On Tuesday, Jeff Dean, Sanjay Ghemawat, Quoc Le, and Oriol Vinyals — four engineers whose work forms the load-bearing walls of Google’s empire — announced they were leaving the company after a combined century-plus of tenure to start Discovery Loop, a public benefit corporation that promises to automate scientific research with AI. The funding, co-led by Radical Ventures and Khosla Ventures, includes Lightspeed, Kleiner Perkins, Doerr Capital, and, in the detail that should have led every headline, Alphabet itself.

Read that again. The parent company is writing a check to the startup its own legends are fleeing to. This is not a defection. It is a restructuring, and it reveals more about the future of industrial R&D than any amount of hand-wringing about brain drain ever could.

The Cap Table Is the Story

The narrative that writes itself is the one about talent walking out the door. Dean spent 27 years at Google. Ghemawat was his collaborator for more than two decades. Vinyals and Le are not exactly replaceable either. When four people of this caliber leave simultaneously, the instinct is to frame it as a verdict on the mothership — too slow, too bureaucratic, too allergic to the kind of long-term research that made it great.

But Alphabet is on the cap table. It is also, according to the announcement, providing compute. So the company is simultaneously losing four of its most valuable researchers and paying to rent them back their own infrastructure. If that sounds like a contradiction, it is only because we are still using the old vocabulary. This is not a departure. It is a spin-out with a venture patina, and it is likely to become the template for how the AI giants handle the next decade of foundational research.

A partner at a rival venture firm, texting after the news broke because his fund had passed on the deal, put it plainly: “Alphabet gets the equity upside without the quarterly earnings drag of a moonshot division. The founders get autonomy and founder equity. The only people who lose are the shareholders who thought Google was still doing this work internally.”

The Public Benefit Corporation Fig Leaf

Discovery Loop is incorporated as a Delaware public benefit corporation. That designation has become the standard costume for AI startups that want to signal seriousness about societal impact while retaining the freedom to do whatever the board decides is in the “public benefit.” It is a legal structure that allows a company to pursue a mission without the fiduciary handcuffs of pure profit maximization — which, in practice, means it can burn cash on research that might never ship a product without getting sued.

For Alphabet, this is the point. Google’s own research arms have spent years under pressure to show product impact. DeepMind, before the merger with Google Brain, was a money furnace that produced breathtaking papers and negligible revenue. The public benefit corporation structure gives Discovery Loop a formal permission slip to be the same kind of money furnace, except this time the losses sit on someone else’s consolidated financials. Alphabet gets the option value. The public gets the papers. The founders get rich either way.

The Real Restructuring Is Happening Off the Balance Sheet

There is a version of this story in which Discovery Loop is a genuine startup — lean, hungry, unshackled from corporate overhead. But a startup founded by four Google Fellows with Alphabet as an investor and compute provider is not lean in any meaningful sense. It is a wholly owned subsidiary of the Google extended universe, just with a different legal address and a compensation structure that would be impossible inside a public company.

This is not a criticism. It is probably the only way to do the kind of work Discovery Loop is proposing. The company’s pitch — AI systems that propose experiments, run them, learn from the results, and iterate recursively — is the kind of thing that requires years of patience and a tolerance for failure that quarterly capitalism does not permit. The question is not whether the work is valuable. The question is who gets to claim it, and on what terms.

What Tuesday’s announcement really signals is that the AI industry’s center of gravity is shifting from internal R&D labs to a network of founder-led, venture-funded, corporate-backed entities that look like startups but function like the old Bell Labs — if Bell Labs had been structured to make its researchers billionaires. The talent isn’t leaving the building. The building is being redrawn around the talent, and the rest of us are still reading the old floor plan.

Sources