Advertorial
Market warning

An AI company’s bonds now pay over 15%. In 2007, credit cracked three months before stocks peaked.

The AI boom runs on borrowed money. What the bond market is saying now, what happened the last two times a borrowing boom cracked, and the dates to watch.

David | Senior Analyst @ Exent Research Desk6 min read

What lenders charge

Yield on each borrower’s bonds. The riskier the borrower, the more lenders want.

  • US government (10-year Treasury)5.10%
  • Oracle (bonds due 2032)7.24%
  • CoreWeave (bonds due 2032)15.06%

Treasury: Oct 16 close. Oracle and CoreWeave: Oct 14, as reported. CoreWeave's yield was about 13.4% on Oct 1.

15%+What an AI cloud company’s bonds yieldCoreWeave; its shares −25% this week
−4.5%The Nasdaq this weekIts worst week since June
5.10%10-year Treasury yieldJust off a 24-year high (Oct 7)
−$260BValue lost by AI software stocksReported Oct 16

Lenders decide every day whether they will get their money back. When they start to worry, they charge more. This week they charged one AI company more than 15% a year to borrow, about three times what the US government pays.

One company's bonds are not a crash signal. But in 2007 the first cracks also showed up in one corner of the bond market, three months before stocks peaked. Here is what happened then, what is different now, and the dates to watch.

The AI boom runs on borrowed money. This week, the lenders got nervous.

AI needs data centres, chips and power, and the bills are enormous. The cash the big tech companies earn no longer covers all of it, so more and more of the build-out is paid for with debt. Here is the chain, from the spending to the people who carry the risk.

  1. Spend

    The big cloud firms spend about $800 billion this year, much of it on AI (JPMorgan estimate)

  2. Borrow

    Alphabet, Amazon, Meta and Oracle sold over $228 billion of bonds in 2026, more than double 2025

  3. Stretch

    Smaller AI clouds such as CoreWeave borrow at junk rates. It paid $640 million in interest last quarter

  4. Hold the risk

    Bond funds, insurers, banks and private lenders, and everyone who owns the stocks

CoreWeave’s interest bill, per quarter

An AI cloud that rents out Nvidia chips, rated junk (B+)

  • Apr–Jun 2025$267M
  • Apr–Jun 2026$640M
  • Jul–Sep 2026 (its forecast)$860–940M

Interest expense from company results. Jul–Sep 2026 is the company's own forecast range.

It is not only CoreWeave. The Financial Times reported this week that bonds which paid for big data-centre projects were trading below face value. When lenders get nervous, a company that relies on borrowing has three choices: pay more, build less, or sell shares. Stockholders feel all three.

In 2007, credit cracked first. Stocks hit a record anyway, then fell 57%.

The S&P 500 since 1999, and the last two borrowing booms

Shaded: from the record close to the lowest close that followed. Log scale, so equal drops look equal.

−49%2000–02−57%2007–09Now ???Record, Oct 6, 2026

2007. The borrowing boom was in mortgages. Banks bundled home loans into bonds and sold them around the world. The first cracks showed up in June, inside two Bear Stearns hedge funds that had borrowed heavily to buy those bonds.

Stocks barely noticed. The S&P 500 wobbled over the summer, then closed at a record on October 9. Lenders had been asking for more for months by then: the extra yield on medium-grade company bonds over Treasuries rose from 1.59 points in mid-July to 2.13 by mid-September. Seventeen months after the record, the S&P 500 was down 57%.

  1. Jun 22, 2007

    Bear Stearns commits up to $3.2 billion to rescue one of its mortgage hedge funds

  2. Jul 17, 2007

    Investors are told the two funds are worth almost nothing

  3. Aug 9, 2007

    BNP Paribas freezes three funds it can no longer value

  4. Oct 9, 2007

    The S&P 500 closes at a record: 1,565

  5. Mar 9, 2009

    The S&P 500 bottoms at 677, down 57%

2000. The borrowing boom was in telecom. Companies raced to lay fiber-optic cable for an internet they were sure would explode, and US telecom firms borrowed more than $320 billion between 1999 and 2001.

The internet did change everything. But by April 2001, less than 3% of the long-distance fiber was in use. The loans could not be paid, telecom defaults soared, and the Nasdaq fell 78% from its March 2000 record to its October 2002 low.

The last two borrowing booms, and now

What was borrowed, and what stocks did after the peak

The borrowing boomWhat followed
2000⁠–⁠02Telecom fiber: over $320B borrowed in 1999–2001Nasdaq −78% · S&P 500 −49%
2007⁠–⁠09Mortgages: credit cracked in June and July, stocks peaked Oct 9S&P 500 −57%
NowAI data centres: $228B+ of new bonds in 2026, one AI cloud pays 15%+?

