1
AI demand is visible
Orders, backlog, contracts, or management commentary are already pointing at data-center demand.
AI infrastructure picks-and-shovels
The next Bloom trade is a bottleneck trade.
A Leopold-style framework for hunting public-market exposure to whatever becomes scarce if frontier AI scaling continues: power, electrical gear, optical bandwidth, storage, cooling, packaging, and entitled compute sites.
7
hunting grounds where physical scarcity matters
10
highest-signal names ranked for follow-up diligence
$830B
estimated 2026 capex from top cloud and service providers
155 GW
estimated installed data-center power capacity in 2026
Core Logic
The target is not generic AI exposure. The target is scarce, physical, mission-critical capacity that hyperscalers cannot conjure with software.
2
Supply takes years
Capacity cannot be added quickly: transformers, substations, turbines, lasers, packages, and sites.
3
The product is boring
Prefer products that look industrial, cyclical, or component-like before the market fully rewrites the story.
4
The customer needs speed
Microsoft, Oracle, Meta, Amazon, xAI, CoreWeave, and similar buyers are paying for time-to-capacity.
5
Crowding is controlled
The best setup is validated by data but not yet treated like an obvious AI winner.
Bottleneck Map
Click a bottleneck to filter the stock dashboard. The map moves from power supply and electrical delivery into compute fabric, thermal control, and chip assembly.
Interactive Stock Dashboard
Use this as a diligence cockpit. Scores are qualitative and reflect thesis fit, current evidence, and timing/risk as of this research pass.
Live quotes load from the local server.
No matching names. Widen the filters or clear the search.
Ranked Takeaways
The first table is the "not yet fully crowded" research queue. The second list contains strong businesses where the AI-infrastructure angle is already more widely priced.
| Rank | Name | Why It Is Interesting | Main Problem |
|---|
Most Crowded Watchlist
These are real beneficiaries, but the market already treats them as AI-infrastructure winners or close to it.
POWL, NVT, GEV, LITE, COHR, FN, SNDK, STX, IREN, APLD, MOD, AAON, CAMT, ONTO.
Near-Term Catalyst Watch
MOD: fiscal Q4/full-year 2026 results scheduled for May 27, 2026.
MU: fiscal Q3 likely in June 2026, useful for the memory read-through.
Most Q1 reporters: next reporting windows likely late July through August 2026 unless company calendars update sooner.
Diligence Checklist
Before turning any candidate into a trade, check whether it is selling a scarce input or just borrowing AI language.
Market Checks
- Stock move since last earnings and since January 1, 2026.
- Short interest and borrow availability, especially for smaller or speculative names.
- Current EV/EBITDA and P/E versus each company's own history.
- Insider selling after AI-related announcements.
Fundamental Checks
- Backlog quality: cancellable or non-cancellable, gross margin, duration, and customer concentration.
- Customer exposure to hyperscalers, CoreWeave, Oracle, Microsoft, xAI, Meta, Amazon, or NVIDIA partners.
- Capacity expansion timing and whether customers can multi-source the product.
- Upcoming earnings call topics likely to confirm acceleration or expose execution issues.
Sources
Primary links used in the research pass. Company releases should be rechecked before any live decision.