Meta Fell 7%. Your Mortgage Rate Fell Too. Here's the Link.
Meta beat Wall Street's forecasts and still lost a big slice of its value in a day, because its AI building spree took more cash than its business made. The money that left AI stocks ran into Treasury bonds, and the mortgage rate, which follows them, posted its first weekly fall since August.
-$2.12 billionMeta's free cash flow in its third quarter: it spent $34.91 billion on data centers and gear, more than the $32.79 billion its business produced
4.96%the 10-year Treasury yield on October 22, its first close under 5% in a month
7.32%the 30-year mortgage rate on October 22, down from 7.40%: the first weekly drop since August 20
Figures as of October 23, 2026: Meta's and Tesla's third-quarter results (October 21), Alphabet's second-quarter results, the Treasury's daily yield curve to October 23 and Freddie Mac's survey of October 22. Freddie Mac's next survey and the Fed's decision are due after this was written.
On Thursday, October 22, 2026, Meta’s shares fell 7.0%, even though its results beat analysts’ expectations1. The Nasdaq Composite lost 2.60% that day, its steepest daily drop since early June2. The same day, the average 30-year mortgage rate fell to 7.32% from 7.40%, its first weekly drop since August 203.
A falling tech stock and a falling mortgage rate look unrelated. They are the same money moving.
Meta made more money and ran out of cash
Meta’s revenue was $64.60 billion in the third quarter, up 26% on a year earlier, and its net income was $9.86 billion4. Its business produced $32.79 billion of cash from operations, while capital spending, mostly data centers and gear for AI, was $34.91 billion5. Free cash flow was negative $2.12 billion: more cash went out than came in5. Reuters counted it as Meta’s first cash outflow in more than a decade1.
Meta, July-September 2026, $ billion
Source: Meta Platforms, Q3 2026 results (Form 8-K), Oct 21, 2026 5
Follow $100 of that revenue. About $51 came back as cash from the business, and about $54 went into data centers and gear, so Meta ended about $3 short for every $100 it took in6. Put another way, it spent about $106 for every $100 of cash its business made6.
The spending is not slowing. Meta expects capital spending of $135-145 billion in 20267. Its release also announced a new non-voting class of shares under the heading “Creating a New Class of Shares for Capital Raising”5. Chief executive Mark Zuckerberg said: “Our investments are paying off, the core business is strong”5.
Meta is not alone. In its second quarter, Alphabet spent $44.9 billion on property and equipment against $39.1 billion of cash from operations, so its free cash flow was negative $5.9 billion8. That is about $115 spent for every $100 of cash in9. In June, Alphabet raised $49.6 billion of new stock, to be used partly for AI infrastructure, and set up a program to sell up to $40 billion more10.
When the cash from the business runs short, the rest comes from investors and lenders. Goldman Sachs estimates that AI-related financing is nearly one-quarter of all U.S. investment-grade bond issuance this year, and much of the lending comes from insurance and retirement money12.
The question lenders ask is whether the revenue will arrive. Gartner says AI data center owners would need $8.5 trillion of revenue through 2030 to cover announced spending, financing costs and a 12% return, and its forecast of AI revenue over that period is no more than $3.6 trillion13.
AI data centers through 2030 (Gartner, Oct 14, 2026), $ trillion
Source: Gartner press release, Oct 14, 2026; Deloitte, Oct 22, 2026 13
On Thursday the money ran to Washington
On October 22, the chip index fell 6.7%1. Lenders started charging AI borrowers more: Deloitte notes that the cost of insuring AI-related debt and the yields on AI bonds rose sharply, with the smaller AI clouds under the most pressure14. Blackstone rose 8.4% after saying wider spreads on AI debt could offer a chance to lock in higher returns1.
The money that left did not disappear. As investors sold AI bonds, Treasury yields fell14. Joe Saluzzi of Themis Trading described investors “rotating out of techs, out of the AIs, and into the more safety trades”15.
