Investment Memo · Situational Awareness Fund

Leopold's Q1 portfolio, decoded

Thesis layers, the new put-hedge signal, and a $100K allocation framework.

filed · Q1 disclosure · today

Good timing on this — the filing just dropped. Below is a breakdown of the thesis, the five layers his book maps to, and one way to think about a $100K allocation that mirrors his conviction without copying it line-for-line.

The core thesis

one-sentence version AGI is a 2027 event; superintelligence is 2028–2030; markets have not priced in either the exponential trendlines or the industrial mobilization they require. Therefore: buy the bottlenecks of the buildout (power, compute, memory, optics, fabs) and short what AI eats.

The key insight — Leopold is not betting on AI models themselves. His portfolio reflects a singular thesis: the real bottleneck to AGI is not algorithmic, but physical. Electricity generation and computing capacity will be the most valuable assets of the coming decade.

He envisions "Trillion-Dollar Clusters" — gigawatt-scale datacenters that will consume more electricity than small nations, necessitating a complete overhaul of the global power grid and a supercycle for specialized equipment that cools, connects, and powers these facilities.

The Q1 portfolio, mapped to thesis layers

His Q1 holdings break into five conviction buckets:

Power / Energy

biggest conviction

On-site generation and grid-bypass plays. The cleanest part of the thesis.

$BE Bloom Energy — $878.7M $TE T1 Energy — $43.9M $SEI Solaris — $62.5M $BW Babcock & Wilcox — $19.9M

Compute infrastructure / data centers

core build

The picks-and-shovels of the cluster era.

$CRWV CoreWeave — $556.1M $CORZ Core Scientific — $389.1M $APLD Applied Digital — $320M $IREN Iren Limited — $401M

Bitcoin miners

repurposed capacity

Cheap power + existing facilities → pivot to AI compute. The spiciest layer of the book.

$RIOT Riot Platforms — $142.2M $CLSK Cleanspark — $104.5M $BITF Bitfarms — $38.8M $BTDR Bitdeer — $29.8M $HIVE Hive Digital — $6.4M

Memory / storage

HBM thesis

The under-covered constraint on inference at scale.

$SNDK Sandisk — $724.4M $MU Micron — $5.9M

Semiconductors / optics

smaller weights

Diversified exposure to fabs and the optical interconnect supercycle.

$INTC Intel — $8.9M $AMD — $20.2M $TSM Taiwan Semi — $7.6M $ASML — $6.1M $SMH — $10.3M $GLW Corning — $21M $WYFI WhiteFiber — $20.9M $PSIX Power Solutions — $26.3M $SHAZ SharonAI — $18.1M $PUMP Propetro — $13.1M

The new development — filed today

On May 18, 2026, Situational Awareness LP disclosed a $137 billion notional portfolio — over 60% allocated to put options on AI hardware stocks including $NVDA, $AMD, and $SMH. A bearish near-term overlay on the semiconductor sector specifically.

So he's simultaneously long the infrastructure layer and hedging against near-term AI hardware multiple compression. That's a nuanced read — he still believes in the buildout, but thinks the chip stocks got way ahead of themselves.

A $100K allocation framework

Not financial advice. This is one way to mirror the thesis — sanity-check it with a financial professional before pulling the trigger, especially the miner exposure.
Bucket Allocation Rationale
Power / Energy $30K Highest conviction in his book; $BE + $IREN as core plays.
Compute Infra $25K $CRWV + $CORZ; more volatile, higher ceiling.
Semiconductors $20K $SMH ETF for diversified exposure vs single-name risk; $INTC if you want the contrarian call.
Memory $10K $MU over $SNDK — more liquid, same thesis.
Bitcoin Miners $10K $RIOT or $CLSK; high-risk / high-reward repricing as compute demand hits.
Cash / optionality $5K Given his put-hedge signal, keep some dry powder.

My take

The power thesis is the cleanest

Bloom Energy's solid-oxide fuel cells offer a clever workaround for powering data centers: on-site, modular power generation that bypasses the traditional grid. Aschenbrenner identified this energy shortage before Wall Street caught on. That's the part of the thesis most people still underestimate.

The miner repricing is the spiciest bet

High volatility, but if agentic compute demand hits the numbers he's projecting, those cheap power assets become extremely valuable fast. Asymmetric upside, real downside.

The put hedge is the timing signal

This is the part I'd weight most for your entry — he's saying "buildout is real, but chip multiples might compress in the near term." Could mean better entry points coming on $NVDA / $AMD if you have patience.

Does that framework make sense, or want to go deeper on any specific layer?