GM. This is Milk Road Stocks, the newsletter that upgrades your market IQ in 5-minute daily doses.
Today we’re talking hyperscaler CapEx: who’s spending what, the risks that come with such large expenditures, and where the new opportunities lie…
First, a quick detour.
Incogni helps remove your sensitive information from all broker types, including those tricky People Search Sites. Get 55% off Incogni using code MILKROAD.
CAN BIG TECH PAY ITS AI BILL? 💸
Amazon is in talks to sell ~$8B worth of Nvidia chips to outside investors... and then rent them straight back.
The chips would sit inside an SPV (a separate company set up to own them and borrow against them), while Amazon keeps running them in its own data centers.
That takes $8B off Amazon's books in a year where it expects to spend ~$220B on CapEx.
And they’re not alone…
Many of the richest companies in the world are now borrowing to build AI - Big Tech has sold roughly $220B of bonds so far this year, already more than it sold in all of 2025.
And the build those bonds are paying for is huge...
Apollo put hyperscaler CapEx (think: Amazon, Microsoft, Alphabet, Meta and Oracle) at 1.4% of U.S. GDP last year.
That's already above the 1.2% peak of the late-1990s telecom boom, with Wall Street's forecasts showing it will more than doudle to ~3% by 2027. 👇
Problem is, spending like that eats through cash fast.
Free cash flow (the cash left over once the build is paid for) at those five companies fell from ~$170B a year in early 2024 to ~$35B by mid-2026.
Consensus forecasts compiled by Morgan Stanley's Counterpoint Global have it bottoming at roughly -$265B in Q3 2027, before rebounding to ~$505B by 2030.
Microsoft is the only one of the five expected to stay cash-positive the whole way through.
And lenders have noticed. By late July, 78 of the 91 hyperscaler bonds sold this year were trading at higher yields than at launch, meaning investors wanted more interest to keep lending.
But Michael Mauboussin (co-author of that report) argues burning cash is fine, as long as each dollar spent earns more than it costs to raise.
His example is Walmart, which ran negative free cash flow for 14 straight years from 1973 to 1986. Its stock returned 33% a year over that stretch (about 3x the S&P 500).
The hyperscalers clear that bar for now. Returns on their new investment are forecast to bottom at ~23% in 2027, against a cost of capital of ~8%.
I.e. Borrow at 8%, earn 23%.
And the demand is there to back it up. As of June, Microsoft, Google Cloud and AWS were sitting on ~$1.7T of signed contracts they hadn't delivered yet.
Amazon is even raising the price it charges customers to reserve Nvidia chips on AWS by ~15%.
The catch is who signed a lot of those contracts...
Microsoft's backlog grew 84% in a year, but only 25% once you strip out OpenAI.
Our analyst Vincent pointed out that frontier labs (OpenAI, Anthropic and co.) account for roughly half of the AI buildout planned for 2027, even as their share of total AI usage shrinks.
If their revenue growth slows, he expects them to commit to less compute.
He's watching the lab IPOs for the first look at their gross margins, compute costs and cash burn - and if those disappoint, he expects big price swings across AI infrastructure stocks.
Last week he trimmed some of his AI infrastructure names (Bloom Energy, Corning and Infineon among them) and moved the money into what he calls AI adoption plays.
Long story longer: Big Tech's AI bill is enormous, it's being paid with borrowed money through 2027, and the returns say it's worth it... for now.
Here's what decides whether it stays that way:
- Q3 cloud earnings at the end of this month will show us whether that $1.7T backlog is still turning into solid revenue.
- Hyperscaler bond yield will tell us how risky lenders think these loans are getting.
- The AI lab IPO filings will give us the first proper look at their margins and cash burn.
Btw - Vincent's Bloom Energy trim locked in a ~100% gain on that position.
He doesn't think the next 3-5x is in the AI infrastructure names the market already knows, and he's laid out where he's putting the money instead inside Milk Road PRO.
If you want to see what he’s buying, try Milk Road PRO for a buck for 7 days.
KEEP YOUR SSN OFF THE DARK WEB
Every day, data brokers profit from your sensitive info—phone number, DOB, SSN—selling it to the highest bidder. What happens then? Best case: companies target you with ads. Worst case: scammers and identity thieves breach those brokers, leaving your data vulnerable or on the dark web. It's time you check out Incogni. It scrubs your personal data from the web, confronting the world’s data brokers on your behalf. And unlike other services, Incogni helps remove your sensitive information from all broker types, including those tricky People Search Sites.
Help protect yourself from identity theft, spam calls, and health insurers raising your rates. Plus, just for MILK ROAD readers: Get 55% off Incogni using code MILKROAD
