We tracked how many people leave Nvidia against how many join, month by month. In quiet months it's low. In January 2025, just 6 people left for every 100 who joined.
Then come the spikes. In July and August 2025, nearly 3 left for every 10 who joined. In July 2026, it happened again.
The timing isn't random, Nvidia pays much of its staff in shares, released every March, June, September and December. Three of the biggest spikes come straight after a payout.
Throughout 2026 so far, leavers relative to joiners has been rising.
Nvidia - Leavers as a Percentage of Joiners by Month
Now its founder Jensen Huang has resigned as CEO after 33 years, and CFO Colette Kress is interim CEO while the board searches for a successor. For many long-serving staff, working with Huang was part of why they stayed.
At Nvidia, the shares perk that is meant to keep people has become a reason to go.
The next payout is in a few weeks on 9 December.
The weeks after a payout are when its people move, and this December comes with a leadership vacuum attached. Watch this space.
Researchers leave top AI labs at twice the overall rate
At both Anthropic and OpenAI, research churn runs at twice the overall staff churn. The people the labs value most are the ones most likely to walk.
Over half of the researchers who left OpenAI joined just three rivals: Meta, Anthropic and Thinking Machines Lab.
Far fewer researchers leave Anthropic. Researcher departures there run at 11.1% of researcher hires, against 19.1% at OpenAI. Of those who do leave Anthropic, most join Google DeepMind.
Research Departures as a Percentage of Research Hires, 12-Month Rolling Average, 2024–2026
The data shows us that research talent at the top labs is mobile, but it moves between a handful of names.
Anthropic holds on to its researchers better than OpenAI, which matters as it heads into an IPO, and the labs are now hiring fewer researchers and more engineers to scale infrastructure.
AI is now a sales race, and Anthropic has fewer founders selling
At the big labs, sales hiring now matches technical hiring. Anthropic and OpenAI each add one go-to-market hire for every technical hire.
The app makers are pushing harder still. ElevenLabs and Lovable hire three to four go-to-market people for every technician.
Former founders are the chosen hires for the sales race. They build from zero to one and co-design products with clients, which makes them natural go-to-market hires. The app makers draw the most of them into these roles, while Anthropic draws the fewest.
Go-To-Market Hires Per Technical Hire, 12-Month Rolling Average, 2024–2026
The labs are hiring from start-ups to close the gap. Anthropic has hired from more than 350 companies with fewer than 50 staff, and OpenAI from more than 590.
The race to sell AI is now as fierce as the race to build it. Former founders are the go-to-market talent that start-ups win and the labs are still chasing, so the scale-ups that hire them first will be hard to catch.