AI Has a $12.9 Billion Interest Bill

Amazon, Alphabet, Meta and Oracle can find the money. That does not automatically mean shareholders will earn a good return on it.

The Bill Before the Payoff

The AI buildout is usually discussed in chips, models and data-center capacity. The bond market tells the same story in a language investors should not ignore: fixed claims on future cash flow.

Across 12 dollar-denominated fixed-rate bond deals, Amazon, Alphabet, Meta and Oracle raised $244 billion. Their annual cash coupons total $12.864 billion—before another server rack produces another dollar of revenue.

$12.864B per year — The annual coupon bill on the bonds already issued. It gets paid before equity holders debate whether the AI return on capital lived up to the slide deck.

This is not a solvency alarm. These companies can access capital. The sharper question is whether the return on that capital will comfortably clear a much higher hurdle rate.

Rates Explain Most—But Not Everything

Using issue yields rather than cash coupons produces a $12.939 billion annualized comparison measure. About 83.2% came from Treasury benchmarks and 16.8% from company spreads.

So yes, the risk-free rate is doing most of the damage. But the extra spread investors demanded also climbed as these companies returned to market:

• Alphabet: 47 → 62 → 63 → 85 bp

• Amazon: 55 → 75 → 80 bp

• Meta: 78 → 90 bp

• Oracle: 97 → 105 → 145 bp

That is not a clean AI credit-risk premium. Deal size, maturity, liquidity and new-issue concessions all matter. Still, the direction is useful: repeated access to capital is becoming more expensive at the margin.

Equity investors love optionality. Bondholders prefer contractual payments. When companies fund an uncertain payoff with fixed obligations, the bondholders do not need AI to exceed expectations. Shareholders do.

Every additional basis point on a fresh $244 billion would represent about $24.4 million of annual interest. A rate cut can help the Treasury component, but it does not guarantee tighter company spreads—or erase the coupons already locked in.

This is where the AI story becomes a return-on-invested-capital story. The demand can be real, the infrastructure can be necessary, and the equity can still disappoint if financing and depreciation absorb too much of the economics.

Hans's Take

The AI boom does not have a funding problem. It has a hurdle-rate problem. The winners will not simply be the companies spending the most—they will be the ones that convert expensive infrastructure into durable free cash flow.

Portfolio Read-Through

• Favor balance-sheet quality. The bond market is already separating AA borrowers from BBB borrowers.

• Watch spreads alongside yields. Lower Treasuries do not guarantee cheaper all-in funding.

• Separate demand from economics. Data-center demand can accelerate while individual equity returns disappoint.

• Demand a margin of safety. Long-duration AI winners remain sensitive to financing costs and execution.

Source

Analysis adapted from Haircut's review of bond pricing sheets and final supplements for Amazon, Alphabet, Meta and Oracle (source thread). Figures cover the dollar-denominated fixed-rate deals included in that review, not every form of company borrowing.

Trade smart, stay hedged.

Hans Albrecht • Options With Hans

Educational content only. Not financial advice. Options involve risk and are not suitable for every investor.