A bond pays a fixed coupon, so when more buyers want it, its price rises and its yield falls. The SEC’s investor bulletin says market interest rates and bond prices generally move in opposite directions16. On October 22, the 10-year Treasury yield fell from 5.13% to 4.96%, its biggest one-day fall since August 2024 and its first close under 5% since September 2217.
10-year U.S. Treasury yield, % (daily close)
Source: Federal Reserve via FRED (DGS10); U.S. Treasury par yield curve 17
The October 5 peak of 5.31% was the highest 10-year yield since 200218. The market also pulled back its bets on the Fed: the odds of a rate rise on October 28 fell to 7.2%, about 1 in 14, from 27.7% a week earlier15,19.
Market odds of a Fed rate rise on October 28 (CME FedWatch), %
Source: CME FedWatch, as reported by 24/7 Wall St., Oct 23, 2026 15,19
Your mortgage follows the 10-year
The 10-year Treasury yield acts as a benchmark for other long-term interest rates, and the gap between the mortgage rate and the 10-year is called the mortgage spread, the Boston Fed explains20. Since 2000 that gap has ranged from more than 300 basis points in the financial crisis to less than 100 in 202120. On October 22 it was 2.36 points: 7.32% against 4.96%20.
October 22, 2026, %
Source: U.S. Treasury par yield curve; Freddie Mac PMMS, Oct 22, 2026 17,3,20
Freddie Mac’s survey averages loan applications from the previous Thursday through Wednesday, so the October 22 reading of 7.32% closed before Thursday’s bond move3. The rate had risen every week from August 27 to October 8, when it reached 7.40%, the highest since November 20233.
30-year mortgage rate, Freddie Mac weekly average, %
Source: Freddie Mac, Primary Mortgage Market Survey via FRED (MORTGAGE30US) 3
The effect on a bill is real but small. On a $400,000, 30-year loan, the payment is $2,769.52 at 7.40% and $2,747.72 at 7.32%, about $22 a month less21. It is still about $300 a month more than at a year ago’s 6.19%21.
Monthly payment on a $400,000, 30-year mortgage
Source: Freddie Mac PMMS rates; Economy Studio arithmetic (principal and interest) 3,21
The catch: oil and the Fed pull the other way
The same Thursday, Houthi strikes on Saudi targets sent crude to a five-month high, and investors worried that it could keep inflation high and push the Fed toward more rate rises22. By Friday the oil rally had pushed the 10-year yield back to 5.00%17,22.
Inflation is the reason. Consumer prices were up 3.7% in the year to September, with gasoline up 34.2%23. The Fed raised its target range to 3.75-4.00% on September 16 and said “Inflation remains elevated”24. The job market is not giving way either: 194,000 people filed new jobless claims in the week to October 17, near multi-decade lows25,1.
So two forces pull on the mortgage rate at once: fear about AI debt pushes money into Treasuries and rates down, while oil and inflation push them up.
Britain: the same link, through swaps
In the UK, more than 90% of new mortgages are fixed, and their price follows swap rates set in the financial markets, not the Bank of England’s rate on the day26. When swap rates spike, fixed-rate mortgage prices follow quickly26. Inflation there was 3.5% in September, up from 3.1%, with motor fuels up 29.8% on the year27.
What to watch
None of this says what to do. It shows a chain: when AI companies spend more cash than they make, they borrow, and when lenders get nervous, the money runs to government bonds and the mortgage rate follows.
The Fed decides on Wednesday, October 2824, and Britain’s Autumn Budget is the same day28. Freddie Mac’s next survey, on October 29, will be the first to include Thursday’s bond move3.
The one thing to remember
When the AI money stopped flowing it did not vanish. It ran to the safest place it could find, and the mortgage rate was the first thing to notice.
Educational content only, not financial advice. Charts are drawn from the sources above; the text is written with AI tools and every number is checked against its source before publishing. Spotted a mistake? Tell us on the corrections page.