AI Demand Isn't Cracking. The Borrowing Behind It Is.

In one session, CoreWeave's bonds were priced like distressed debt, Oracle had its worst day in months, and the chip index fell 4%. The same week, ASML said customers are ordering machines for 2030. Those two things don't contradict each other. They describe two different ledgers, and investors are only now starting to read them separately.

HeyTheo Research · Thursday, October 15, 2026

Quick Read

• On Wednesday, Oct 14, CoreWeave (CRWV) fell about 15% after its bonds due 2032 hit a 15% yield. Oracle (ORCL) fell nearly 9%. The AI selloff was led by companies that borrow to build.

• The orders behind AI look as strong as ever. ASML (ASML) beat on profit and said customers are booking through 2030. TSMC (TSM) grew sales 50% last quarter.

• The pressure is on the funding side. CoreWeave's debt rose from about $15 billion to $52 billion in a year, and its quarterly interest bill more than doubled.

• A 10-year Treasury yield above 5% raises the cost of every borrowed dollar. That hits debt-funded builders first and cash-funded giants last.

What happened on Oct 14

The day started with a hot inflation report and ended with the AI trade's weakest links exposed.

Source: Yahoo Finance, CNBC, investingLive, market data as of the Oct 14 close

A 15% yield on a bond is the price lenders charge when they are not sure they will be paid back in full. A US government bond of similar length pays about 5.2%. The gap, roughly 9 points, is what lenders demand for the risk of lending to CoreWeave.

Selling spread to every company in the same position: Nebius (NBIS), the data center operators that converted from crypto mining, and Oracle. All of them finance AI capacity mostly with debt.

The demand ledger looks fine

If AI demand were cracking, the companies at the start of the supply chain would see it first. They are not.

ASML, which makes the machines that print advanced chips, reported a third-quarter profit of €5.0 billion, well above the €4.1 billion analysts expected. It guided fourth-quarter sales to €14.2 billion to €15.2 billion and said customers expect the industry to stay short of supply through at least 2028. Some are already placing orders for 2029 and 2030.

TSMC, which makes most of the world's advanced AI chips, reported third-quarter revenue of about $46.7 billion, up 50% from a year earlier.

And Microsoft (MSFT) disclosed $329 billion of data center lease commitments that have not started yet, up about 67% in one quarter.

Those are orders, contracts and leases. They do not look like a boom running out of buyers.

Research cue · try it in HeyTheo

Ask Theo for the bull case, bear case and what to watch on any AI stock you hold, and add: "how is this company paying for its AI spending: cash flow, bonds or loans?" The answer sorts your holdings by which ledger they sit on. Open HeyTheo

Start with: search CRWV, then ask the funding question above.

The funding ledger is where the strain is

CoreWeave rents out Nvidia-powered computing to AI labs. It buys the chips largely with borrowed money and pays the loans back from rental income.

That works while borrowing is cheap and customers keep paying. Here is how the numbers have moved.

Source: CoreWeave filings via Yahoo Finance, S&P Global Ratings

Revenue doubled. Interest expense grew faster. The interest bill now takes about 25 cents of every revenue dollar. The company's chief executive called the selloff the market "acting irrationally." Lenders are not acting irrationally when they ask for 14.5% from a company with that profile. They are pricing it.

Oracle is a larger, investment-grade version of the same question. Its total debt is close to $167 billion. S&P Global cut its rating to BBB-, one step above junk, and expects a free cash flow shortfall of nearly $42 billion in the coming fiscal year as Oracle spends $90 billion to $95 billion on data centers. Roughly half of Oracle's $638 billion contract backlog is tied to one customer, OpenAI, whose payments are scheduled to start in 2027.

Research cue · try it in HeyTheo

Build two baskets. Cash-funded builders: the large cloud companies that pay for data centers mostly from operating cash. Debt-funded builders: the newer AI cloud operators and Oracle. Track them against each other for the next month. If the gap keeps widening, the market is pricing credit, not demand. Open HeyTheo

Start with: search MSFT for the first basket and CRWV for the second.

