AI's debt bill: who builds, who borrows, who pays

SPEAKER_1 (0:00)

This is Marketing Is Calling, the audio edition of Landline, the daily newsletter from PPC Land. It's Saturday, November 14th, 2026, and today's episode is a special on artificial intelligence and money. Is AI creating value or destroying it? Who wins? Who loses? And who is paying with borrowed money? This episode is read by an AI voice.

Let's start with a bond. On Thursday, Meta finished selling 9 billion euros of debt. The longest slice won't be repaid until 2065, and it pays 6.4 percent a year until then. A week earlier, Meta sold $15 billion of notes, some paying 7.5 percent, running to 2056

Think about that for a second. Meta built one of the most profitable advertising businesses ever. It didn't issue a single bond until August 2022, now it owes bond holders roughly $109 billion. It has sold $37.5 billion of new shares, and it's preparing to sell up to $60 billion more, shares with no vote attached. Alphabet has gone further. It raised about $85 billion of new equity, sold a bond that matures in 100 years, stopped buying back its own stock, and reported two quarters in a row of negative free cash flow.

In the third quarter, it spent $51 billion on property and equipment. That's more than all the cash its operations produced. Advertisers are still paying these companies more every quarter. But that money no longer covers the AI bill. The gap is being filled with debt and new shares.

The full special at ppc.land runs to more than 20,000 words, with a dozen tables, a timeline, and every source linked. If you want the whole picture, that's the version to keep.

So the ad tech companies, are they printing money through debt? Mostly no. The Trade Desk signed a $750 million credit line with JP Morgan in April, secured on almost everything it owns. It hasn't said what the money is for. In November, it reported its first revenue decline since listing in 2016

Yahoo, owned by Apollo Funds, refinanced $1.6 billion in May at junk bond rates. And Moody's warned that AI will change the digital media landscape. Teads ended September with $657 million of debt and $90 million in cash. And it's talking to its lenders. Here's the pattern. The independents borrow to survive. The platforms borrow to build. And the platforms' AI products are part of why the independents are struggling. Now, the part that doesn't show up as dead at all.

Meta's Hyperion campus in Louisiana was financed with a $27.3 billion bond issued by a joint venture, not by Meta. It priced at par in October 2025 By September this year, it had fallen to 94.4 cents on the dollar. On the same day, Meta's stock jumped 9%.

Equity investors loved the AI story. Bond investors wanted more money to lend against the buildings. A second deal for a campus in El Paso paid 7.5% and drew thin demand. And Moody's counts $1.2 trillion of data center leases at the six biggest builders, with $820 billion of that not yet on their balance sheets.

The demand signals are shifting too. Torreson Slock, the chief economist at Apollo, tracks how many times over hyperscaler bond deals are covered by orders. That ratio fell from about five times in February to under two times in July. Amazon's $25 billion sale that month drew the weakest order book of this whole AI wave. Investors haven't stopped lending. They've started charging more and asking harder questions. At the riskier end sit the Neo Clouds, companies that rent out chips and borrow against them. Nscale, which filed a list in New York, lost just over a billion dollars in six months on revenue of $140 million. Two customers, Microsoft and Anthropic, make up about 85 percent of its $103 billion of contracts. Its earliest loans cost up to 15 percent. And then there's NVIDIA. In Nscale's case, NVIDIA is an investor, a guarantor, a customer, a lender, and the supplier of the chips, all five at once. In August, NVIDIA also guaranteed up to $105 billion for an open AI campus in Ohio.

When a supplier finances its own customers, its sales figures get harder to read.

Oracle is the stress case. S&P cut it to the lowest investment grade rating in July, forecasting a cash shortfall of nearly $42 billion this fiscal year. Its bondholders are suing, led by a Carpenters Pension Fund in Ohio.

So is any of this creating value? Yes, in places. Anthropix Revenue reached almost $1.7 billion in a single quarter, about 12 times a year earlier, though it still made an operating loss. Google Cloud doubled. And the cost of using AI has collapsed. Running a million tokens through a model at the level of GPT 3.5 cost $20 in late 2022, and about $0.07 two years later. For users, that's an enormous gain. But the returns at companies using AI are thin. In one survey of about 6,000 executives, 89% saw no effect on productivity. And the losses are concrete. People Inc. lost 26% of its core web sessions as AI overviews spread and cut 195 jobs. Small publishers lost 60% of their search traffic over two years. Software stocks fell into a bare market. Prices are falling fast for AI models too, which is great for buyers and hard on the margins of the companies that finance the capacity.