AI infrastructure, decoded
A free newsletter breaking down the companies building AI: chips, memory, networking and power. Real numbers, clear theses, and the risks most posts leave out.
Subscribe free →Free. No spam. Unsubscribe anytime.
> scan memory --valuation
2yr fwd P/E 4x (10yr median 8x)
read cheap, or pricing a peak?
> macro --rates
10Y yield 5.18% highest since 2007
pressure on high-multiple AI names
> watch CRDO
revenue +115% y/y
fwd P/E 31x vs MRVL 62x
the catch beta 3.2, gap at $206-220
>
Key levels, trend and volume on the names that matter. What the price is actually saying, not what the hype says.
Valuation, growth and analyst targets side by side, so you can tell a great company from a great price.
Every idea comes with the risks and the level where it's wrong. The part most finance posts skip.
Margin of Signal is written by an independent investor who tracks the AI build-out every day, from memory pricing and data-center spending to the charts of the companies riding it.
The goal is simple: show the reasoning, show the data, and be honest about what could go wrong.
Memory was the squeeze of 2026. The constraint is already moving outward from the chip: to packaging, optics, storage and power. Here's the order, the numbers, and what could break it.
High-bandwidth memory is the stacked DRAM sitting next to every AI accelerator. Here's how it works, who makes it, and why it's squeezing the entire memory market.
Spending is still accelerating, but the pricing power is shifting from chips to the bottlenecks: memory and power. Here are the numbers, the thesis, and what could break it.