This 24 Year-Old Made $4.5B Investing in These Companies
Summary
- The "Situational Awareness" Pivot: Leopold Aschenbrenner, who turned \$1B into \$5.5B in a year, has fundamentally shifted his portfolio from AI Chips (GPUs) to AI Infrastructure (Energy & Power).
- The Chip Trade is "Priced In": The fund dumped over \$1B in chip stocks (NVIDIA, Broadcom, TSMC, Micron), believing the market has correctly priced the hardware, but has severely underpriced the energy constraints required to run that hardware.
- Energy is the New Bottleneck: The thesis identifies "Power Generation" as the critical constraint. The grid cannot handle AI demand; therefore, on-site power generation (fuel cells) and entities holding power permits (Bitcoin miners) are the new alpha.
- AI Deflationary Pressure: A major short position highlights the belief that AI coding agents will decimate the traditional IT outsourcing business model.
Trade Ideas
The fund has "dumped NVIDIA... dumped Broadcom... dumped TSMC... dumped Micron." He sold \$300M in NVIDIA puts (profit taking) and exited the equity. The "easy money" in the hardware layer has been made. The market has fully priced in GPU production. The capital cycle is moving from the *processor* (chips) to the *enabler* (energy/hosting). Holding these names now offers lower risk-adjusted returns compared to the infrastructure layer. AVOID (Rotation). This is a sector rotation call: Sell Chips, Buy Power. The AI capex boom could continue to surprise to the upside for chipmakers; NVIDIA remains the monopoly supplier.
Leopold has built a massive position in Bloom Energy to the tune of "\$855 million" (approx. 20% of the portfolio). AI data centers face a power crisis; the public grid is too slow and congested to meet demand. Bloom Energy manufactures solid oxide fuel cells that convert natural gas to electricity *on-site*, allowing data centers to bypass the electrical grid entirely. LONG. This is the "NVIDIA of Energy" play—betting on the hardware required to generate power locally for hyperscalers. Manufacturing execution risk (can they build enough turbines to meet the \$20B backlog?) and reliance on natural gas prices.
Ejaaz Ahamadeen
Co-Host, Limitless Podcast (Bankless)
Leopold owns around "10% of a company called Core Scientific" and is buying Bitcoin miners because "they own the two critical pieces of infrastructure required for AI buildouts... real estate and power." Building new power infrastructure takes years due to permitting. Bitcoin miners (like Core Scientific) already possess high-capacity power interconnects and grid rights. The play is an arbitrage: repurposing crypto-mining facilities into AI data centers (hosting for CoreWeave) to skip the multi-year regulatory wait times. LONG. A play on "power arbitrage"—buying the rights to electricity access rather than the crypto asset itself. Bitcoin price volatility affecting the stock's beta; execution risk in pivoting infrastructure from mining to HPC (High Performance Computing).
Ejaaz Ahamadeen
Co-Host, Limitless Podcast (Bankless)
Leopold has taken out a "massive short" on Infosys, a company specializing in IT outsourcing. Infosys's business model relies on labor arbitrage (cheaper human engineers). New AI agents (like Claude Code and GPT Codex) are becoming capable enough to automate menial IT and coding tasks. This technological deflation destroys the value proposition of human-heavy outsourcing firms. SHORT. A bet on AI displacing entry-level and mid-level coding labor. AI adoption in enterprise environments may be slower than anticipated; Infosys could pivot to integrating these tools (though this cannibalizes their billing model).
