Sources of Stock Ideas
How we systematically generate the next stock to analyze. Updated May 2026.
With ~5,500 publicly-traded US stocks (and many more globally), we cannot analyze them all. Our framework has two parts:
- Two thesis buckets — what kind of investment we are looking for.
- Discovery methods — how we surface candidates that might fit one of those buckets.
Every analyzed stock should slot into Bucket 1 or Bucket 2. If it fits neither, we should not be looking at it. The market prices near-term fear; we price multi-year fundamentals — that gap is the edge.
The Two Buckets
Every investment we make falls into one of these two categories. They are the only theses we underwrite.
1AGI BeneficiariesBucket 1
Anything that benefits from our core AGI worldview. AGI is the single biggest economic shift in our lifetime — companies positioned to capture value from it are the first place we look.
What qualifies
- Compute infrastructure — GPUs, custom AI chips, networking, packaging (NVDA, AVGO, AMD, TSM, MU, SK Hynix)
- Power for compute — Nuclear (CEG, CCJ, LEU), gas (EQT, LNG), grid equipment (GEV, ETN), behind-the-meter (BE)
- Physical bottlenecks AGI cannot replicate — Mining (FCX, MP, copper, rare earths), real estate near data centers, irreplaceable physical assets
- Companies AGI directly enables — AI software platforms with data moats, robotics enablers
How to underwrite
Identify the layer of the AGI stack the company sits in. Confirm the demand is structural, not cyclical hype. Estimate downside (what is the floor if AGI demand disappoints, capex pauses, or competition intensifies?) and bull case (what is this worth in 5 years if AGI plays out as expected?). Then decide: invest, watch, or pass.
Example: When NVIDIA disclosed B200 needs 8 HBM stacks vs H100's 5, that immediately made HBM capacity the binding constraint — leading us to MU and SK Hynix.
2Punished & Too CheapBucket 2
The mirror image of Bucket 1. The market routinely overshoots when it punishes a stock — sometimes because of AGI fears, sometimes for ordinary reasons (cyclical trough, scandal, guidance cut, sector rotation, forced selling). When the punishment exceeds the actual damage, the floor is high and the upside is asymmetric. The market prices near-term fear; we price multi-year fundamentals.
The two flavors of "punished"
- Punished by AGI fears — IT services (INFY, ACN, CTSH), staffing firms, legacy SaaS, traditional education. The market may have correctly priced in disruption — but it may also have overshot, mispriced the timing, or missed an offsetting tailwind. Always worth checking before agreeing with the market.
- Punished by non-AGI reasons — Cyclical troughs (memory, oil, homebuilders), special situations (spin-offs, post-bankruptcy, activist targets), deep value (P/TB < 1, cash > market cap, EV/sales < 0.5x), sectors in the doghouse (office REITs, life-science REITs).
What we require
- The floor must be real — tangible book backing, real cash on the balance sheet, hard assets (aircraft, real estate, ore reserves), or a regulatory moat that protects cash flow.
- The business is not structurally broken — revenue not collapsing, no fraud red flags, not in terminal secular decline.
- Optionality for a 10x — a credible path to a much larger market cap if things go right (cycle turns, asset re-rates, AGI threat overstated, activist forces value).
How to underwrite
Start with the floor: what is the worst case? If we can construct a case where we do not lose money, then look for the upside. Estimate bear/base/bull scenarios over 5 years. Decide: invest, watch, or pass. The hardest discipline is distinguishing temporary pain from terminal decline. Pre-revenue companies are an automatic kill.
Examples: AerCap (AER) at 0.86x book — world's largest aircraft lessor with 21% ROE, hard-asset floor. ARE (Alexandria Real Estate) crashed 79% as life-science labs went into oversupply, but AI drug discovery actually increases wet-lab demand. TK trading at $11.80 with 73% of market cap in cash and zero debt.
Discovery Methods
How we surface candidates that might fit Bucket 1 or Bucket 2. Any of these methods can produce ideas for either bucket.
AFirst-Principles ThinkingHighest Yield
Start from a structural force in the world (AGI, cycle, demographic shift, regulatory change) and reason forward to who benefits and who gets hurt. Then go find the listed equities that fit. This is the highest-yield method because the conclusions are ours, not borrowed — and we get there before the market re-rates.
What to do
Maintain a running list of structural forces and their second-order effects. When AGI news hits (new model release, hyperscaler capex announcement, nuclear restart, chip export rules), reason through the chain: who benefits at each layer of the stack, what binding constraint emerges, who gets disrupted. Add candidates to the analysis queue and slot them into Bucket 1 or Bucket 2.
Example: Reasoning forward from "training runs need 5 GW by 2028" → grid is the binding constraint → behind-the-meter power is the only fast solution → fuel cells (BE) and gas turbines (GEV) become first-order beneficiaries. This chain led us to BE before Leopold's 13F confirmed it.
