Private AI companies like OpenAI and Anthropic are achieving unprecedented valuations and revenue growth, reaching milestones two to three times faster than previous tech giants.
The primary investment thesis has shifted from "follow the GPU" to "follow the gigawatts," identifying electrical power as the main bottleneck and driver of the AI buildout.
The architecture of AI is evolving to support persistent, agent-based computing, causing a potential flip in server architecture from being GPU-heavy to CPU-heavy (e.g., from 1:4 CPU:GPU to 4:1).
Infrastructure components, particularly memory, are experiencing unprecedented supply tightness, with commitments extending years into the future, making "sellers of shortage" (e.g., memory and power suppliers) a key investment area.
12 quotes
Concerns Raised
A significant technological breakthrough (a 'Deepseek moment') could make current hardware and infrastructure less necessary, disrupting the 'sellers of shortage' thesis.
The potential for unforeseen regulation that could dampen the current unconstrained growth of the AI sector.
Opportunities Identified
Investing in the suppliers of key bottlenecks, particularly power infrastructure and memory manufacturers (e.g., Samsung, Hynix).
The potential resurgence of CPU manufacturers like Intel as AI workloads shift towards agent-based, serial processing tasks.
The massive expansion of the total addressable market for compute as thousands of AI agents are deployed per person, exponentially increasing the technology footprint of individuals.