Megan Skiendel: Zara, long week — tell me you have something good for me.
Zara Reyes: I have something unhinged for you. Six days. I need you to hold that number.
Megan Skiendel: Six days what?
Zara Reyes: Six days in late June 2026. Four senior researchers leave Google DeepMind. We're talking John Jumper — Nobel Prize in Chemistry, 2024, led AlphaFold — gone to Anthropic. Jonas Adler, Gemini coding, Anthropic. Alexander Pritzel, Gemini training systems, Anthropic. And Noam Shazeer, like, co-author of the actual Transformer paper, he bounces to OpenAI. All of this while Demis Hassabis is standing at an event in Cannes telling people Google wins its 'fair share' of talent.
Megan Skiendel: Wait — a Nobel laureate left within a year of winning?
Zara Reyes: Within a year. And that's the thing — this isn't attrition, it's not a slow drain. Four people from the same flagship project, Gemini, in six days? That's a cultural moment crystallizing in public. Researchers watched each other leave and went — yeah, that's what winning looks like now. That's brand momentum, and it's moving away from Google.
Megan Skiendel: Okay, but — I want to pump the brakes on the brand death framing for a second. Because I think there's something more specific doing the work here. It's not vibes. It's math.
Zara Reyes: What's the math?
Megan Skiendel: Anthropic just raised at a $965 billion valuation. They're considering an IPO as soon as fall 2026. So — honestly, think about it like buying a house before the subway stop gets announced. That's the whole thing. You get in now, the neighborhood reprices, you didn't work for Alphabet for thirty years accumulating RSUs that are already baked into a two-trillion-dollar market cap.
Zara Reyes: Right — Alphabet stock literally rose four percent the same week these people left. On its Dow Jones debut. Stability is the problem.
Megan Skiendel: Exactly. That's the paradox. And the SignalFire data — eleven times more likely to flow toward Anthropic than back — that's not brand sentiment, that's people running a personal finance calculation and arriving at the same answer.
Zara Reyes: No but — wait, who's writing the checks to Anthropic?
Megan Skiendel: Google is. Google is a partial investor in Anthropic. They are literally — I mean, this is the part that should be a scandal — they are funding the entity using pre-IPO equity to pull their own researchers out the door. Demis Hassabis is in Cannes saying they win their fair share while Alphabet's own capital is subsidizing the competition.
Zara Reyes: No but that's actually where I want to push, because — okay, the equity story is real, but I think it's downstream of something more specific. DeepMind pivoted hard into coding AI. Like, that's a resource allocation decision. Compute flows toward the coding work, and the researchers who came there to push the actual frontier — Adler, Pritzel, Jumper — they're watching their mandates narrow.
Megan Skiendel: The coding pivot drove six more departures. Not just those four — six additional researchers went to Meta, OpenAI, Anthropic combined.
Zara Reyes: Right. And that's — wait, that's the Tuesday morning mechanism. A senior researcher opens their calendar, sees their compute allocation got cut again, checks Slack, sees Pritzel just posted from Anthropic. That's not a slow drift. That decision crystallizes fast.
Megan Skiendel: And Shazeer going to OpenAI proves it's not just Anthropic's pull.
Zara Reyes: Exactly — same IPO upside story, totally different choice. If this were purely pre-IPO math, everyone ends up at the same place. They're scattering. Meta, OpenAI, Anthropic. That scatter pattern tells you the push is real.
Megan Skiendel: Honestly, that's where the SignalFire eleven-to-one number stops being just a headline. It's a compute dispute quantified.
Megan Skiendel: And honestly — the part that I keep tripping over is that Anthropic's own footing isn't that solid. They're in the middle of renegotiating their Amazon deal right now, like, actively, over token-based pricing. That's not a company with nothing but upside. And researchers are still choosing it over a two-trillion-dollar balance sheet.
Zara Reyes: No but that's — wait, that's actually the sharpest version of this whole story. It's not that Anthropic is a sure thing. It's that Google built conditions where even an uncertain bet somewhere else lands better than a certain one at home. Dario Amodei is running contract disputes with Amazon and still winning the talent math.
Megan Skiendel: Which means Google's problem isn't equity. It's that it wrote the playbook — transformer architecture, AlphaFold, the whole thing — and now everyone else is running it back at them. Funded, partly, by Google.
Zara Reyes: Lowkey the most expensive irony in tech right now. Good talk.
