OpenAI Sweetens Pay Packages as Meta Poaches Top Researchers

OpenAI, long known for luring gifted researchers with a mission-first mystique and eye-popping model launches, is now having to wield a more traditional weapon: cash. A leaked internal memo reviewed by The Information and confirmed by multiple employees says the San-Francisco lab is “recalibrating compensation across all technical ladders” after losing a mini-exodus of senior scientists to Meta’s Generative AI group in Menlo Park and London. Effective immediately, staff can access higher equity ceilings, four-year stock-refresh grants and an enlarged cash band that reportedly tops out at US $975,000 for principal researchers—numbers that start to resemble the pay packets once associated only with FAANG incumbents.

The timing is no mystery. In the past six months Meta has poached at least seven OpenAI and DeepMind alumni, including reinforcement-learning specialist Sharon Zhou, speech-generation lead Arun Narayan and the widely cited transformer-efficiency researcher Zhuohan Li. Sources inside Meta say the company’s new offers routinely exceed US $1.5 million a year in total comp for staff-level hires and balloon to US $5 million–US $7 million for director titles—plus a side bet that Meta’s soaring ad revenue will keep RSUs liquid. By contrast, OpenAI’s hybrid for-profit/non-profit cap table limits common-stock pools, leaving it more reliant on mission and brand to win résumés.

But mission alone is no longer enough in a world racing toward ever-larger frontier models. While GPT-5 pre-training is rumoured to be in final stages on a multimodal corpus, OpenAI insiders say the project’s schedule slipped twice this spring when the company scrambled to restaff model-evaluation and alignment teams. Sam Altman has publicly argued that a “modest brain drain” is normal as the field matures; privately, managers admit Meta’s wins have spooked recruiters in London, Lagos and Berlin, where start-ups see any OpenAI churn as a once-in-a-decade hiring window.

The new pay plan tries to plug those holes. First, equity ceilings for senior engineers rise from roughly 0.15 percent to 0.25 percent of diluted shares; second, a four-year refresh cycle kicks in after year two, bringing OpenAI in line with Apple and Nvidia; third, base-salary bands jump 15 percent across the board, with staff-plus researchers now eligible for a US $35,000 annual “compute stipend” they can allocate to side projects, academic collaborations or Kaggle challenges of their choosing. The stipend is novel—less about cash than about making sure talent doesn’t feel compute-starved compared to Meta’s or Google’s sprawling GPU clusters.

For readers in the United States and United Kingdom, the move underscores how the AI arms race has morphed into an HR slug-fest; recruiters are telling PhD candidates in Cambridge and Berkeley to expect bidding wars previously reserved for hedge-fund quants. In Germany, where the government is bankrolling sovereign LLM projects at Fraunhofer and TU Munich, the fear is that any uptick in U.S. comp will drain local labs unless Berlin fast-tracks equity-option reforms. Norwegian AI shops, buoyed by sovereign-wealth seed money, may actually benefit, marketing Oslo’s quality-of-life plus competitive cash as a calmer alternative to Bay-Area churn. And in Nigeria’s booming Lagos tech scene, where top AI graduates already command salaries unheard-of five years ago, OpenAI’s refresh grants raise the salary ceiling yet again, pressuring domestic firms to dangle dollar-linked offers or risk constant brain drain.

The bigger question is whether money alone can insulate OpenAI from attrition head-winds. Meta’s open-source gambit with Llama 3 has earned it deep goodwill among researchers who prefer public checkpoints over proprietary black boxes. Google DeepMind is whispering that its upcoming Gemini 2 will leapfrog GPT-4o in multilingual reasoning. Smaller labs like Anthropic and Mistral tout culture and agility rather than raw pay. Meanwhile, Sun Microsystems legend Bill Joy recently invested in a stealth startup promising “parallel foundation models” trained at tenth-scale cost, hinting that the next recruitment battlefield might not be ultra-large models at all but ultra-efficient ones.

Still, short-term optics matter. The leaked memo landed one day before OpenAI’s annual “Dev Day Europe,” signalling to attendees—and Wall Street—that the ChatGPT creator has the war chest (bolstered by Microsoft’s evergreen credit line) to keep marquee names in house. Whether that holds true six months from now may depend less on the size of pay cheques and more on the lab’s ability to ship genuinely differentiated leaps in intelligence. For now, at least, OpenAI has answered Meta’s talent raid with the bluntest instrument in Silicon Valley’s arsenal—cash plus options—proving that even the most mission-driven AI frontier shop has to play old-school compensation ball when the poachers come calling.