Lower interest rates are supposed to be good news for technology stocks and crypto. But there is a wrinkle in the AI boom: The companies building the infrastructure are increasingly borrowing long-term to buy assets that can become obsolete far faster. That turns the bond market into an increasingly important test of the AI spending thesis.
Lee Sees Easier Money Ahead
Fundstrat’s Tom Lee expects bond yields to normalize over the next six months as inflation pressures ease, arguing that technology stocks and crypto are already signaling easier financial conditions ahead. He said a 10-year Treasury yield below 5% would be "really positive for risk-on."
The equity-market logic is straightforward. Lower yields reduce the cost of borrowing and increase the value investors place on future earnings, potentially giving high-growth technology stocks another tailwind.
But the AI infrastructure boom creates a complication.
The world’s largest technology companies are no longer funding all that growth from their own cash flow. Amazon.com Inc (NASDAQ:AMZN), Microsoft Corp (NASDAQ:MSFT), Alphabet Inc (NASDAQ:GOOGL)(NASDAQ:GOOG), Meta Platforms Inc (NASDAQ:META) and Oracle Corp (NYSE:ORCL) issued roughly $200 billion of investment-grade debt during the first half of 2026, almost twice their issuance during all of 2025, according to iShares.
AI Is Becoming a Bond Trade
Oracle provides perhaps the clearest example. The company said it expects to raise $45 billion to $50 billion in gross cash during 2026 through a combination of debt and equity to expand its cloud infrastructure for customers including Advanced Micro Devices, Inc. (NASDAQ:AMD), Meta, Nvidia Corp. (NASDAQ:NVDA), OpenAI and xAI.
Alphabet has also been spending at extraordinary levels. The company spent $80.6 billion on capital expenditures during the first half of 2026, more than double the $39.6 billion spent during the comparable period a year earlier. Its long-term debt stood at $98.2 billion at the end of June.
The issue isn’t necessarily that this spending is reckless. AI infrastructure can generate revenue for years, and companies such as Alphabet and Oracle have substantial existing businesses supporting their balance sheets.
The question is duration.
Some of the GPUs being purchased for today’s AI buildout can have an economic life of roughly two to three years before newer, more powerful chips make them less competitive for the most demanding workloads. That creates an unusual financing mismatch: A hyperscaler could be paying off a 10-year bond long after the two-year-old hardware it helped finance has been replaced or relegated to less demanding uses.
A data center may be financed with debt lasting a decade or more, while the computing hardware inside it can have a much shorter economic life as AI chips become faster and more efficient.
The Bond Market Wants Proof
That mismatch is becoming harder for investors to ignore. Goldman Sachs expects the five hyperscalers to issue roughly $250 billion of bonds this year and $400 billion in 2027.
At the same time, the 10-year Treasury yield has moved above 5%, raising the cost of capital even for companies with investment-grade credit. Reuters reported that investors are demanding higher yields as hyperscaler borrowing grows.
For equity investors, that creates a crucial test for the AI trade. If rates fall, Lee’s argument gets stronger. But if AI capex keeps rising faster than the cash flows it generates, lower rates may simply make it easier to finance an even larger infrastructure bill.
That makes the next phase of the AI trade less about whether companies can borrow and more about whether the returns from those billions in GPUs, data centers and power infrastructure arrive before the debt comes due.
Investors should watch free cash flow, capex guidance and borrowing costs together—not in isolation.
Tom Lee image created by artificial analysis via DALL-E; and Dow Jones Industrial Average, via Shutterstock
Login to comment