Anjney Midha, CEO of AMP, discusses the critical importance of output maximization in AI infrastructure. He argues for compute pooling modeled on the electric grid, emphasizes common sense in scaling, and calls for aligning community interests with data center growth. The conversation also explores cultural alignment in AI labs and Midha's personal mission to use AI for improving end-of-life care.
Summarized by Podsumo
Node utilization should be at 95% (Google's standard), but MFU (model flop utilization) best-in-class is only 60-70%—indicating massive waste.
AMP proposes a 'compute grid' to pool resources across clouds, making megaflops flow like megawatts, similar to the power grid's independent system operators.
Up to 20% of US data centers face community backlash; Midha suggests a $0.50/hour surcharge paid directly to local communities to build trust.
Anthropic's success is attributed to 'prepared luck' and a culture of safety focused on coding, contrasting with labs that overfund without hardship.
Midha's personal mission is using AI for end-of-life prediction to reduce 30% of Medicare/Medicaid spend and empower patients.
"Luck favors the prepared mind. Anthropic has been the most prepared company for four years."
— Anjney Midha
"Culture is not a set of beliefs, it’s a set of actions."
— Anjney Midha (citing Bushido)
"If you have to bring up a new data center in a new community, why not just say we're gonna charge $4.50 an hour and give that marginal increase to the local community as cash?"
— Anjney Midha