Key Takeaways
- 💥 OpenAI CRO Mark Chen described Meta's talent poaching as feeling like “someone has broken into our home” in an emotional Slack memo
- 💰 Meta offered $100M+ signing bonuses to critical OpenAI researchers, with Zuckerberg personally recruiting for Meta’s superintelligence lab
- 🔁 OpenAI is “recalibrating comp” and creating “creative ways to reward top talent” to counter offers
- 🧠 At least 8 OpenAI researchers joined Meta in June 2025, including Trapit Bansal (o1 reasoning model architect)
- ⚖️ Chen vowed retention efforts won’t compromise “fairness to others” amid intense workload complaints (80-hour weeks)
The “Home Invasion” Memo That Shook OpenAI
Mark Chen didn’t mince words. When four senior researchers left for Meta’s superintelligence lab, OpenAI’s Chief Research Officer sent a Saturday Slack message that hit like a thunderclap. “I feel a visceral feeling right now, as if someone has broken into our home and stolen something,” he wrote . This weren’t just corporate speak—it was raw, human reaction to what felt like a violation.
Chen promised teams leadership wasn’t “sitting idly by”. He and Sam Altman were working ’round the clock talking to those with offers. “We’ve been more proactive than ever before,” he stressed, revealing plans to “recalibrate comp” and find “creative ways to recognize top talent” . But he also drew a line: “While I’ll fight to keep every one of you, I won’t do so at the price of fairness to others” .
The timing stung. OpenAI was about to shut down for a recharge week when Meta pounced. One leader warned in Chen’s memo: “Meta knows we’re taking this week... and will take advantage” .
$100 Million Bonuses and Zuckerberg’s Personal Pitches
Let’s talk numbers. Sam Altman publicly exposed Meta’s jaw-dropping offers on his brother’s podcast: “They started making giant offers... like $100 million signing bonuses, more than that in compensation per year” . Multiple OpenAI sources confirmed these figures to WIRED .
Zuckerberg ain’t delegating this talent war. He’s:
- Personally reaching out to targets 📞
- Leading Meta’s new AI superintelligence unit after a $14.3B stake in Scale AI brought its CEO, Alexandr Wang
- Hunting “top talent from OpenAI and Google” while avoiding Anthropic due to “culture fit” issues
Altman claimed “none of our best people” accepted yet . But that confidence wobbled when Trapit Bansal—key architect behind OpenAI’s o1 reasoning model—joined Meta days later .
OpenAI’s Three-Pronged Counterattack
Facing existential pressure, Chen’s memo outlined OpenAI’s retention strategy:
- Comp Recalibration: Adjusting salaries/bonuses to match or exceed Meta’s sky-high offers
- Executive Engagement: Altman and Chen doing 1:1 talks with wavering staff
- “Creative Rewards”: Unspecified incentives recognizing contributions beyond cash
Table: OpenAI vs. Meta Talent Tactics
Research leaders also urged targets to resist “exploding offers” (short deadlines forcing rushed decisions). “Tell them to back off,” one wrote in Chen’s memo .
Who Did Meta Actually Poach?
Names matter. Here’s the confirmed brain drain:
- Lucas Beyer, Alexander Kolesnikov, Xiaohua Zhai: Zurich-based researchers hired in late June
- Trapit Bansal: Crucial to OpenAI’s o1 reasoning model; joined Meta’s reasoning team
Bansal’s exit cuts deep. His work on reinforcement learning alongside Ilya Sutskever helped shape OpenAI’s technical edge . At Meta, he’ll build “frontier AI reasoning models” alongside ex-Google/DeepMind hires.
Meta’s also tried acquiring whole startups like Safe Superintelligence (Sutskever’s new venture) and Perplexity, though talks stalled .
AGI vs. “Skirmishes”: The Philosophical Split
Beneath the money war lies a culture clash. Chen’s memo dismissed Meta’s moves as “side quests” distracting from OpenAI’s “main quest”: achieving artificial general intelligence (AGI) .
Altman echoed this on “Uncapped”: “Meta will be... saying ‘we’re just gonna copy OpenAI.’ That basically never works... You don’t build a culture of innovation” .
Meta’s retort? Scale AI’s Newman told CNBC: “They built the rails for open-source AI... so much is built on Meta” . Zuckerberg’s betting open-source models like Llama can out-innovate closed systems if given elite talent.
80-Hour Weeks and “Recharge” Breaks
Burnout’s the elephant in the server room. OpenAI staff reportedly grind 80-hour weeks . The scheduled shutdown (while leaders kept working) hints at unsustainable pressure—something Meta exploited during the break.
One leader’s plea in Chen’s memo acknowledged this: “If you’re feeling pressure, don’t be afraid to reach out” . Whether “recalibrating comp” includes work-life balance fixes remains unaddressed.
What “Fairness” Means in a $100M Talent War
Chen’s “fairness” line is fascinating. Does he mean:
- Not overpaying some stars while others feel undervalued? 💸
- Avoiding bidding wars that distort internal equity?
- Ethical limits to retention tactics?
Unclear. But it signals OpenAI won’t match Meta at any cost. Some see this as principled; others as a fatal handicap.
The Supercomputer Arms Race Continues
Amidst the chaos, Chen reminded staff: “A lot more supercomputers are coming online later this year” . Compute power remains core to OpenAI’s strategy—something bonuses can’t buy.
Meta’s countermove? Loading its new lab with ex-Google DeepMind talent like Jack Rae and ex-Scale AI engineers . The goal: Close the reasoning model gap with OpenAI and DeepSeek’s R1 .
FAQs
Q: Did any OpenAI researchers actually accept Meta's $100M offers?
A: Yes. At least eight joined in June 2025, including Trapit Bansal (core to o1 reasoning models) and three Zurich researchers .
Q: How is OpenAI fighting back?
A: Three ways: 1) Salary/bonus adjustments, 2) Altman/Chen personal retention talks, 3) Unspecified “creative rewards” .
Q: Is Zuckerberg personally involved?
A: Yes. He’s directly calling targets and leads Meta’s new superintelligence lab .
Q: Does OpenAI really work 80-hour weeks?
A: Per internal sources, yes—prompting recharge shutdowns. Meta reportedly exploited one such break to pressure staff .
Q: What’s “recalibrating comp”?
A: OpenAI’s adjusting pay structures to compete with Meta’s massive offers, though details remain confidential .
Q: Who did Meta fail to recruit?
A: Altman claimed “none of our best people” left initially, but that shifted. Meta also couldn’t acquire startups like Safe Superintelligence .
The real battle isn’t over bonuses. It’s which vision wins: OpenAI’s focused AGI quest or Meta’s open-source empire. For talent caught in between, 2025’s gonna be a wild ride.
Comments
Post a Comment