5 questions to ask your AI stocks when borrowing gets expensive

Hey {{first_name|Investor}} -

Higher rates expose the AI companies that have to raise more money before their investments start paying for themselves. Size helps, but it doesn't settle the question: this week I found one of the largest companies in the world raising about $70 billion from outside investors in a single quarter, and a small patent licensor whose loan resets with SOFR.

On October 7 the 10-year Treasury yield traded as high as about 5.36%, its highest level since 2002 (NBC News; Forbes, October 7). That was the high during trading. It ended the week at 5.24% (Forbes, October 9).

The 10-year yield is what the US government pays to borrow for 10 years. Lenders start from it when they price longer-term loans to companies, so when it moves, the cost of new long-term borrowing moves with it.

The reasons people point to: inflation that won't settle, oil pressure from the Middle East, and a Fed that raised its target range to 3.75% to 4% on September 16 (Federal Reserve). Most officials expect one more quarter-point increase, depending on the data (Reuters; FOMC minutes). There's also a lot of new borrowing. Goldman Sachs strategist Amanda Lynam estimated $489 billion of AI-related debt had been issued this year by late July, above Goldman's $322 billion estimate for all of 2025 (Yahoo Finance, July 23). That's one bank's estimate of a category that's hard to define, and I wouldn't read it as the cause of the bond selloff.

Employers added just 29,000 jobs in September (Forbes). Weak hiring usually pulls yields down. This time it didn't.

"Rates went up" covers 3 different things, and they move separately.

The 10-year yield. The benchmark for longer-term, fixed-rate borrowing. A higher 10-year raises the cost of a company's next fixed-rate bond. Bonds it already sold keep their rate.

Short-term rates like SOFR. Floating-rate loans are priced off these, and they follow the Fed. When short-term benchmarks rise, unhedged floating-rate borrowing generally gets more expensive as its rate resets.

The credit spread. The extra interest a lender demands from a particular company, on top of the government rate. A strong company pays a small spread. A risky one pays a big one, and spreads can widen even when Treasury yields fall.

CoreWeave ($CRWV) shows what this looks like at scale. It rents out AI computing power and builds data centers to do it. It carried about $35 billion of debt at June 30, and it sold senior notes this year with coupons of 9.75% and 9.625% (CoreWeave Q2 filing). Net interest expense was $640 million in the second quarter alone (CoreWeave Q2 release).

The question I keep coming back to is simple. Does a company generate enough cash to pay for the spending it has planned? Most of that spending is CapEx, meaning buildings and equipment: data centers, chip plants, servers.

Companies that can fund their buildout from their own cash face less immediate financing pressure. They still need the investment to earn an adequate return.

Companies that can't fund it have to fill the gap from somewhere: cash already in the bank, customers paying in advance, partners, government incentives, new debt or new shares. Higher rates make the last 2 more expensive, and that's where the pressure lands.

Customers are a separate question, and an important one. A supplier selling to the biggest tech companies knows its customers can pay. That doesn't mean they'll keep ordering. A wealthy customer can still cancel, postpone or renegotiate. So I ask about both: ability to pay, and willingness to keep spending.

A lot of AI companies borrow with convertible bonds, and their interest rates can look tiny, sometimes under 1%. The lender accepts low interest in exchange for the right to swap the bond for shares.

If the bond is still outstanding when it comes due, the company has to repay it. If the lender converts it first, the company settles in shares, cash or both, depending on the contract. So a low interest bill can come with a big cash repayment later, or with dilution, meaning existing shareholders end up owning a smaller percentage of the company. There are also fees, and some companies spend extra on hedges to limit that dilution.

1. Does the business generate enough cash to fund its planned spending? If not, what fills the gap? Compare cash from operations with CapEx, then look at what covers the difference: existing cash, customer advances, partners, incentives, debt or shares.

2. How much debt can get more expensive in the next 2 years? Separate floating-rate loans, which reset as short-term rates move, from fixed-rate debt coming due, which gets refinanced at whatever rates are then.

3. Can customers pay, and how firmly have they committed? Look at how concentrated the customer list is, whether contracts allow cancellation, and what has to be delivered before the money arrives.

4. Will the investment earn enough to justify its financing cost and risk? This means the return on the money invested, which is a different number from a profit margin. For a data center, that return depends on power bills, maintenance, capacity sitting idle, and replacing older chips as newer ones arrive.

5. Can the company finish what it has committed to without an emergency fundraise? If it can't, the options are spending cuts, slower growth or new shares, and you want to know which one it would pick.

The earnings release is where to start. Debt terms and spending commitments usually sit in the notes of the quarterly filing, the 10-Q.

The rest of this issue is for Insiders.

Below the line: 8 names run through the 5 questions. The giant that raised about $70 billion from outside investors in one quarter, the AI cloud funding its buildout 5 different ways, the hardware company whose cash went into inventory and customer invoices, and the patent licensor whose loan follows SOFR. Plus a table for the rest.