Five Rules for Treating AI Infrastructure Stocks as Tactical, Not Core
Five Rules for Treating AI Infrastructure Stocks as Tactical, Not Core
Five Rules for Treating AI Infrastructure Stocks as Tactical, Not Core
Debt-to-equity across the five AI infrastructure names runs from 31% (Coherent) to 387% (CoreWeave). Treat them all with the same position size, and the weakest link ends up holding the most weight in your book. Here are the five rules I run them through.
1. Separate core holds from tactical bets explicitly
Nvidia and TSMC are core. Coherent, CoreWeave, Nebius, Iren, and Applied Digital are tactical. Mixing them in your sizing means your highest-risk names get core-sized exposure. I keep the combined weight of these five inside a low-to-mid double-digit percentage of the portfolio, no more.
2. Make D/E the first filter
One-line rule: if D/E crosses 100%, tighten the macro scenario by one notch.
| Company | D/E |
|---|---|
| Coherent | 31.1% |
| Nebius | 108.3% |
| Applied Digital | 110.1% |
| Iren | 148.8% |
| CoreWeave | 387.6% |
CoreWeave's 387.6% alone is enough to disqualify it as a core hold.
3. If operating cash flow is negative, price in dilution upfront
All five have negative OCF. They can't pay their own bills from operations. That leaves two paths — issue more debt or issue more shares. Both are bad for existing shareholders. My base case is mid-to-high single-digit annual dilution for this group.
4. Pre-set position sizes by macro scenario
Build three macro paths — rates higher, sideways, lower — and pre-commit a size for each name in each path. Deciding after an event is too late. CoreWeave-style leverage only gets full size in a rate-cut path.
5. Separate "story" from "business"
CoreWeave with negative gross margin isn't a business yet — it's a story. Stories are fine to bet on, but only at "story size." Businesses (think Coherent — profitable, low leverage) can carry "business size." Treating them the same is how your largest exposure ends up on the most fragile name.
FAQ
Q: Should I just own Coherent? A: It's the safest, but its 21.4% growth is the lowest in the group. If you want growth exposure, pairing in a small Nebius position is reasonable. The "single safe name" isn't always optimal.
Q: Is Iren's 148.8% debt too risky? A: On its own, yes — but it pairs with Nvidia's $2.1B right to invest and a PEG of 1.44, the cheapest in the group. "Cheap risk" and "expensive risk" are different trades.
Q: Can Nebius keep growing at 506%? A: Most of that is the base effect — growth off a small revenue base. It will likely normalize to double digits over the next 12-24 months. That normalization itself is the biggest risk to the stock.
Q: Should I buy all five at once? A: All five share the same macro exposure (rates, hyperscaler capex), so diversification within the group is weaker than it looks. Building in tranches against scenario triggers beats a single full-size entry.
More in this Category
Coherent Wins — A Six-Round Scorecard for Five AI Infrastructure Stocks
Coherent Wins — A Six-Round Scorecard for Five AI Infrastructure Stocks
I scored Coherent (COHR), CoreWeave (CRWV), Nebius (NBIS), Iren (IREN), and Applied Digital (APLD) across six rounds. Coherent took it with 10 points, driven by the only debt-to-equity ratio under 32%.
Memory Sold Out Through 2027: Why Micron Now Prices Like a Utility
Memory Sold Out Through 2027: Why Micron Now Prices Like a Utility
Micron's HBM lines are effectively sold out through 2027, and that supply-demand gap is flowing directly into quarterly margins. Why waiting for a $480 pullback beats chasing the vertical line, and the three scenarios that would actually break the thesis.
The Korean Memory Trio: Three US-Listed Routes Compared (MU vs DRAM ETF vs EWY)
The Korean Memory Trio: Three US-Listed Routes Compared (MU vs DRAM ETF vs EWY)
The real suppliers of AI memory sit in South Korea. Three US-listed ways to capture that exposure — Micron, the DRAM ETF, and EWY — compared on volatility, valuation, and entry zones to help fit each layer into a portfolio.
Previous Posts
Palantir Isn't a SaaS Company — It's Infrastructure
Palantir Isn't a SaaS Company — It's Infrastructure
Classify Palantir as SaaS and the valuation looks insane next to a 19% YTD drop. But +85% revenue, a 145% Rule of 40, and 150% net retention say this isn't software — it's industrial-grade infrastructure.
Palantir Just Printed Record Numbers — Why Did the Market Yawn?
Palantir Just Printed Record Numbers — Why Did the Market Yawn?
Palantir posted $1.63B in quarterly revenue (+85% YoY), 60% operating margin, 53% net margin, US business +104%, and a 145% Rule of 40 — all in one quarter. The stock is still down ~19% on the year.
Palantir Is Down 19% This Year — Should You Have Sold?
Palantir Is Down 19% This Year — Should You Have Sold?
Palantir is down ~19% YTD even as it printed its best quarter ever. Here's how I think through whether to hold, sell, or add — and why the real problem is usually anchor, not analysis.