Thesis of the Week: The Hardware Wall and the End of Cheap Growth
This week underscored a growing realization: AI scaling won’t be halted by a lack of data, but by physical constraints. Energy and compute are becoming the ultimate bottlenecks.
1. AI & Compute: The Battle for Data Centers
Major cloud providers are reporting increasingly long lead times for building new data centers. The power requirements of current GPU clusters are overwhelming local grid capacities in many hotspots.
2. Energy: A Nuclear Renaissance?
Tech giants are investing heavily in alternative energy sources. The discussion around Small Modular Reactors (SMRs) is gaining momentum, as traditional renewables often struggle to provide the baseload power needed for 24/7 AI operations.
3. Robotics: The Next Wave of Automation
While digital AI masters text-based tasks, we’re seeing a surge in startups training foundation models for physical robots. The “Embodied AI” sector is attracting massive capital.
4. Crypto: DePIN as a Solution to Infrastructure Bottlenecks?
Decentralized Physical Infrastructure Networks (DePIN) are positioning themselves as alternatives to centralized cloud providers. Projects aggregating underutilized GPU capacities are experiencing strong growth.
5. Longevity: Data-Driven Prevention
The use of AI in drug discovery continues to accelerate, but in the near term, we’re seeing the biggest impact in personalized, AI-driven prevention programs.
Signal vs Noise
Signal: Strategic investments by tech companies in energy infrastructure. Whoever controls the power, controls the AI. Noise: Announcements of “AGI in 2 years.” The physical reality of construction timelines and power grids will dictate the actual pace.
Portfolio Moves
- Focus on infrastructure providers (compute, energy, cooling).
- Monitoring DePIN projects in the GPU space.
- Caution regarding pure software AI startups lacking a clear moat against hyperscalers.