AGI-thesis investment research. Best ideas, sector research, and deep dives.
High-conviction positions. Click "View Report" for full analysis with live price chart.
| Ticker | Company | Thesis | Score | Report |
|---|---|---|---|---|
| TSM | Taiwan Semiconductor Manufacturing | World's only advanced-node foundry. Every AI chip is made here. Physical moat: $100B+ and 10+ years to replicate. | 9 | View Report → |
| MU | Micron Technology | HBM memory for AI GPUs. Cycle play — buy at trough ($90-140). Currently at peak earnings. Supply shortage extends to 2027-2030. | 8 | View Report → |
| SKHYNIX | SK Hynix (000660.KS) | HBM market leader (62% share). Best memory business in the world. Korean-listed, harder to access but better positioned than Micron. | 8 | View Report → |
| TLN | Talen Energy Corporation | Nuclear IPP with 1,920 MW AWS PPA through 2042. 11.3× fwd P/E, 8% FCF yield. Discount to CEG (19.6×) may reflect single-plant risk and post-bankruptcy stigma — or mispricing. | 8 | View Report → |
| ASML | ASML Holding N.V. | Literal monopoly on EUV lithography. €38.8B backlog. Every advanced AI chip passes through an ASML machine. 2 High-NA systems shipped. 60+ EUV target for 2026. | 9 | View Report → |
| TLT/EDV | Long-Duration Treasury Puts (options program) | Bond market prices ~0% AGI probability (Chow–Halperin–Mazlish). Rolled 18-24mo puts, 10-15% OTM, ~0.5-1%/yr premium budget. If AGI capital demand breaks the rates regime (30y → 8-10%), TLT falls 35-50% and puts pay 10-30x. Defined risk — no carry bleed of an outright short. Complements existing 2.5%/3.125% fixed mortgages (already short duration). | 8 | No report yet |
| C / BANKS | Citigroup & US Big Banks | AI cost-side transformation — banks are in nobody's AI basket. ~55-60% of opex is automatable white-collar labor (compliance, KYC, ops). Convexity = opex/net income: Citi ~3.4x (10% expense cut ≈ +34% EPS) vs JPM ~1.2x. Citi is a convergence trade: closing the efficiency gap to peers re-rates ~0.8x → 1.2-1.5x TBV even if industry-wide savings get competed away. Charter scarcity + deposit/trust moat gate new entry; AI as fixed-cost tech favors the big four. Key risk: 15 years of failed self-transformation. | 8 | No report yet |
Q1 2026 13F (SEC filing date 2026-05-18). $13.7B AUM, 42 positions. Major reversal: dropped INTC calls, added massive put book on NVDA/AVGO/AMD/MU/TSM/ASML/SMH. SNDK now top long.
INVESTORLi Lu, Ted Weschler, Todd Combs, Mohnish Pabrai, Leopold Aschenbrenner.
Capacity and contract model for CRWV, NBIS, IREN, HUT, CORZ, APLD, WULF, CIFR, and GLXY. Direct GPU-cloud pricing power vs fixed data center rent.
INDUSTRYLIN, AIQUY, APD, NIPNF essentially didn't move while NVDA did 6x. Honest reasons: semis are ~10% of revenue, take-or-pay contracts cap upside, wafer starts grew slowly, APD's hydrogen blowup, oligopoly too big to re-rate.
INDUSTRYDetective-style analysis of whether AI memory demand can stay ahead of qualified supply, and what that means for Micron's normalized earnings.
INDUSTRYEvery power source (grid, gas, fuel cells, solar, nuclear, SMRs, geothermal, fusion), real costs, timelines, and what actually matters.
INDUSTRY$150B foundry market. TSMC dominance, Intel trust deficit, CHIPS Act.
INDUSTRYEthernet switches, InfiniBand vs Ethernet for AI, NVIDIA Spectrum-X threat, Broadcom ASIC monopoly. ANET/AVGO/NVDA.
INDUSTRYGW capacity estimates for Anthropic and OpenAI. Google/Broadcom 3.5 GW deal. Projections: 2 GW per training run by 2027, 5 GW by 2028.
INDUSTRYHow biotech works, drug development process, sub-sectors, AI/AGI disruption, Anthropic's Coefficient Bio acquisition, AI labs entering biotech.
AI CHIPSGory technical comparison of Blackwell/Rubin, TPU v7 Ironwood, Trainium2/3, Cerebras WSE-3, Groq LPU, Meta MTIA v3, OpenAI Titan. Includes Nvidia's $20B Groq acquisition (Dec 2025), Cerebras's pivot from G42 to OpenAI, and strategic outlook on who threatens Nvidia.
AI CHIPSInvestment framework for evaluating new AI chip startups. 12-factor scorecard, capital math, startup graveyard, red/green flags, decision tree. Updated with Groq-Nvidia deal and current Cerebras state.
AI CHIPSBackground reading: the stack of abstractions, design pipeline end-to-end, industry structure (fabless / foundry / EDA / IP / lithography). Time-invariant material to build on.
Post-AGI compute demand, training vs inference, who captures value.
QUESTIONPermanently scarce, AGI-proof assets. All publicly traded teams.
QUESTIONPost-AGI land value thesis. Scarce physical assets.
QUESTIONPower, chips, cooling, networking — what AGI needs day one.
QUESTIONEvery component from electricity to token — and which survive when AGI can design anything but needs time to build it.
QUESTIONHow long will AGI self-improve before solving real-world problems? The interstellar analogy, and why the answer is "not long."
Funnel from 513 AI-exposed stocks to a valuation-adjusted top five and 50-stock winner pool for Jan 1, 2029.
FUNNEL RUNFirst end-to-end deep-value funnel run. Pre-filter → rank → 16 parallel Opus agents. Best ideas: OSG, OM, ACTG, FPH, CCRN, CABO, GASS, XPER, CYRX, SRBK.
FRAMEWORKThe 5 systematic sources we use to generate the next stock to analyze: AGI thesis, deep value, superinvestor 13Fs, news movers, crashed stocks.
SECTOR REPORTCapacity expansion plans for Samsung, SK Hynix, Micron. Bottleneck sequence (CoWoS → stacking → wafers). Earliest oversupply: H2 2027.
DEEP DIVEBusiness, financial, management, risk, AGI impact, valuation.
DEEP DIVEOwned vs leased vs neocloud GW. Hyperion, Prometheus, fleet, MTIA. Replacement value $800B–$1.2T vs implied market value $200–400B.
STRATEGY8 analysis layers, funnel from 8,049 to ~10 companies.
PILOTAsset-based valuation and downside protection methodology.
FRAMEWORK24 categories — the master checklist for deep-diving any company.
Data sources: SEC EDGAR (10-K, 13F), company filings, Forbes franchise valuations. Live prices: Finnhub. Fair values: first-principles segment analysis by Claude.
Last updated: 2026-06-12