BSuperinvestor 13FsQuarterly
Track the portfolios of investors whose judgment we respect. Their best ideas often outperform the market and we get to leverage their research for free. Filed 45 days after each quarter-end.
The four investors we follow actively
Leopold Aschenbrenner
Situational Awareness Investments. AGI-pilled. Top: BE, CRWV, INTC. Mostly Bucket 1.
Ted Weschler
Berkshire portfolio manager. Private portfolio is what we want — separate from BRK 13F.
Mohnish Pabrai
Pabrai Wagons Fund. Concentrated value investor, copies Buffett style. Mostly Bucket 2.
Li Lu
Himalaya Capital. Munger's protégé. Concentrated, long-term, often China-focused.
What to do
Each quarter, pull the new 13Fs. Look for: (1) new positions, (2) significantly added positions, (3) any holding where price has not run up much from the estimated cost basis. For each candidate, decide which bucket it fits.
Example: Leopold's $1.7B BE position validates the behind-the-meter data center power thesis (Bucket 1). His CORZ activist 13D was a Bucket 2 idea (BTC miner trading near floor) that became Bucket 1 once the BTC→AI pivot crystallized.
CNews-Driven Movers (Up)Active
Stocks that have moved up on news. The market reacted to something — investigate whether the catalyst is structural (and could sustain) or whether it is a one-time pop. Catching a structural trend early can mean buying before the full repricing.
What qualifies
- Stocks up 20%+ in a week or month on company-specific news (not market beta)
- Sector-wide moves driven by industry data (e.g., memory prices spiking)
- Earnings beats that signal a structural inflection (margin expansion, new product traction)
- Regulatory changes that benefit a specific group of companies
- New customer wins or contract announcements that change the trajectory
What to do
Daily/weekly scan of biggest movers. For each significant mover, ask: "Is this a one-time pop or the start of something structural?" If structural, prioritize for analysis. Speed matters — entering after the first 30% move is often still very early in a multi-year trend. Most candidates here will be Bucket 1.
Example: When DRAM spot prices spiked 65% QoQ and SK Hynix hit 72% operating margins, that was not a one-quarter event — it was the start of a multi-year HBM-driven supercycle. Investigating in real time would have meant catching the first leg of the move.
DCrashed StocksPeriodic
Periodic scan for stocks that have dropped significantly. The most asymmetric bets often come from temporarily distressed quality businesses. Mr. Market hates uncertainty — when good companies face temporary problems, the stock overshoots to the downside. Almost every candidate here is Bucket 2.
What qualifies
- Stocks down 30%+ in the last month
- Stocks down 50%+ in the last 6 months
- Quality companies that crashed on guidance cuts (vs structural deterioration)
- Sectors in the doghouse (recently: office REITs, life science REITs, education stocks)
- Stocks that hit 52-week lows where the business is fundamentally intact
What to do
Monthly scan: filter our universe for stocks down >30% in the last month. For each, determine: "Is the business broken, or just the stock?" If just the stock, this is the entry point. If the business is broken (revenue collapsing, secular decline, fraud), pass.
Example: ARE crashed 79% from peak as life science labs went into oversupply. But AI drug discovery actually increases wet lab demand — the crash created the entry point for a structural growth story (Bucket 2 with a Bucket 1 tailwind).
EPeriodic ScreensQuarterly
Brute-force scans across the full universe for statistical cheapness. Almost all candidates here are Bucket 2. The April 25, 2026 funnel run is the canonical example — pre-filter the universe, rank, then parallelize 16 Opus agents to do the deep dives.
What we screen for
- Price-to-tangible-book < 1.0x with positive cash flow
- Cash > market cap with a real operating business
- EV/Sales < 0.5x
- Cyclicals at trough where we have conviction the cycle is about to turn
- Special situations — spin-offs, post-bankruptcy, activist targets, buyout candidates
What to do
Run quarterly. Pre-filter the universe to a manageable set, then send the survivors through the multi-agent funnel. The Round 1 universe screen ($50M–$5B) surfaced AER, ECPG, NEN. The April 25, 2026 run surfaced OSG, OM, ACTG, FPH, CCRN, CABO, GASS, XPER, CYRX, SRBK.
Summary — How It All Fits
Two buckets (what we buy):
- AGI Beneficiaries — companies positioned to capture value from AGI.
- Punished & Too Cheap — companies the market has overshot on, AGI-related or otherwise. High floor, asymmetric upside.
Five discovery methods (how we find them):
- First-Principles Thinking — reason forward from structural forces.
- Superinvestor 13Fs — Leopold, Weschler (private), Pabrai, Li Lu. Quarterly.
- News-Driven Movers — distinguish one-time pops from structural inflections.
- Crashed Stocks — quality businesses with temporary problems. Monthly scan.
- Periodic Screens — statistical cheapness across the full universe. Quarterly funnel.
This document is the source of truth for stock idea generation. Update as the framework evolves.