Why a $100 Million Salary for an Elite AI Researcher is a Bargain

Dylan Patel argues a $100M salary for a top AI researcher is rational because a 5% compute efficiency gain saves billions across training and inference fleets.

• A researcher who finds a 5% efficiency gain saves that percentage across both training runs and the entire inference fleet, compounding over repeated improvements.
• Adding more researchers to AI training slows progress; experiments are sequential and require gut feel to interpret trend lines, not parallelizable headcount.
• Rune (OpenAI) proposed aggressively recruiting process-knowledge holders from Shenzhen and other global hubs, framing it as a US national competition.
• Intel’s semiconductor talent pipeline dried up because PhDs in nano-chemistry chose Google ($800K) or Meta ($10M+) over chip fab roles ($200K).
• Jensen Huang told Patel: America is rich because it exported labor but kept all the value — Nvidia and Apple capture gross profit while Asian manufacturers do not.
• ML research and semiconductor manufacturing are structurally identical: thousands of interdependent knobs, impossible to exhaustively test, dominated by intuition-guided search.
• Sam Altman publicly claimed Meta didn’t poach OpenAI’s best people, but privately ran counter-offers — signaling the talent war is already at crisis level internally.

2025-10-02 · Watch on YouTube