Can Scale AI and Alexandr Wang reignite Meta’s AI efforts?

Meta is reportedly investing practically $15 billion within the data-labeling agency Scale AI and taking a 49% stake within the startup, whereas additionally bringing on CEO Alexandr Wang to assist lead a brand new “superintelligence” lab throughout the firm.

The deal is paying homage to Meta’s earlier massive and dangerous bets, reminiscent of its $19 billion acquisition of WhatsApp and its $1 billion Instagram buy. On the time these mergers closed, many individuals urged Meta had grossly overpaid for the platforms — and in the present day’s discourse isn’t any completely different. There’s no scarcity of traders and founders who had been left scratching their heads this weekend over Meta’s newest tie-up.

In the long run, WhatsApp and Instagram grew to become an integral a part of CEO Mark Zuckerberg’s empire. The query is whether or not the Scale AI deal will equally work in Meta’s favor, proving Zuckerberg’s prescient technique but once more, or whether or not the corporate is greedy at straws in a misguided effort to catch as much as rivals like OpenAI, Google, and Anthropic.

On this case, Meta isn’t betting on an rising social media app, however on the information used to coach high AI fashions. For the final a number of years, main AI labs reminiscent of OpenAI have relied on Scale AI to supply and label knowledge that’s used to coach fashions. In current months, Scale AI and its knowledge annotation rivals have began hiring extremely expert folks, reminiscent of PhD scientists and senior software program engineers, to generate high-quality knowledge for frontier AI labs.

It might profit Meta to have a detailed relationship with an information supplier like Scale. Meta’s leaders have complained a few lack of innovation round knowledge within the firm’s main AI groups, in accordance with an individual conversant in the matter.

Earlier this yr, Meta’s generative AI unit launched Llama 4, a household of AI fashions that did not match the capabilities of Chinese language AI lab DeepSeek’s fashions and that was broadly seen as a disappointment. Not serving to issues, Meta is making an attempt to fight an attrition downside. In keeping with knowledge compiled by SignalFire, Meta misplaced 4.3% of its high expertise to AI labs in 2024.

Meta isn’t betting on solely Scale AI to reignite its AI efforts, but in addition on Wang to guide the aforementioned new superintelligence group. The 28-year-old CEO has confirmed himself to be a robust startup founder — he’s recognized round Silicon Valley to be bold, salesman, and really well-connected. Over the previous few months, Wang has been assembly with world leaders to debate AI’s affect on society.

Wang hasn’t led an AI lab of this type earlier than, nonetheless, and he doesn’t have the identical AI analysis background that many different AI lab leaders do, like Protected Superintelligence’s Ilya Sutskever or Mistral’s Arthur Mensch. That’s maybe why Meta can also be stated to be recruiting high-profile expertise like DeepMind’s Jack Rae to spherical out its new AI analysis group.

The post-acquisition destiny of Scale AI the corporate is a bit murky. The function of real-world knowledge in AI mannequin coaching is altering — some AI labs have introduced knowledge assortment efforts in-house, whereas others have elevated their reliance on artificial (i.e., AI-generated) knowledge. In April, The Info reported that Scale AI had missed a few of its income targets.

In keeping with Anyscale co-founder Robert Nishihara, a number of frontier AI labs are exploring novel methods to leverage and optimize knowledge, lots of that are fairly compute-intensive.

“Knowledge is a transferring goal,” Nishihara informed TechCrunch in an interview. “It’s not only a finite effort to catch up — you must innovate.”

It’s doable that Meta and Wang’s relationship might scare off different AI labs which have historically labored with Scale AI. In that case, this deal may very well be a boon for Scale AI’s rivals, reminiscent of Turing, Surge AI, and even nonconventional knowledge suppliers such because the lately launched LM Enviornment.

Turing CEO Jonathan Siddharth informed TechCrunch by way of e-mail that he’s acquired elevated curiosity from clients in gentle of the rumors round Meta’s take care of Scale AI.

“I feel there shall be some purchasers who will want to work with a accomplice that’s extra impartial,” he stated.

Solely time will inform how Meta’s funding will pan out for its AI efforts, however the firm has important floor to make up. In the meantime, the competitors isn’t slowing down. OpenAI is gearing up for the discharge of its subsequent flagship mannequin, GPT-5, in addition to its first brazenly accessible mannequin in years — a mannequin that’ll compete with Meta’s present and future Llama releases.