Rather read first? Our free daily brief lands in your inbox before the open. Send it to me

Why rates turn this from a story into a problem

The biggest AI spenders are borrowing too. Alphabet (GOOGL), Amazon (AMZN), Meta (META) and Oracle had issued more than $228 billion of bonds this year through Oct 13, more than double their total for all of 2025. JPMorgan estimates hyperscaler spending at about $800 billion this year, rising above $1 trillion a year by 2027.

The difference is the price. A company rated AA borrows at a small premium to Treasuries. A company rated B+ borrows at nearly three times the Treasury rate. When the 10-year yield sits above 5%, as it has since September's inflation report, that premium is charged on top of an already high base.

That is the link to the macro picture. As we laid out in our guide to reading inflation this fall, a hot core reading pushes bond yields up. Higher yields make every borrowed data center more expensive, and the companies with the thinnest margins feel it first.

When a credit problem becomes a demand problem

The two ledgers are connected. Debt-funded operators are real buyers of AI chips and servers. If they cannot borrow, they buy fewer.

So the line to watch is whether funding stress shows up in orders. Three places it would appear first:

• Neocloud capacity plans. A delay or cut in planned data center capacity from CoreWeave or its peers.

• Chip-backed lending. CoreWeave has said it will keep borrowing against its chips and wants insurers as lenders. Higher rates on that debt, or a failed deal, would be an early warning.

• Customer concentration. OpenAI is a major customer for Oracle and CoreWeave. Any change in its commitments travels through both.

Memory is part of the same chain. As we covered in our note on CXMT and Micron, memory makers are earning record margins because AI servers absorb every advanced chip they can make. That holds only while the buyers can pay for the servers.

Research cue · try it in HeyTheo Set a trigger on any AI stock you hold: "tell me if its bonds or its stock fall more than 10% in a week." Then use the backtester to see how the stock traded the last time 10-year yields rose half a point in a month. It shows how sensitive your holding is to borrowing costs. Open HeyTheo

Start with: search the AI stock you hold, then set the 10% trigger.

Research it yourself: rules HeyTheo tracks

• Cash vs debt builders. Two baskets of AI infrastructure names, split by how they pay for capacity, tracked against each other.

• Interest burden trigger. Flags any holding whose interest expense grows faster than its revenue for two quarters in a row.

• Rating change trigger. Flags any holding whose credit rating is cut, or moved to negative outlook.

• Ask Theo. Try: "Which of my holdings depend most on borrowed money to grow?" or "Give me the bear case for AI cloud stocks if the 10-year yield stays above 5%."

Check the rule behind any trigger before acting on it. You trade through your own broker; HeyTheo helps you decide.

The Two Ledgers

The AI trade is now two trades. One is demand: chips, machines and leases, and it still reads strong. The other is funding: who pays for the build-out, at what rate, and with whose money.

On Oct 14 the market sold the second and only grazed the first. The question for any AI holding is which ledger it sits on. Read its interest bill next to its revenue, check its rating, and watch what lenders charge it. Those numbers move before the orders do.

Start your own first pass. Download HeyTheo, add the AI names you hold, and run the three cues above to see which ledger each one sits on. Get the app

Reading on a laptop, or not ready to install? Get the free daily brief by email, and open the app on your phone when you are ready. Get the daily brief

Sources

ASML, TSMC, Microsoft, CoreWeave, S&P Global Ratings, LSEG, JPMorgan, CNBC, Bloomberg, Yahoo Finance, investingLive, 24/7 Wall St., BigGo Finance, Startup Fortune, BabyPips

Disclaimer

Disclaimer: HeyTheo is a research and education platform, not an investment adviser or broker-dealer. Nothing here is advice to buy, sell, or hold any security. You trade through your own broker; HeyTheo helps you decide. Backtested results are hypothetical and do not guarantee future returns. References to governments, officials, or policies are for market context only and are not political endorsements. All investing involves risk, including loss of principal. Data is as of the dates noted.