In partnership with
In today’s post:
💰 $20T. Gone If AI Cracks
💉 -18% On Good News
📉 Why Beats Stopped Working
What the People Building AI Actually Read
The people building AI aren't only reading about AI. Sam Altman's pick for getting more done is Zero to One, and it has nothing to do with machine learning.
Shortform breaks down books like The Beginning of Infinity and Thinking, Fast and Slow into the key ideas, the context behind them, and what to do with them.
Get the ideas behind the breakthroughs in about 15 minutes, then decide which ones deserve your whole weekend.
💰 $20T. Gone If AI Cracks
An AI crash would vaporize $20 trillion.
That's the price TS Lombard's Dario Perkins put on a collapse in AI spending, in a scenario analysis published Thursday.
Also on the bill: $125B to $190B in credit losses and a recession. The good news? He thinks it would be a mild one.
📉 Who Gets Flattened First?
In Perkins' crash scenario, the S&P 500 drops 30%. Tech takes the biggest hit at down 40%, and every other sector falls 20%.
Why so brutal? Market concentration. When a few AI giants are holding up the whole index, they drag it down with them on the way out.
The hyperscalers should survive. Meta, Amazon, Google and Microsoft have balance sheets built like nuclear bunkers.
The credit damage lands on the weaker links in the AI chain:
Neoclouds like CoreWeave
Standalone model developers like OpenAI and Anthropic
Construction suppliers building the data centers
And they're running up quite the tab. By late 2027, Perkins projects the non-hyperscaler AI crowd will be carrying:
$545B in bonds
$920B in bank loans
$650B in private credit
That's over $2 trillion of borrowed money betting the party never ends.
🏚️ So How Bad Does The Recession Get?
Ending the data center boom would knock around 2 percentage points off U.S. GDP and wipe out roughly 800K construction-related jobs.
Still, Perkins expects something like 1990 or 2001. Painful, survivable, mostly forgotten by the next bull run.
His reasoning: private-sector balance sheets are healthy and the Fed has plenty of room to cut rates. AKA there's still gas in the emergency tank.
👀 Thursday Felt Like A Trailer
The note landed on a jumpy day. Meta and IBM both flashed warning signs after earnings, and investors bolted out of tech stocks and into government bonds.
Worries about AI's costs, sustainability and actual payoff are back. Wall Street is finally asking the awkward question: when does all this spending make money?
🧠 What It Means For Your Money
Perkins' big message: don't try to time the pop. He admits nobody knows when the AI capex splurge ends, and warns that shorting U.S. tech too early will "surely underperform."
His playbook looks like this:
Keep normal risk exposure, spread globally. International stocks should do fine if the boom rolls on, and outperform if it crashes.
Be patient with bonds. He'd add government bonds only once signs of fragility show up. If AI cracks and the Fed rushes to cut, long-term rates should fall significantly.
Watch the dollar. He sees it ultimately heading lower, since the damage centers on the U.S. financial system.
Why lean international? Perkins says this bubble is far more concentrated than Dotcom, piled into the U.S. and the AI supply chain. Non-US markets haven't bet the house on one theme.
The rest of the world forgot to go all in on AI. That could turn out to be its best trade.
If AI pops tomorrow, you’re...
Invest Alongside a $12B Real Estate Owner/Operator
Choose individual real estate investments alongside Lightstone. 20%+ GP co-investment in every offering, the same assets we pursue with our own capital. Our scale, your advantage. $100K minimum investment. For accredited investors.
Any message that includes model/proforma/targeted figures need a disclosure at the end such as: "This communication is for informational purposes only and does not constitute an offer to sell or a solicitation of an offer to buy any securities. Any such offer or solicitation will be made exclusively through the definitive offering documents. All investments involve risk of loss, including the potential loss of principal. Past performance is not a guarantee of future results. Any targeted returns or projections are forward-looking statements, are based on current assumptions, and are not guarantees of future performance. Actual results may differ materially.
This stock has lost half its value since June 2025.
Its earnings per share just grew +45.9%.
One of those numbers is wrong about where this goes next. We think we know which.
The market is fixated on the bad news. Revenue growth slowed to +11.8%, the third straight quarter of slowing growth, and the company is losing share at home.
What it's mostly ignoring: operating margin hit 34.6%, free cash flow reached $11.36B, and churn improved even after repeated price hikes.
That last part is the engine. When customers stay through price rises, most of each hike drops straight to profit.
Add an ad business that management expects to hit roughly $2.7B (+80%), and profits can keep compounding even while revenue growth slows.
The stock still hasn't found a floor, which is why the entry plan matters as much as the idea.
Inside today's Premium+ deep dive:
Whether the selloff has gone too far or has further to fall
The price zone we're watching, and why
What consensus is missing about the lever that drives most of the business
Exactly what would make us walk away
Here's the case for why we're buying, laid out so you can decide for yourself.
