Big Tech Is Borrowing Like It's 1873

A few days before his bank failed, Jay Cooke had the President of the United States as a houseguest.

Ulysses S. Grant came to Philadelphia in September 1873 to put one of his sons in a school nearby, and he stayed at Ogontz, Cooke's mansion north of the city. The place had more than 50 rooms.

Then Grant left for Pittsburgh. On Thursday morning, September 18, Cooke was at his office downtown when a telegram came in from New York.

His New York branch had closed its doors before 11 a.m. Cooke shut Philadelphia right behind it, and his biographer says he wept.

Two days later, the New York Stock Exchange closed for ten days.

It was the first time a financial panic had ever forced it shut.

Over in Pittsburgh, Grant's aide told a reporter the President had no opinion on the failure, "for he has none to express."

Cooke had helped finance the Union's side of the Civil War. What sank him was a railroad he couldn't stop lending to.

He Sold the Civil War to Farmers

When the war started, the government sold its bonds the way it always had, mostly to banks and wealthy investors. Cooke sold them to everybody.

He sent about 2,500 agents into nearly every county and town in the North, and by May 1865 his offices were taking more than 10,000 orders a day for $50 and $100 notes. All told, his agencies sold more than $1 billion of Union debt.

I've spent about 26 years in financial publishing and direct-response marketing. Cooke was doing direct response before anybody had a name for it, and he was very, very good at it.

After the war, he pointed that same machine at the Northern Pacific, a railroad meant to run from Lake Superior to Puget Sound. His writers pitched the land along the route as mild, almost tropical.

Newspapers started calling the frozen northern plains "Jay Cooke's Banana Belt."

Then the buyers dried up. After Vienna's stock market crashed in May 1873, a dispatch from Frankfurt said an American railroad bond "even if signed by an angel of Heaven would not sell."

So Cooke's bank kept the railroad going with its own money. His contract capped those loans at $500,000. By the time the bank failed, the railroad owed it about $5.6 million.

His London partners had warned him a year earlier, in a four-word cable: "You cannot carry company."

The depression that followed lasted 65 months, the longest economic contraction on record in the United States.

Google Went Cash-Negative for the First Time Since 2004

I've been thinking about Jay Cooke ever since Meta reported earnings last Wednesday.

Over the three months through September, Meta's business brought in about $32.8 billion in cash. It spent about $34.9 billion on data centers and equipment. So it came up about $2.1 billion short, a year after the same quarter left it $10.6 billion to spare.

Wall Street calls what's left over free cash flow. It's the money a company has after paying to run and grow the business, and it's what pays for dividends and buybacks.

Meta wasn't first. In July, Alphabet, Google's parent company, reported its first quarter of negative free cash flow since it went public in 2004.

It spent $44.9 billion on data centers and equipment in three months, against $39.1 billion of cash coming in.

Alphabet now expects to spend as much as $205 billion this year. That's roughly $600 for every person in the country, from one company.

Alphabet's Debt Doubled in Six Months

When a company's cash runs short, it can borrow or it can sell more stock. Big Tech is doing both.

Alphabet's long-term debt more than doubled in six months, from $46.5 billion at the end of last year to $98.2 billion by June 30.

In February, it sold a bond that doesn't come due for 100 years, which Bloomberg says is the first century bond from a tech company since Motorola sold one in 1997. In June, it raised another $49.6 billion by selling new shares.

Meta's long-term debt has nearly tripled, from $28.8 billion at the end of 2024 to $80.9 billion at the end of September.

That's just the debt on Meta's books.

It pays for some of its biggest data centers through separate companies that borrow the money to build them, then rent them back to Meta. And these companies are named after fried dough.

Beignet Investor raised $27 billion for a data center in Louisiana in October 2025, and Sopaipilla Investor raised about $12.5 billion this August for one in El Paso, Texas.

I run a few businesses. When one of them needs a loan, nobody names it after a pastry.

Now look at who's lending.

According to Apollo, the big private-credit firm, buyers put in orders for nearly five times what Big Tech wanted to borrow back in February. By July, orders covered less than two times.

That's how 1873 started. The tracks were still going down when the buyers for the bonds started thinning out.

CoreWeave Expected $4 of Interest for Every $1 of Operating Profit

The giants can carry this for a long time. Meta still had about $82 billion in cash and investments at the end of September.

