What Microsoft, Alphabet, Amazon, Meta, Oracle and CoreWeave pay to borrow, what that price says about risk, and why interest rates matter more than you think
Figures: bond pricing terms filed with the SEC, Treasury yields for November 17, 2026, and Meta's Form 10-Q for the quarter ended September 30, 2026. Education, not investment advice.
The companies building AI are borrowing more than ever, and some of their bonds now pay eye-catching rates. A yield above 15% for about six years can sound close to risk-free. It isn't. A bond's yield is the market's price for its risk, and the higher the yield, the more risk you're being paid to carry.
Here are the latest US-dollar bond deals from five of the biggest builders, with the coupons and the "spread": the extra yield over a US Treasury bond of the same length.
| Company | Priced | Size | Coupons (first to last maturity) | Spread over Treasuries |
|---|---|---|---|---|
| Microsoft | Oct 29, 2026 | $25B | 4.85% (2028) to 6.35% (2066) | +0.25 to +1.05 points |
| Alphabet | Aug 6, 2026 | $25B | 4.50% (2028) to 6.50% (2066) | +0.33 to +1.30 |
| Amazon | Jul 7, 2026 | $25B | 4.60% (2029) to 6.25% (2066) | +0.40 to +1.25 |
| Meta | Nov 3, 2026 | $15B | 5.85% (2031) to 7.50% (2056) | +1.10 to +2.20 |
| Oracle | Nov 2, 2026 | $10B | 7.00% (2029) to 8.65% (2046) | +2.40 to +3.35 |
For comparison, on November 17, 2026 the 10-year Treasury yielded 5.10% and the 30-year 5.46%.
None of these investment-grade borrowers pays anywhere near 15%. And the spread is moving: Meta's spreads were +0.50 to +1.10 points in October 2025, +0.53 to +1.47 in April 2026, and +1.10 to +2.20 in November 2026. Lenders are asking more each time.
CoreWeave, which rents out AI computing power, issued 9.625% senior unsecured notes due in 2032. As CNBC reported on November 6, 2026, the bond's yield "has climbed above 16%." When a bond with a 9.6% coupon yields over 16%, its price has fallen well below face value, and the market is pricing in real trouble.
There's a rule of thumb to turn a spread into a default risk. The yearly chance of default is roughly the spread divided by the share of your money you'd lose in a default. The market convention is about 40 cents back on the dollar, so a 60-cent loss.
| Spread over Treasuries | Default risk the price implies |
|---|---|
| +0.65 points (Microsoft, 10-year) | about 10% over 10 years |
| +1.10 points | about 17% over 10 years |
| +2.20 points (Meta, longest) | about 31% over 10 years |
| +3.35 points (Oracle, longest) | about 43% over 10 years |
| about +11 points (CoreWeave, at a 16% yield) | about 67% over 6 years |
The investment-grade rows overstate the real risk, because spreads also pay for other things, such as how easy the bond is to sell. Under this rule of thumb, CoreWeave's price implies a better-than-even chance of default over the bond's life. That's a price, not a forecast: part of the spread pays for risk and liquidity, not default alone. A 16% yield there is the price of that risk, not a gift.
"Senior" helps, but it doesn't make a bond safe. Senior bondholders are paid before shareholders in a default. But they still usually lose a large part of their money, which is why the market works with a 40% recovery.
For an investment-grade bond, the bigger danger is usually rates, not default. When market interest rates rise, the price of an existing fixed-rate bond falls, so that its yield matches new bonds.
| If market yields rise 1 percentage point | Price change |
|---|---|
| A 10-year bond paying 6% | about −7% |
| A 30-year bond paying 6.4% | about −12% |
That's more than a year's interest gone. If you hold to maturity and the company pays, you still get every coupon and your money back. But if you need to sell early, the price you get depends on rates that day.
Meta shows why the borrowing is happening. Its free cash flow (cash from operations minus capital spending) has dropped fast as it builds data centers:
| Meta ($B) | Cash from operations | Capital spending | Free cash flow |
|---|---|---|---|
| Q4 2025 | 36.2 | 21.4 | 14.8 |
| Q1 2026 | 32.2 | 19.0 | 13.2 |
| Q2 2026 | 31.9 | 30.1 | 1.7 |
| Q3 2026 | 32.8 | 33.9 | −1.2 |
After about $1.0 billion of finance-lease payments, Q3 free cash flow was −$2.1 billion. At September 30, 2026, Meta held $82.4 billion of cash and marketable securities against $83.7 billion of debt, so it had slightly more debt than cash. It bought back no stock in the first nine months of 2026.
Since then, Meta has raised money in three ways:
In Capital Copilot:
You're the pilot: decide how much of each risk you want to carry before you reach for the yield.
Education, not investment advice. We don't say whether to buy, sell or hold any security. Bond terms are from pricing term sheets (free writing prospectuses) filed with the SEC by each company, and CoreWeave's Form 8-K of June 18, 2026. Treasury yields are from the US Treasury's daily par yield curve for November 17, 2026. Meta figures are from its Form 10-Q for the quarter ended September 30, 2026 and its Forms 8-K of October and November 2026. The CoreWeave yield quote is from CNBC, November 6, 2026. Default-risk figures are our estimates from the rule of thumb above.
Capital Copilot turns market research into audio you can listen to. Free plan, no time limit.
Free beginner course, lesson 3: what a share is, common and preferred stock, dividends, splits and buybacks
What the filings show about cloud revenue, capital spending and backlog at Google, Amazon and Microsoft, and the signals to watch
What return on equity and return on capital really measure, and the two numbers that decide your return
Three checks on the $1.64 payout, what Nike's own plan says, and what to watch next
Briefings, reports and audio TL;DRs — free, with no time limit.
Comments
Loading comments