violent that the New York Stock Exchange shut down for 10 days.
Afterward, investors swore nobody could have seen it coming. Except the warnings had been visible for a year. Cooke’s great project – the Northern Pacific Railway – couldn’t sell enough bonds to the public, so his bank quietly absorbed them itself, financing its own customer. Money had been getting tighter all summer. Good railroad news had stopped lifting railroad stocks.
Nobody rang a bell at the top. Nobody ever does. But the signals were there.
We’re staying bullish on AI today – but we’re not blind. Every great boom ends. The railroads proved it in 1873. The dot-coms proved it in 2000. And someday, the AI buildout will prove it, too.
Most investors find out about a peak the hard way, months after it’s passed, while watching their gains evaporate. We don’t intend to be those investors.
That’s why we’re introducing a new tool we’ll be tracking closely: Luke Lango’s Peak Watch Checklist.
How We’ll Know When the AI Boom Is Peaking
Booms don’t die all at once. They usually fail in a predictable order, and the checklist tracks 13 warning signals grouped into the three stages a boom goes through on its way down.
Stage 1 is Froth: money gets cheap, plentiful, and a little circular – late-boom behavior that can last a long time.
Stage 2 is Cracks: the top forming, as good news stops moving stocks and money gets harder to raise.
And Stage 3 is Confirmation: the top is in, as chip orders, data-center occupancy, and project pipelines start rolling over.
Each signal gets one of four ratings: Clear (green), Watch (yellow), Flashing (orange), or Triggered (red). Each also comes with a specific tripwire – a concrete event that would flip it to red. No vibes, no guesswork. When a tripwire hits, we’ll know it, and you’ll know it. We roll it all into a single number – the Peak Watch Score – which counts how many of the 13 signals are lit (orange or red).
Today’s Peak Watch Score: 7 of 13 signals are lit.
That puts the AI boom firmly in late-cycle territory. But with most of the fundamental confirmation signals still clear, it does not yet tell us the top is in.
Let’s walk through the board, stage by stage.
The 13-Signal AI Bubble Checklist
Stage 1: Froth – All 3 Signals Are Flashing
Debt-funded capex: Flashing. Big Tech is now spending roughly all of its cash flow on AI and borrowing to cover the rest. Alphabet’s (GOOGL) $180-190 billion 2026 capex plan roughly equals its operating cash flow, and hyperscaler bond issuance passed all of 2025’s total by June.
Vendor financing: Flashing. This is when a supplier finances its customer’s purchases of its own products. Broadcom (AVGO) agreed to provide Anthropic up to $42 billion in financing tied to a $125 billion chip lease, and it’s reportedly arranging roughly $50 billion more for OpenAI. That’s a classic late-cycle tell.
Record stock issuance: Flashing. Smart companies sell stock when it’s expensive. Alphabet’s $85 billion share sale and SpaceX’s (SPCX) $75 billion IPO both set records, and Anthropic has filed to go public.
Stage 2: Cracks – 4 of 7 Signals Are Lit
Capex growth slows: Clear. AI spending isn’t just growing; it’s still accelerating. Alphabet’s capex jumped from $35.7 billion in Q1 to $44.9 billion in Q2. It turns red if two or more of the big four guide to slower capex growth in the same quarter. This is the one to watch this week, when Microsoft (MSFT), Alphabet, and Amazon (AMZN) report.
Good news gets sold: Flashing. When stocks fall on great earnings, the market is telling you the good news is already priced in. Nvidia (NVDA) fell after four straight earnings beats before breaking that streak in August. Last week, Nvidia and Broadcom slid after a report put OpenAI’s annualized revenue near $50 billion rather than the higher figure that had been widely cited – a gap that comes down to gross-versus-net accounting. Taiwan Semiconductor (TSM), Meta (META), and Lam Research (LRCX) have all now slid after strong reports this month as well.
Revenue lags spending: Watch. AI revenue is ramping fast – but spending is ramping faster. The big five are tracking roughly $750 billion in capex this year.
