In partnership with
Good morning, capex worriers.
Every time Meta says it's spending more, investors wince.
Last week Meta said $145 billion, and a quiet bank note said "that's actually cheap."
Today we're breaking down why the number that scared everyone is the most bullish thing Meta has done.
Write docs 4x faster. Without hating every second.
Nobody became a developer to write documentation. But the docs still need to get written — PRDs, README updates, architecture decisions, onboarding guides.
Wispr Flow lets you talk through it instead. Speak naturally about what the code does, how it works, and why you built it that way. Flow formats everything into clean, professional text you can paste into Notion, Confluence, or GitHub.
Used by engineering teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works system-wide on Mac, Windows, and iPhone.
A leaked internal memo shows Meta is about to start building its own AI chip, and doubling how much computing power it runs.
Two big moves are hiding in that memo.
Move one: Meta is making its own brain.
The chip is code-named Iris and it goes into production this September.
Until now, Meta rented its "brains" from Nvidia (the company that sells the expensive AI chips everyone fights over).
Iris is Meta's plan to build them in-house instead.
It's not doing it alone.
Meta designed Iris with $AVGO ( ▼ 3.25% ) (the quiet chip firm big tech hires to build custom silicon) and will manufacture it at TSMC (the Taiwanese factory that makes most of the world's advanced chips).
If Broadcom sounds familiar, it should. It's the same partner we flagged when OpenAI started designing its own chip.
One by one, Nvidia's biggest customers are building their way around it.
Move two: Meta is doubling its compute.
Compute (the raw computing power AI runs on, measured in gigawatts of electricity) is Meta's fuel. It runs about 7 gigawatts today. The memo says 14 by 2027.
One gigawatt roughly powers a million homes.
To feed all that, Meta locked in long-term supply deals with Samsung (memory), SanDisk (storage), and Sumitomo (the cables that connect it all).
Here's the wild part: Meta tested Iris in six weeks with no major problems, and plans to ship a brand-new AI chip every six months through 2027.
The rest of the industry ships one about every year and a half.
Meta just tripled the speed limit.
Wall Street keeps scolding Meta for how much it spends.
Start with the number that scared everyone: $145 billion this year on AI.
That's about a fifth of all Big Tech AI spending combined.
It sounds terrifying.
Divide that spending by the new capacity it buys, and each gigawatt costs Meta about $22 billion to build.
The market had been assuming $45 billion. That's roughly half price.
And the reason is Move One.
When you buy Nvidia's chips, you pay Nvidia's markup. When you design your own with Broadcom, you cut that markup out.
The Iris chip is the discount.
Now stack that against a number from two weeks ago.
We showed that a gigawatt of Meta's compute can be rented out for about $20 billion a year, at an 85% profit margin.
Put the two side by side, on Wall Street's own numbers, and the machine appears:
It costs Meta $22 billion to build a gigawatt, once.
That same gigawatt can throw off $20 billion a year, again and again.
For a specific group, this matters a lot: anyone who sold Meta on "capex fear." The fear only makes sense if the spending is a black hole.
Cut the cost per gigawatt in half, and the same spending becomes the cheapest path to up to $150 billion in new revenue.
Meta reports earnings on July 29.
If management puts a real dollar figure on compute sales, capex stops looking like a waste and starts looking like a revenue engine.
Nvidia sells the chips.
Meta owns the infrastructure and monetizes the excess.
Know someone still panicking about Meta's spending? Forward this to them →
Meta building its own chip is...
That’s it for today!
How did you like today’s newsletter?
Second Order is for information and entertainment only and is not financial advice.
We are not licensed financial advisors.
Any tickers or companies mentioned are our opinions, not recommendations to buy or sell.
Do your own research and consult a professional before making investment decisions.