What The Best AI Investors Are Buying Right Now
Speakers
GavinSBaker
— Portfolio Manager, Atreides Management
Ejaaz Ahamadeen
— Co-Host, Limitless Podcast (Bankless)
Josh
— Co-Host
Summary
The hosts analyze the AI investment thesis of veteran investor Gavin Baker, contrasting it with Leopold Aschenbrenner's. They walk through Baker's portfolio, which focuses on AI infrastructure bottlenecks—connectivity, memory, inference chips, and world models—while holding a bearish hedge on the broad market via QQQ puts. The core thesis is that AI is in a supercycle constrained by physical supply, not a dot-com bubble.
- Gavin Baker's 13F portfolio emphasizes infrastructure bottlenecks like connectivity (Astera Labs), memory (Micron), inference (Cerebras), and world models (Unity).
- Baker is heavily invested in NVIDIA for 20+ years and sees a path to $10 trillion market cap.
- He holds a large put position on QQQ as a hedge against broad market downside.
- The hosts compare Baker's decades-long track record to Leopold Aschenbrenner's high-octane approach.
- Baker argues AI is a supercycle, not a bubble, because demand is funded by cash flow and supply is physically constrained.
- Key constraints include chip fabrication (TSMC), memory (SK Hynix), energy, and deployment speed.
- The shift from pre-training to inference and reasoning creates 5-10x more compute demand.
- Unity is highlighted as a world model builder for training AI agents and robots.
Trade Ideas
Astera Labs solves AI connectivity bottleneck
Astera Labs provides the critical connectivity plumbing between GPUs in large-scale AI clusters. As data centers scale to hundreds of thousands of chips, the bottleneck shifts from GPUs to data transfer, and Astera Labs' solutions solve that, making it a key infrastructure play.
NVIDIA long, path to $10T
NVIDIA will maintain its high profit margins and strong demand as the AI supercycle unfolds, with a clear path to a $10 trillion market cap given its dominant GPU position and the insatiable demand from hyperscalers for training and inference compute.
Micron benefits from AI memory bottleneck
Micron is a leading memory manufacturer essential for AI workloads. Given the massive demand for memory chips from AI training and inference, and the company's recent 10x market cap growth to over $1 trillion, the memory bottleneck creates a huge opportunity for Micron.
Unity powers AI world model training
Unity Software is a world model builder with deep physics and 3D rendering capabilities. As AI moves toward AGI and humanoid robots, simulated environments for training become critical, and Unity's engine is one of the best platforms for creating these virtual datasets, making it a unique AI play.
Cerebras captures inference compute growth
Cerebras designs specialized inference chips that are crucial as the AI industry shifts from pre-training to post-training and reasoning. Inference compute demand is estimated to be 5–10x larger than pre-training, and Cerebras is positioned to capture that growth alongside other inference-focused companies.
Bearish on broad market via QQQ puts
Gavin holds a large put position on QQQ (the Nasdaq 100 ETF) as a hedge against broad market downside. While he is bullish on specific AI infrastructure companies, he is bearish on the general market, believing the QQQ index may decline, and uses puts to express that view and protect against systemic risk.
