mike

AI has created a novel memory type (HBM) that is structurally supply-constrained, required in exponentially growing quantities per GPU generation, and consumed by agentic inference workloads, ending the 40-year commodity 'pig cycle'.

Heron Power is adapting Model 3/Y power electronics for grid deployment, delivering power to data centers faster and cheaper than legacy transformer tech. This represents a fundamental rethink of grid infrastructure using modern power electronics.

Radiant Nuclear's 1.5MW container microreactor (5-year no refuel) and Form Energy's iron-air battery (100-day duration, rust-based chemistry) address data center needs for firm, long-duration power. Both leverage physics advantages for stationary grid applications.

The shift from training to agentic inference — where each agent requires its own KV cache — creates a multiplicative memory demand that dwarfs training requirements and is still in early innings.

AI infrastructure demand is triggering unprecedented capital formation: Intel's first equity raise in 55 years, Microsoft's hundreds of billions in data center projects, and Anthropic partnering with GIC to build dedicated AI compute facilities. Capital access is becoming a primary competitive differentiator.

Beijing is mobilizing domestic capital markets and household savings to finance AI infrastructure at scale, attempting to replicate the state-directed capital model that succeeded in solar and EVs. However, export controls on semiconductor equipment continue to constrain China's ability to replicate TSMC/SK Hynix process knowledge regardless of funding.

Small drone threat requires both speed and accuracy against Group 1/2 drones; 90% COTS components enable thousands/month production; fully autonomous edge compute essential for unmanned platforms; government procurement accelerating via Joint Task Force 401 across all services.

Robotics will deploy first in structured environments (factories, warehouses) for unboxing, boxing, material handling — not homes. Home robotics is 'hard mode' due to unstructured environments, safety, liability. Factory redesign around robot feeding/utilization yields 5-10x productivity vs 1:1 human replacement.