Can Mega Salaries Build a World‑Class AI Team? Lessons from Meta, Anthropic and DeepMind

Meta has been pouring unprecedented capital into AI, allocating $660‑720 billion to data centers and AI infrastructure while offering up to $200 million per hire to poach top researchers from OpenAI, Apple and Google.

Despite this spending, Meta faces serious challenges: a lawsuit over the use of 2,396 pirated adult films for training its MovieGen and LLaMA models, high turnover among AI leadership, and a lower engineer net‑add rate and retention compared with Anthropic and OpenAI.

01 Zuckerberg’s "Superintelligence" Bet

Zuckerberg aims to assemble a 50‑person elite "Superintelligence" group to close the frontier gap, offering salaries that can exceed $200 million per person and recruiting entire teams from competitors.

02 Hiring ≠ Strength – Look at Anthropic, DeepMind, DeepSeek

Anthropic retains talent through a clear mission, safety‑first culture and high psychological safety, achieving ~80 % employee retention without relying on sky‑high salaries.

DeepMind/FAIR pursues long‑term research goals, stable leadership and abundant compute, enabling breakthroughs like AlphaGo and AlphaFold, but frequent reorganizations at Meta threaten research continuity.

DeepSeek demonstrates that a small, flat team can deliver high‑performance models through engineering efficiency and reproducibility, even with limited compute resources.

The common lesson is that talent acquisition is an accelerator, not the engine; without a stable, mission‑driven environment, even the brightest hires cannot sustain breakthrough AI research.

Recommendations include establishing a three‑year "North Star" focused on long‑term challenges such as video generation consistency or tool‑using AI, providing continuous research windows of 18‑24 months, and emphasizing open‑source milestones to attract broader community contributions.

Ultimately, Meta must shift from short‑term KPI pressure to building a resilient research ecosystem that can consistently produce world‑class AI advances.