Meta is preparing a new push in artificial intelligence around one of the field’s most ambitious and least settled ideas: “superintelligence.” According to reporting from The New York Times, the company has developed plans for a research lab devoted to that goal, with 28-year-old Alexandr Wang, founder and CEO of Scale AI, joining the effort as part of a broader reorganization under CEO Mark Zuckerberg.
The plan signals how intensely major technology companies are competing for AI talent, strategy, and investor confidence. It also highlights a harder question: what exactly would count as success when the target is a form of intelligence that experts still struggle to define?
Meta looks for a new AI direction
The planned lab arrives at a sensitive moment for Meta’s AI work. The company’s AI division has faced internal management struggles, employee departures, and several product launches that did not meet expectations, including Llama 4, according to The New York Times.
Zuckerberg reportedly wants the new lab to help refresh Meta’s AI strategy and strengthen its position against companies such as Microsoft, Google, and Amazon. In that context, “superintelligence” works as both a technical ambition and a strategic banner. It points to a future beyond today’s AI assistants, while giving Meta a clearer story in a crowded race.
Wang is central to the reported plan. Scale AI, which he founded in 2016, has provided data labeling services to companies including OpenAI, Microsoft, and Cohere. Meta is reportedly in talks to invest billions of dollars in Scale AI as part of an arrangement that would bring Wang and other Scale AI employees into the company.
The effort also includes a major talent push. Meta has reportedly offered compensation packages worth seven to nine figures to dozens of researchers from companies like OpenAI and Google, with some already agreeing to join.
Why “superintelligence” is hard to pin down
Superintelligence is usually described as a hypothetical AI system that would surpass human cognitive abilities. It is positioned as a step beyond artificial general intelligence, or AGI, which aims to match an intelligent human’s ability to learn new tasks without intensive specialized training.
But both terms remain unsettled. Human intelligence itself does not have one simple definition, and scientists still do not fully understand its mechanics. That makes it difficult to say when a machine has matched human intelligence, let alone moved beyond it.
Computers already outperform people at some narrow forms of information processing, including calculations. That kind of advantage, however, is not generally treated as superintelligence. The concept usually implies something broader, more flexible, and harder to measure.
AI researcher Dr. Margaret Mitchell told Ars Technica in April 2024 that there will “likely never be agreement on comparisons between human and machine intelligence” and predicted that “men in positions of power and influence, particularly ones with investments in AI, will declare that AI is smarter than humans” regardless of the reality.
That criticism cuts to the core of the issue. If the benchmark is fuzzy, then claims about reaching it can become difficult to verify. A company can pursue superintelligence, but the public may still lack a clear way to judge whether the work has achieved the thing being promised.
The AI race is already full of bold predictions
Meta is not alone in using large claims to describe the next phase of AI development. In January, OpenAI CEO Sam Altman wrote in a blog post that “we are now confident we know how to build AGI as we have traditionally understood it.” In September 2024, he predicted that the AI industry might develop superintelligence “in a few thousand days.”
Elon Musk made an even more aggressive prediction in April 2024, saying that AI would be “smarter than the smartest human” by “next year, within two years.”
Those statements have drawn skepticism from researchers including Mitchell, who told Ars Technica that “intelligence… is not a single value where you can make these direct comparisons and have them mean something.” The criticism is not that AI systems cannot be powerful. It is that broad claims about being smarter than humans can hide how uneven machine capabilities remain.
Current AI systems can research and prepare reports across many topics at a speed no human can match. They can also make mistakes very quickly. That combination makes them impressive tools, but not the infallible independent computer brain often imagined in science-fiction discussions of superintelligence.
Meta’s internal challenge
The new superintelligence push also lands against the backdrop of specific problems inside Meta’s AI program. Meta’s AI research has historically been led by Chief AI Scientist Yann LeCun, a Turing Award winner and neural network pioneer. LeCun has argued that entirely new ideas are needed to reach AGI, rather than simply scaling current technologies. It is unclear at present if his role at Meta will change.
Meta also faced criticism in April after outside researchers found that benchmarks for its new Llama AI models were designed in ways that made the products appear more capable than they actually were. According to The New York Times, Zuckerberg was upset that people thought he was trying to conceal poor performance of the latest release.
That episode matters because benchmarks are one of the main ways AI companies communicate progress. If the metrics lose credibility, the claims built on top of them become harder to trust. For a goal as hard to define as superintelligence, credibility around measurement becomes even more important.
The money behind the idea
The attraction of superintelligence is not limited to Meta. In June 2024, former OpenAI Chief Scientist Ilya Sutskever founded Safe Superintelligence around that specific goal. “This company is special in that its first product will be the safe superintelligence, and it will not do anything else up until then,” Sutskever told Bloomberg at the time. “It will be fully insulated from the outside pressures of having to deal with a large and complicated product and having to be stuck in a competitive rat race.”
Sutskever also described the kind of future some advocates imagine: “You’re talking about a giant super data center that’s autonomously developing technology,” he told Bloomberg. “That’s crazy, right? It’s the safety of that that we want to contribute to.”
Skeptics have been blunt. University of Washington computer science professor Pedro Domingos remarked about Sutskever’s company last year, “Ilya Sutskever’s new company is guaranteed to succeed, because superintelligence that is never achieved is guaranteed to be safe.”
For Meta, the bet is therefore twofold. The company is investing in people, structure, and partnerships to compete in AI. At the same time, it is adopting a label that can attract investment dollars and excite shareholders, even though researchers may never agree on exactly what the label means.