Meta is building a new lab for superintelligence

Meta is going nuclear in the AI talent war, and the price tag has officially hit nine figures.

According to a report from The Indian Express citing four people familiar with the plans, Mark Zuckerberg is building a brand-new AI lab dedicated entirely to pursuing "superintelligence"—hypothetical systems that exceed the capabilities of the human brain. To lead the charge, Meta is bringing in Scale AI's 28-year-old founder and CEO, Alexandr Wang, as part of a multi-billion-dollar deal that could also bring over key talent from his startup.

The nine-figure talent raid

Meta isn't just poaching Wang. The company has reportedly dangled compensation packages ranging from seven to nine figures to dozens of top researchers at rival outfits like OpenAI and Google, with several already agreeing to jump ship.

To sidestep antitrust regulators, Meta is in talks to invest billions directly into Scale AI in an unusually structured deal, echoing similar acqui-hire maneuvers recently pulled by Microsoft and Google. The Federal Trade Commission is currently taking Meta to trial over its historic Instagram and WhatsApp acquisitions, meaning an outright buyout was dead on arrival.

Wang, who previously roomed with OpenAI CEO Sam Altman, has built Scale AI into an industry-critical data powerhouse that labels and cleans training data for players like OpenAI and Microsoft. In January, Wang even bought an ad in The Washington Post urging Donald Trump to accelerate AI investments so America can "win the AI war."

Management turmoil and benchmark fallout

This aggressive reshuffle comes on the heels of deep internal turbulence inside Meta. Despite Zuckerberg boasting in May that Meta AI reached over 1 billion monthly active users, the AI division has been bleeding talent due to brutal product cycles and leadership friction.

Meta also suffered an embarrassing blow in April after outside researchers caught the company juicing benchmark scores for newly released Llama models to make them look more capable than they actually were. Zuckerberg was reportedly furious that the community believed Meta was deliberately covering up lackluster results.

Meanwhile, Meta's long-serving Chief AI Scientist Yann LeCun has consistently maintained that current LLM architectures will never reach AGI and that entirely new technical foundations are required. Zuckerberg's new superintelligence lab signals he is no longer willing to wait for academic breakthroughs alone.

Why it matters

For developers relying on Meta's open-source Llama models, this shift raises critical questions. Meta gained massive goodwill by giving away frontier-adjacent models for free, but pursuing superintelligence with nine-figure salaries requires astronomical capital expenditure. If Meta feels forced to pivot toward proprietary frontier systems to justify these investments, the golden age of open-weight LLMs could face serious headwinds.

For the broader industry, Zuckerberg just detonated compensation standards across Silicon Valley. Offering up to hundreds of millions in compensation to individual researchers will force OpenAI, Google, and Anthropic into defensive bidding wars, squeezing smaller labs entirely out of the premier talent market.

We think Meta's approach is technically shrewd because acquiring Scale AI's operational DNA solves one of Meta's biggest bottlenecks: high-quality proprietary data pipelines. However, the cultural hazard is massive. Dropping an outside wunderkind and heavily compensated mercenaries into a company already wrestling with internal strife and LeCun's faction could ignite even worse infighting before any superintelligence sees the light of day.