Pricing power erodes: Watch. The price of AI tokens – the units AI models charge by – hit a record low in early September, down by more than half from its summer peak. GPU rental prices, though, are still rising, with H100 rents at $3.49 an hour, up 4.5% in four weeks.
AI credit spreads widen: Flashing. A credit spread is the extra interest a borrower pays over safe government debt, and it’s how lenders price risk. Bond markets tend to sniff out trouble before stock markets do. CoreWeave’s (CRWV) bonds yield near 14%, and its latest loan priced a full point wider than the one before.
Capital raises fail: Triggered (selectively). Australian AI cloud firm Firmus pulled its roughly $5 billion IPO last week after weak investor demand. Now, that’s one company, not a system-wide freeze; but it’s a real crack. It goes system-wide if a second AI infrastructure raise gets pulled or slashed. Upcoming IPOs from Nscale and Lambda are the next tests.
The Fed tightens: Flashing. Booms end on interest rates more often than on demand. The Fed hiked in September for the first time since 2023, and its projections point to one more hike this year.
Stage 3: Confirmation – The Fundamental Signals Remain Mostly Clear
Chip orders digest: Clear. Micron (MU) just posted a record $54 billion quarter and guided higher, and Broadcom guided its AI chip revenue up again.
Capacity sits idle: Clear. North American data-center vacancy is just 1.4%, even with supply up about a third. Every rack being built is getting filled.
Projects get canceled: Watch. Firmus and its partner scrapped a planned 1.6-gigawatt Australian buildout. It’s the same company that pulled its IPO, so this isn’t yet a trend. But it could become one if several developers cancel or a hyperscaler walks away from signed leases.
The AI Boom Looks Late Cycle, Not at a Confirmed Peak
In summary, the froth is fully lit, and cracks are starting to show in financing, credit, and rates. But the confirmation signals are clean. Chips are still selling out, data centers are still full, and Big Tech is still accelerating its spending.
That’s late-cycle behavior, not a confirmed peak. A late-cycle reading isn’t a sell signal either – some of a boom’s biggest gains historically come late. What it means is that we’re no longer flying blind.
What Moves the Board Next
We’ll update the Peak Watch Checklist every week – and the next two weeks should move it.
Microsoft, Alphabet, and Amazon report third-quarter earnings this week: a direct test of “capex growth slows.” The Fed also meets Oct. 27-28, though hike odds have collapsed to around 8% since the latest CPI report – good news for the “Fed tightens” signal. The Nscale and Lambda IPOs will tell us whether Firmus was a one-off or the start of something system-wide. And this week’s news split the difference: Meta’s plan to raise potentially hundreds of billions through a new share class nudges “record stock issuance” the wrong way, while Nebius (NBIS) closing a $3.5 billion facility at reasonable rates is a point for the bulls.
All of which raises the question sitting behind all 13 signals: will this spending ultimately pay off?
Our math says the odds favor yes. But patience is the price of that payoff.
A Long Payback Period Leaves Plenty of Room for Scares
In our base case, revenue doesn’t clear the required-return threshold until 2035, with full payback around 2039. That leaves nearly a decade of room for financing scares, disappointments, and sharp selloffs between here and there.
Which is exactly why this checklist exists. Late-cycle booms reward the investors who stay. Some of history’s biggest gains come in a boom’s final innings; but they go only to the investors who can suss out a shakeout from a top.
The AI trade has clearly moved beyond the clean early innings. Capital is getting more expensive, some investors are pushing back, and credit markets are separating strong borrowers from weak ones.
However, the signals that would confirm a fundamental peak have not arrived together. That’s why we’re watching instead of running.
We’re staying bullish and keeping our eyes on the scoreboard.
We update the full Peak Watch checklist – every signal, rating, and tripwire – each week inside Innovation Investor and Early Stage Investor. The easiest way to start getting it is through my free XPANSE briefing. It lays out the Musk-centered buildout we believe drives this boom’s next leg, the three moves I’d make today, and one lesser-known stock I believe sits at the center of it all. It also shows you how to join us as a member, so the Peak Watch Score lands in your inbox every week before the crowd notices the music changed.
Nobody rings a bell at the top. We built the next best thing.
