How the smart money is positioned — and how we are
The iconic AI, semiconductor, infrastructure, nuclear and space books — hedge funds, ETFs, the marquee VC labs — ranked by performance, each with holdings and our read. Then the QAI Fund: our house model portfolio, built from the universe below. Analyst work product, never advice.
Read the thesis before it’s consensus — every board, dossier & note the moment the desk ships it.
How we’d build it — model portfolio
The durable AI margin sits in compute and the physical buildout that feeds it — not in any single name. The QAI Fund overweights the semiconductor engine and the power/grid/datacenter layer, rents the frontier (space, robotics, private-AI labs) for optionality, equal-weights the mega-caps to dampen single-stock (NVDA) concentration, and holds a deliberately large ~25% cash reserve as dry powder for the boards’ next shift-points and any drawdown. Built from the funds tracked below; rebalanced on the boards.
The engine. Diversified semiconductor exposure to the compute super-cycle — design, equipment, and foundry.
Risk: One cyclical sector; a chip de-rate hits the whole sleeve at once.
Equal-weight the Mag 7 in one ticker — owns the hyperscaler capex + AI platforms while damping single-name NVDA concentration.
Risk: Still seven correlated mega-caps; equal-weight only partly diversifies the AI factor.
The binding physical ceiling on the buildout — transformers, switchgear, transmission. Picks-and-shovels of electrification.
Risk: Rate-sensitive industrials; a capex pause slows the order book.
24/7 power for datacenters — utilities + fuel cycle, the cleanest "AI-buys-nuclear-power" expression with less commodity whipsaw than miners.
Risk: Policy- and PPA-dependent; sentiment reverses fast off the uranium re-rating.
The real estate + memory of compute — data-center REITs plus the HBM/memory sleeve that physically enables AI.
Risk: The memory sleeve adds compute-cycle beta a pure-REIT name would not carry.
AI leaves the datacenter — autonomy, industrial robotics, and defense-tech, the physical-AI edge of the buildout.
Risk: Active/thematic; a robotics capex cycle that has not yet re-accelerated.
Frontier optionality — the SpaceX-adjacent launch economy and the rearmament cycle, sized as optionality not core.
Risk: High thematic-fad volatility; concentration in a few new-space names.
An explicit GenAI buildout basket — platforms, compute, and (via CHAT) genuine international AI exposure the US-only funds lack.
Risk: Overlaps the compute core; differentiation is thinner than the label implies.
The frontier labs (Anthropic, OpenAI) + SpaceX (xAI)/Anduril via the marquee VC books — the upside the public market cannot yet price.
Risk: Access-constrained private positions held through the marquee VC books; carried at unrealized marks, IPO-timing dependent.
A deliberately large dry-powder reserve (~20–30% target) — held to add into the boards’ next shift-points and to buy drawdowns in a high-beta book.
Risk: Cash drag in a melt-up — the deliberate cost of keeping real optionality.
The QAI Fund is a real private fund; the model-portfolio breakdown above is its house positioning across the tracked universe, vintage-stamped to Q1–Q2 2026 and moving with the market. This is not an offer to sell or a solicitation to buy any security, and not investment advice; interests are offered only to qualified investors under formal fund documents. Pre-IPO sleeves hold access-constrained private positions carried at unrealized marks.
Consensus book
The names the tracked funds hold most — an aggregate of their books, not a QAI recommendation. Tap one to filter the funds below to who holds it.