Sequoia's David Cahn on the AI talent arms race: $100M signing bonuses, 'con-sized' teams, and the gentle singularity shift

Summary

David Cahn, a partner at Sequoia Capital, frames the AI talent arms race through sports team economics and argues we are still only in the second inning.

The immediate trigger is reports of $100 million signing bonuses offered to top AI researchers. Cahn ties this to Meta's effort to assemble a roughly 50-person superintelligence lab. The revealed preference of the Magnificent Seven is unmistakable: they are betting on talent at any price because they genuinely believe AGI is a must-win. Whether that belief reflects accurate calibration or arms-race psychology is unclear, but the self-perpetuating logic works either way. Once one company sets a price floor, everyone else must respond.

The 50-person model

Cahn sees the 50-person ceiling as strategically coherent rather than arbitrary. He points to Steve Jobs's original Mac team and Elon Musk's Tesla Autopilot group, both around 50 people and both producing outsized results. The pattern works because at that size everyone still talks to everyone. For Meta's lab, the math becomes tolerable: 50 researchers at $100 million each equals $5 billion, a rational number if the prize is AGI. Cahn jokes that this unit of measurement should be called a "con" — a con-sized team.

On whether any of this is economically rational, Cahn is direct about the uncertainty. The sports franchise analogy breaks down in AI research. In sports, a star player's revenue impact can be modeled with reasonable precision. In AI, the marginal value of any individual researcher is unknown. Companies are effectively betting that if someone is among the 50 people capable of achieving AGI, the marginal benefit is extraordinarily high. That is a belief, not a proven return.

IP leakage

High-frequency trading offers a useful contrast. In HFT, noncompetes, garden-leave clauses, and strict IP controls keep a trader's edge inside the firm. AI research is far more porous. Researchers move between labs, foundational ideas like RLHF leak quickly into public literature, and there is effectively no proprietary IP in the traditional sense. The entire AI ecosystem functions more like a shared body of ideas than a collection of walled gardens. This raises a genuine question about what a $100 million signing bonus actually buys.

The gentler narrative

Sam Altman's "gentle singularity" framing, published on his personal blog rather than as official OpenAI content, marks a notable shift from the AGI-is-imminent narrative that dominated a year ago. A year back, vocal figures insisted AGI would arrive in 2026 and dismissed skeptics as outside the social circle. Now the framing is gradual, ambient, consumer-friendly. Cahn connects this directly to OpenAI's commercial position. A company with a large subscription base and real revenue has obvious incentives to frame AI progress as steady, multi-year diffusion rather than sudden discontinuity.

Cahn's broader concern applies to labs without that revenue cushion. How many multi-billion-dollar unprofitable AI labs can capital markets support over five years, particularly if progress against benchmarks like ARC-AGI stalls? ARC Prize v2 remains completely unsaturated, and progress this year has been slower than last year. OpenAI and Anthropic have enough revenue runway to weather a multi-year plateau. Others may not.

The $600 billion question Cahn raised last year still lacks a clean answer. Using Nvidia revenue as a proxy for data center spend, the industry needs to generate roughly $600 billion in revenue off $300 billion in annual data center investment just to achieve 50% gross margin. The talent arms race adds a new cost layer on top of the compute arms race. Both are running simultaneously, and the revenue to justify either is still being built.