In short
The departure of four leading Google researchers appears to be a classic case of brain drain, but the terms of the deal paint a different picture. We explore why the company was able to transform a former internal team into an external startup while retaining access to the research results and computing resources.
The departure of four key researchers would normally represent a painful brain drain for Google. But this time, the company didn’t just let the team go: it became a founding investor in Discovery Loop, retained a stake, and agreed to serve as a cloud partner.
On August 5, Jeff Dean, Sanjay Gemawat, Oriol Vinyals, and Quoc Le left Google. For Dean, this marks the end of his 27-year tenure at the company. All four are founding a new startup.
The unusual part of the story is that Google seems to have moved its research lab outside its own structure, but retained several key levers for itself. For at least the first year, Discovery Loop will have access to Google’s computing power, and Google itself will contribute to the project not only with funding but also with cloud infrastructure.
This doesn’t seem like a typical case of employees leaving for a competitor, but rather a limited-risk bet: Google loses direct control over the team but gains the opportunity to participate in what it creates going forward. For the researchers, this likely means more freedom, and for the company, it’s a chance to maintain a connection with a strong team.
However, the details of the deal remain unknown: the size of the investment, Google’s stake, and the terms of the partnership after the first year have not been disclosed. Therefore, it’s too early to call this a guaranteed success. So far, only the approach itself is clear—transforming a former in-house team into an independent project without severing financial and infrastructure ties with it.
If a research team leaves but their former employer continues to pay for their cloud services and retains a stake, is that talent retention or a new model for a corporate lab? Source: All Articles / Machine Learning / Habr