In June 2025, Mark Zuckerberg reorganized Meta’s whole AI effort around a single word: superintelligence. The new division, Meta Superintelligence Labs (MSL), came with one of the most expensive hiring sprees in tech history and a bold goal, “personal superintelligence for everyone.” Here is what it is, why it exists and what critics say.
Why Meta created a superintelligence lab
By early 2025 Meta had a problem. Its open Llama models had made it a leader in open-source AI, but the Llama 4 launch in April 2025 got a lukewarm response, and OpenAI, Google and Anthropic seemed to be pulling ahead at the frontier.
Zuckerberg’s answer was to stop competing incrementally and bet the company on superintelligence. That meant:
- Buying talent and data through a huge investment in Scale AI.
- Recruiting the best researchers in the world, whatever it cost.
- Putting all AI work under one new organization with a single goal.
The Scale AI deal
In June 2025 Meta agreed to invest about $14 billion for a large minority stake (reported at 49%) in Scale AI, the company that supplies labeled training data to most major AI labs. As part of the deal, Scale’s 28-year-old founder Alexandr Wang joined Meta to lead its superintelligence push as chief AI officer. Former GitHub CEO Nat Friedman also joined in a senior role.
Some of Scale’s other customers reportedly pulled back afterward, wary of a supplier now partly owned by a competitor.
The talent war
MSL became famous for its recruiting. Meta hired researchers from OpenAI, Google DeepMind, Apple and Anthropic, and some offers were reported at tens or even hundreds of millions of dollars over several years. OpenAI’s Sam Altman publicly criticized the approach, while Zuckerberg argued that a small, elite team with huge compute was the fastest path to the frontier.
Not everyone stayed. Several high-profile hires left within months, and the lab was reorganized more than once in 2025. That is a reminder that money alone doesn’t guarantee a coherent research culture.
What is “personal superintelligence”?
In a July 2025 letter, Zuckerberg set out Meta’s vision. He contrasted two futures:
- Centralized superintelligence: AI that automates most valuable work, with people living on its output.
- Personal superintelligence: AI that knows you deeply, understands your goals and helps you “achieve your goals, create what you want to see in the world… and grow to become the person you aspire to be.”
Meta says it is building the second. A key part of the plan is devices, especially smart glasses like Ray-Ban Meta, which can see what you see and hear what you hear. Zuckerberg has said people without such glasses may be at a “significant cognitive disadvantage” in the future.
Zuckerberg also signaled that Meta might not open-source its most powerful future models, a shift from its earlier Llama strategy, citing safety concerns.
How MSL is organized
Reporting during 2025 described MSL as several groups under one umbrella:
| Group | Focus |
|---|---|
| Frontier research team (“TBD Lab”) | Building the next generation of frontier models |
| FAIR | Meta’s long-running fundamental AI research lab |
| Products & applied research | Bringing models into Meta AI, WhatsApp, Instagram and Facebook |
| Infrastructure | Data centers and compute |
The compute behind it is enormous. Meta has announced multi-gigawatt data center projects, including Prometheus in Ohio and Hyperion in Louisiana, and capital spending running to tens of billions of dollars a year.
The criticism
- Hype vs reality. Critics argue that “superintelligence” is a branding exercise for what are still chatbots and recommendation engines. (The same debate now surrounds the US government’s use of “SI”.)
- Privacy. An AI that “knows you deeply” and sees through your glasses would hold an unprecedented amount of personal data, at a company whose business model is advertising.
- Safety. Yann LeCun, Meta’s longtime chief AI scientist, was openly skeptical of superintelligence risk talk. Safety advocates, including signers of the Statement on Superintelligence, worry about any lab racing to build it.
- Money. Investors have questioned whether the spending will pay off.
Meta vs the other superintelligence labs
| Lab | Founded / pivot | Approach |
|---|---|---|
| Meta Superintelligence Labs | June 2025 | Personal superintelligence via Meta apps and glasses |
| Safe Superintelligence Inc. (SSI) | June 2024 | Research only, “straight shot” to safe superintelligence, no products |
| OpenAI | 2015 | AGI first, then superintelligence. Products like ChatGPT fund research |
| SpaceXSI (formerly xAI) | 2023 / renamed 2026 | Musk’s lab, now part of SpaceX (details) |
The bottom line
Meta Superintelligence Labs is Mark Zuckerberg’s all-in bet that superintelligence is coming soon, and that Meta must be one of the companies building it. It brought record-breaking pay, a $14 billion Scale AI deal and a distinctive “personal superintelligence” pitch built around Meta’s apps and glasses. Whether it produces true artificial super intelligence or just better assistants is one of the big open questions of the AI race.
Enjoyed this? Get the SI Briefing
One short email a week: what changed on the road to super intelligence, and why it matters. No spam, unsubscribe anytime.
Frequently asked questions
What is Meta Superintelligence Labs?
Meta Superintelligence Labs (MSL) is Meta’s AI division, created in June 2025 to pursue superintelligence. It brought Meta’s AI research, foundation models and product teams under one organization.
Who leads Meta Superintelligence Labs?
It was set up under Alexandr Wang, the Scale AI founder who became Meta’s chief AI officer, with former GitHub CEO Nat Friedman also in a senior role at launch.
What is personal superintelligence?
It is Mark Zuckerberg’s term for AI that knows each user deeply and helps them achieve their own goals, delivered through devices such as smart glasses, as opposed to centralized AI that automates work.
How much did Meta spend on superintelligence?
Meta invested about $14 billion for a large stake in Scale AI, offered researchers very large compensation packages and has planned hundreds of billions of dollars in AI data center spending.
Has Meta built superintelligence?
No. Meta Superintelligence Labs is working toward that goal, but no company, Meta included, has built a system that meets the definition of superintelligence.