Unlock it with Premium+ for $19.99/month.
💉 -18% On Good News
Moderna just dropped ~18% in a day. On a trial win.
The cancer vaccine worked. The stock got flattened anyway.
Here's how a leaked dataset turned a victory lap into a faceplant 👇
🕵️ The Leak Nobody Planned
Moderna $MRNA ( ▼ 1.25% ) and Merck $MRK ( ▼ 1.34% ) are developing a personalised cancer vaccine called intismeran autogene.
The full results were supposed to debut on October 24 at the ESMO Congress in Madrid. Big stage, big lights, big reveal.
Instead, cancer news site Oncology Central posted the data early on Thursday, then quietly took it down.
Too late. Once something's on the internet, it's on the internet.
So ESMO lifted the embargo, and Merck and Moderna published the data themselves. Surprise party, ruined by the guest of honour reading the group chat.
📊 What The Data Actually Said
The vaccine was tested alongside Keytruda (pembrolizumab), compared with Keytruda alone. The combo delivered:
42% lower risk of melanoma returning or death after a median follow-up of 24.6 months (just over two years). Primary endpoint: hit.
38% lower risk of distant metastasis or death. Secondary endpoint: also hit.
Two goals, two hits.
So why did the stock fall off a cliff?
🤔 When "It Worked" isn't enough
Rewind to August, when Moderna shares spiked on topline results from the INTerpath-001 trial.
The catch? That announcement had zero details. Just "it worked."
That's like being told you passed an exam without seeing your grade. Everyone quietly assumes it's an A+.
Now the actual grade is out, and investors are grading the grade.
TD Cowen's Tyler Van Buren called the distant metastasis results "less exceptional" than the recurrence-free survival numbers.
The headline stat impressed. The backup stat was simply solid. And when hope has already been priced in, "solid" stings.
Merck got off lightly, closing down just ~2%. Moderna clearly had a lot more riding on this one.
💸 What It Means For You
Good science and a good stock move are two different scoreboards. This trial hit both of its targets and the share price still took a beating.
The lesson is about expectations. When a stock runs up on a vague announcement, the real details have to beat whatever investors imagined. That's a high bar, because imagination has no ceiling.
Worth remembering the next time a company teases "positive results" without the numbers. The spike is the easy part. The full reveal is where the bill arrives.
📉 Why Beats Stopped Working
Earnings were good. Stocks dropped anyway.
Tesla beat. TSMC and Micron put up stellar numbers. Meta delivered great results.
And the Nasdaq still took a beating. 🫠
So what's going on?
📉 The "Show Me The Money" problem
The market has stopped clapping for beats alone. Now it wants to know where the cash goes next.
The drama around CoreWeave made return on invested capital the metric everyone is suddenly watching.
Beat on revenue and profit, then shovel it all straight into more capex? Investors want to see the return first.
It's like getting a raise and immediately buying a bigger house. Impressive, until someone checks your bank balance.
The big trigger was a recent Gartner report flagging a mismatch between AI infrastructure spending and the revenue needed to make the neocloud model work.
Then add rising rates. A lot of this buildout runs on floating-rate debt, so every rate tick makes expansion pricier.
The "recycling debt" model (borrow, refinance, repeat) gets uncomfortable fast if benchmark rates keep climbing.
Throw in the U.S. and Iran flip-flopping in the headlines, and the market reached for the brakes.
🧠 Why We’re Not Panicking
Honestly? We think this sell-off is overdone. The fundamentals don't match the mood.
Demand is still there. TSMC and Micron reported rising AI demand across the board.
Spending keeps growing. Meta says capex increases next year, and I'd bet the other hyperscalers follow.
GPU rental rates keep rising even as gigawatts of new compute come online. New capacity should get more profitable over time.
The picks-and-shovels crowd is scaling up. Siemens Energy, GE Vernova, Caterpillar, Trane, Vertiv and others are expanding production for data centers.
Customers are broadening. About 10% of CoreWeave's backlog came from enterprise customers in Q1 '26. Small, but I expect it to grow as agentic AI spreads.
It's also worth remembering the damage is mostly in tech. Most other sectors are faring much better.
The indexes just look ugly because they're stuffed with AI names. When a handful of stocks run the party, one bad mood clears the room.
📅 What It Means For You
The next big test lands October 28, when Microsoft and Alphabet report.
We’ll be ignoring the headline beats. Cloud revenue growth is the number we’re watching.
If it's strong, the bulls come back. If it misses, it could be rough sailing until Nvidia reports in mid-November.
And if you hold broad index funds, remember how tech-heavy they are. An AI mood swing hits a "diversified" portfolio harder than the label suggests.