Meta Stock Looks to Build on Record High. A New AI Unit Could Hold the Key.

Meta Platforms stock has risen 45% over the past 12 months. (Justin Sullivan/Getty Images)

Meta Platforms was looking to build on its stock rally but wasn’t having much success Tuesday. For extending its success, the social-media company is relying on new AI unit.

In midmorning trading, shares of Meta were down 0.6% at $733.46. On Monday, they closed at a record high of $738.09, valuing the parent of Facebook and Instagram at $1.86 trillion.

Further gains will depend on Meta’s new “Superintelligence Labs” unit, led by former Scale AI CEO Alexandr Wang and former GitHub CEO Nat Friedman. Meta CEO Mark Zuckerberg launched the project Monday.

The formation of the unit follows on Meta’s $14 billion investment in Scale AI earlier this month, giving it a 49% stake in the data-labeling start-up, according to The Wall Street Journal.

To build the “Superintelligence” team, Meta hired AI researchers from Microsoft-backed OpenAI, Amazon.com-backed Anthropic, and Alphabet Google DeepMind. Zuckerberg offered pay of up to $100 million for star talent in the recruitment drive, according to OpenAI’s Sam Altman.

One question about the hires is where they leave Meta’s current Chief AI Scientist Yann LeCun, a prominent industry researcher and noted skeptic of the capacities of current AI models. Meta didn’t immediately respond to a request for comment about LeCun’s role Tuesday.

There have been signs that Zuckerberg feels Meta’s in-house AI talent needs bolstering. Meta’s “Behemoth” AI model, originally meant to be released in April, has been delayed until fall or later, according to the Journal.

A number of AI companies have slowed their rollout plans for new models amid concerns they don’t show sufficient improvement. That suggests the industry’s “scaling law,” the idea that larger and more-complex models are automatically more intelligent, is breaking down.

LeCun has publicly advocated for a different type of AI to existing large-language models, incorporating an internal model of how the world works so that they can learn and adapt more quickly. Last month, Meta presented its latest model based on those principles, called V-JEPA 2, making it available for commercial and research applications.

Write to Adam Clark at adam.clark@barrons.com