The strain shows up first in the smaller borrowers.

CoreWeave buys Nvidia chips and rents out the computing power, sort of a landlord for AI. In June, it borrowed $1.25 billion by selling bonds that pay 9.625% a year. Investors paid full price.

In mid-October, those bonds traded as low as about 80 cents on the dollar, yielding close to 15%.

When a bond falls that far, lenders are telling you they're no longer sure they'll get all their money back.

For the quarter that ended in September, CoreWeave told investors to expect $860 million to $940 million in interest.

It expected its adjusted operating income, roughly the profit from renting out those chips before a few accounting items, to come in at $200 million to $260 million.

That's about $4 of interest for every $1 of operating profit.

Harry Seldon, who writes our technology issues, named CoreWeave one of his three picks for the AI computing boom. In The Coming AI Bailout? on October 15, he also walked through how it could crack. If you own it, that one's worth a read.

On October 14, Gartner, the tech research firm, ran the math for the whole industry.

To pay back what they're spending, cover the cost of borrowing and earn a 12% return, the companies building AI data centers would need about $8.5 trillion in revenue through 2030.

Gartner expects them to bring in $3.6 trillion at most. It also expects more than 80% of companies like CoreWeave to stop operating as independent companies by 2030.

The Railroad Got Built Anyway

The Northern Pacific's tracks finally met in Montana Territory in August 1883.

At the ceremony that September, Grant, a former President by then, helped drive the last spike. It was the same spike they'd used to start construction in 1870.

The railroad got built. 

But in the reorganization, the people holding Cooke's 7.3% bonds had to trade them for preferred stock and give up their place at the front of the line.

I lived through my own version of this. In the dot-com days, I earned a very high salary in tech. Then the bust came, I lost my job, and I ended up filing for bankruptcy.

The internet turned out bigger than anyone promised. A lot of the money that paid for it never came back.

So I'm not predicting that AI fails, or that Meta and Google end up like Jay Cooke. They have huge, profitable businesses underneath all this spending, and if I had to bet, they come out the other side just fine.

What I'm watching is who's lending them the money, and at what price.

Microsoft and Alphabet report after the close on Wednesday, and Amazon reports Thursday. The first thing I'll check is whether each one's spending on data centers ran ahead of the cash its business brought in, and how it covered the gap.

Sort Your Stocks Into Two Piles

Here's your one move this week.

Pull up the three biggest stocks you own and look at each one's free cash flow over the last four quarters. Yahoo Finance shows it for free, under Financials, then Cash Flow.

Now sort them into two piles. Companies paying for their growth with cash they earn go in one. Companies paying for it with money they borrow go in the other.

There's nothing wrong with owning the second pile. Plenty of great businesses borrowed to grow.

But size that second pile so a 50% drop wouldn't keep you up at night, because when lenders get nervous, those are the stocks that feel it first.

Jay Cooke Got Rich Again

Cooke didn't stay broke.

After his bankruptcy, he got a stake in a Utah silver mine called the Horn by promising the owners he'd bring a railroad out to it.

In 1879, he sold the mine and walked away with nearly $1 million. By 1880 he'd paid off his debts, and in 1881 he bought Ogontz back.

It took a silver mine to make him rich again.

In the early 2010s, the gold miners had their own Jay Cooke moment. They borrowed heavily to build monuments. One of the biggest sank billions into a mine that reaches more than 17,000 feet up in the Andes, and it never went into production.

When gold fell about 45%, the miners' main fund fell about 81%.

Today, many of those same companies hold more cash than debt, and they're paying shareholders more than they have in decades.

So while Big Tech goes from paying cash to borrowing, the miners went the other way. On Wednesday, I'll show you how far, and why I think most investors still haven't noticed.

What Premium Members Get Today

Everything above is free and yours to keep.

If your second pile turned out bigger than you'd like, the next question is what to change. That's the part I save for Premium Members, along with exactly what we own and what we do about it.

Since launching Premium, we've closed 14 trades. All 14 made money, with an average gain of 84.5%, against 40.18% for the S&P 500 over the same periods.

Today's Premium section is a review of one of our critical-minerals positions that's gone against us, checked against the exact conditions we published the day we bought it, along with what would make us sell.

When a pick isn't working, you deserve a straight answer, and Premium Members get one below.

Double D

🔓 Premium Content Begins Here 🔒