Anjney Midha's Plan to Radically Lower the Price of Compute

AI development faces significant physical constraints, as the pursuit of frontier models consumes vast amounts of energy, compute, and infrastructure. Anjney Midha, founder of AMP PBC and former Andreessen Horowitz partner, argues that the industry currently suffers from massive inefficiencies due to fragmented compute resources and suboptimal utilization. Rather than a singular "winner-takes-all" model, AI progress follows a "jagged frontier" where specialized, verifiable feedback loops—such as those used in software engineering—drive rapid advancement. To address the compute bottleneck, Midha advocates for a software-defined "grid for compute" that standardizes diverse hardware into a fungible resource. This approach aims to eliminate the deadweight loss inherent in traditional long-term leasing models, allowing research labs to achieve higher utilization rates and focus on output-driven innovation rather than mere hardware acquisition.