Index moves from the record close to the lowest close that followed, price only.

One borrower in trouble, or the start of a credit crack?

In June 2007 the trouble also looked like a special case: two hedge funds at one bank. Most of the time, one borrower's trouble stays one borrower's trouble. Sometimes it spreads, from the weakest borrower to the next and from bonds to stocks. From the inside, the first weeks of both look the same.

If the market is Risk ON

Credit is calm, money is flowing, the trend is up

  • One borrower’s trouble tends to stay contained
  • Dips in AI stocks tend to get bought
  • Investors stay invested and keep their core positions

If the market is Risk OFF

Lenders pull back, credit cracks, the trend has turned

  • Stress spreads from the weakest borrowers to the rest
  • A dip can turn into a long drawdown, like 2000 or 2007
  • Investors trim what is most exposed and keep cash ready

So, is the market Risk ON or Risk OFF right now? That is the one question most investors only answer in hindsight, after the damage is done.

The next four weeks: the Fed, Big Tech’s earnings and CoreWeave’s results.

  1. Oct 28

    Fed rate decision. Higher rates make new borrowing more expensive

  2. Oct 29

    Q3 GDP and September inflation (PCE)

  3. Late October

    Microsoft, Alphabet, Meta, Apple and Amazon earnings, and their spending plans (expected)

  4. November

    CoreWeave and Nvidia earnings (expected)

What this does not mean.

Is this 2007 all over again?

Not so far. In 2007 the bad loans were mortgages spread across the whole financial system. Today the stress sits with a few AI borrowers.

The wider bond market is still calm. Medium-grade company bonds pay 1.44 points more than Treasuries, close to the lowest since the 1990s. In late 2008 that gap passed 6.

Aren’t AI companies making real money?

Yes. Analysts expect S&P 500 earnings to come in about 30% higher than a year ago this quarter, and the index is only 2% below its record.

The question is not whether AI earns money. It is whether that money arrives fast enough to pay for what was borrowed to build it.

So should I sell my AI stocks?

Not on one headline. CoreWeave says its $104 billion of signed contracts more than covers its debts, and one company’s bonds are not the whole market.

What matters is whether the stress spreads, from one borrower to the next and from bonds to stocks. That is what a daily Risk ON / Risk OFF read is for.

Risk ONRisk OFF

Credit cracks show up before stocks. Market intel reads credit markets every morning.

Credit stress is one of the signals it folds into one Risk ON / Risk OFF call each day, so a turn shows early instead of in hindsight.

Current regime

Updated every morning

Live
Today: members only
Risk ON or OFF???.?
Risk OFFNRisk ON
Illustrative only. Not real data.
  • Liquidity??.?

    How much money is moving through the financial system — the fuel behind every rally.

  • Capital Stress??.?

    Early-warning signs from credit markets — where trouble shows up before it reaches stocks.

  • Cycle??.?

    Where the economy sits in its cycle — growth, inflation and interest-rate momentum.

  • Sentiment??.?

    How fearful or greedy investors are — the extremes that mark tops and bottoms.

What you get

  • A daily Risk ON / Risk OFF call on the S&P 500
  • When to hold and when to step aside, in plain words
  • Daily macro briefings: rates, credit, inflation and the AI trade
  • Exposure guidance for your own portfolio
  • Built from millions of macro data points

Built on institutional‑grade data and leading AI.

The same data sources and models the big desks use, read for you every morning.

  • TradingView
  • FRED
  • Morgan Stanley
  • Goldman Sachs
Exent
  • Claude
  • OpenAI
  • BIS
  • SEC

As seen on Yahoo Finance.

S&P 500 vs Exent Regime performance comparison chart from the article

“Exent AI Beat S&P 500 by 10x Over 8 Years”

Yahoo Finance · June 2, 2026

An eight-year performance analysis of Exent AI’s market regime signal against the S&P 500, as published on Yahoo Finance.

Read full article

Backtested results, as published. Past performance does not predict future returns.

Sources

Advertorial, presented by Exent. This article is for information and education only and is not investment advice, a recommendation to buy or sell any security, or a forecast. Market data is as of the dates shown and changes daily. Company names are used for news context only, not as recommendations. Past market episodes are shown for context; history does not predict future returns. Investing involves risk, including the possible loss of principal. Exent AI is an informational research platform and is not an investment adviser, broker-dealer or fiduciary.