scripod.com

Updates From the #1 AI Investor: Leopold's NVIDIA & Anthropic Strategy

This podcast analyzes the investment strategy of prominent AI investor Leopold Aschenbrenner, focusing on his reported short position against NVIDIA and a significant private stake in Anthropic. The discussion explores the shifting dynamics of the AI infrastructure trade, moving beyond just semiconductor chips.
The hosts examine Aschenbrenner's portfolio rotation, which includes shorting NVIDIA, ASML, and Oracle while going long on power, memory, and neocloud sectors. They highlight NVIDIA's $25 billion bond offering, despite its strong cash position, as a potential signal of a market top or a shift to a new funding model. Aschenbrenner's fund holds a 20% stake in Anthropic, valued at $7 billion, reflecting a bet on AI model developers. The conversation identifies the next wave of AI infrastructure investment, moving from chips to power, energy, memory, and physical data-center buildout. Key positions include optical fiber companies and NeoCloud providers like CoreWeave and Iron. The hosts debate whether this represents a bubble, concluding that real customer demand and physical constraints differentiate it from speculative manias. They emphasize energy stocks like Bloom Energy as a critical investment theme, driven by growing electricity needs independent of AI trends.
00:00
00:00
Fund has grown to over $20 billion
01:24
01:24
NVIDIA's first bond raise since 2021 signals a potential top in AI trade
03:43
03:43
Borrowing while returning cash to shareholders
07:35
07:35
AI trade is overcrowded
09:19
09:19
20% of the fund is in Anthropic equity.
12:19
12:19
Shift from bottleneck investments to physical AI infrastructure.
14:28
14:28
Real customers exist and physical constraints prevent over-leverage
17:55
17:55
Shift from copper to optical fibers for data center networking.
22:01
22:01
He is not in a bubble and has rotated his portfolio.
23:25
23:25
Electricity demand drives energy investments
24:47
24:47
AI stocks surge on endorsements