Sequoia's Ravi Gupta: AI researchers are structurally underpaid, Zuck is assembling a team not a star, and great companies compound at 30% forever

Summary

Ravi Gupta, partner at Sequoia Capital, argues that top AI researchers are structurally underpaid, a consequence of comp band systems that assume performance within any given level clusters within a 1–2x range. His view is that reality is far more dispersed, and that the current wave of nine-figure packages — some reaching $100 million — represents a correction rather than an anomaly. Critically, these are not guaranteed contracts. Compensation vests monthly, meaning if a researcher fails to deliver, the commitment ends. The financial risk to the hiring company is more limited than headlines suggest.

Gupta frames the broader compensation distortion as a Silicon Valley structural problem, not just an AI-era one. Andy Jassy, widely credited with building AWS into a business now worth hundreds of billions, joined Amazon in 1997 as a marketing manager and was granted a 10-year equity package valued at $212 million when named CEO in 2021, earning roughly $40 million annually. Tim Cook earns approximately $75 million per year to run a multi-trillion-dollar company where a single successful tariff negotiation can shift market cap by $200 billion. The argument is that platform-level leverage has long exceeded individual compensation across big tech, and the AI talent market is beginning to close that gap.

On Meta's superintelligence push specifically, Gupta views the investment as a team acquisition, not a star acquisition. Alex Wang and Nat Friedman are named as likely leaders of the effort. The relevant unit of measurement is whether Meta becomes a leading player in the race toward superintelligence, not whether any single researcher can be credited with a specific market cap movement. Core research teams at frontier labs are smaller than widely assumed — Gupta puts the number at 30 to 50 people — which makes individual performance more legible than at organizations of thousands. Below Dunbar's number of 150, peer accountability functions differently; researchers know who is carrying weight and who is not.

On the venture side, Gupta acknowledges a "spray and pray on acquihire floors" strategy is logically coherent — seeding pre-product teams with strong AI research credentials, betting that labs will absorb them regardless of product outcomes. He expects that arbitrage to close quickly given how competitive the venture market is. Sequoia's own mandate remains unchanged: back founders building enduring companies. He notes the firm's internal currency is getting a company name onto the conference room wall, something that does not follow from acquihire-oriented bets.

Gupta's macro framework at Sequoia is deliberately company-first. He points to Google's 1999 founding as evidence that macro timing is largely irrelevant at the early stage. The firm monitors valuation multiples relative to historical ranges to gauge how many years ahead they are paying, but the deeper filter is durability of compounding. His observation on what separates the largest outcomes, those reaching hundreds of billions or trillions in market cap, is that they did not sustain 100% growth the longest. They sustained 30% growth the longest. Meta, he notes, grew at roughly 22% on approximately $150 billion in revenue in the most recent year, a data point that no model in 2005 could have projected. The implication for portfolio construction is to weight conviction on long-run compounding over short-run growth optics or macro positioning.

On consumer agents and advertising, Gupta is candid about uncertainty. He references Andrej Karpathy's observation that the internet is currently built for humans and will eventually be rebuilt for agents, and expects entirely new ad units and monetization models to emerge. Platforms like Instacart, where Gupta previously served as an executive, will need to serve two distinct audiences simultaneously. He anticipates agent-facing ad formats will be dramatically more dynamic than anything currently deployed — potentially updating by the minute — but declines to offer a specific structural prediction. He notes Sierra, the customer experience AI company co-founded by Bret Taylor and Clay Bavor, where he sits on the board, as the clearest current proof point that agents are already resolving complex customer service interactions at scale.