PACK NO. 08
Deep pockets
The clouds that can carry a trillion in spending.
MicrosoftMicrosoftMSFT · 45%
AlphabetAlphabetGOOGL · 35%
Amazon.comAmazon.comAMZN · 20%
Interest down 14% this week117k Wikipedia reads about these companies
Tokenized stocks that track the real share price, bought into your own wallet. They can lose value.
Where the idea came from
Hyperscaler capital expenditures came in on trend in Q1 2026, continuing the trajectory that projects them spending $770 billion this year and over a trillion dollars in 2027.
Their post started this thesis. They didn’t pick these holdings, and they aren’t endorsing the pack.
The idea
Spending a trillion is the easy part
Announced capex becomes depreciation on somebody's income statement for years afterwards. The question is not who can spend it, but who can carry it.
Why these three
- Microsoft
MicrosoftDeepest profit pool · 45%
The largest established software earnings base to absorb depreciation against, plus a rental business that converts the assets into revenue directly.
- Alphabet
AlphabetAdvertising cushion · 35%
Search advertising is a profit pool almost unrelated to the buildout, which is precisely the cushion this claim is about.
- Amazon.com
Amazon.comThe hardest case · 20%
Included deliberately as the thinnest-margin operator, which is where this claim would fail first if it is wrong.
The other side
This is an argument for size, and size is already priced. If the assets have shorter useful lives than assumed, a large profit pool delays the damage rather than preventing it — and every holding here is exposed to exactly the same mistake at the same time.
What would change the call
A major operator shortening the assumed useful life of its AI hardware, or writing down datacenter assets.
Record: these holdings at these weights, held from Sep 30, 2025 to Sep 29, 2026, from daily closing prices, before fees. “Live” is the thesis’s public record since its call opened on Sep 25, 2026. “Interest” is human page views of each company’s English Wikipedia article, the last 7 days against the 7 before. Past performance isn’t a promise.
Sources and version history