Zuckerberg’s superintelligence bet: inside Meta’s biggest AI push

In 2025 Mark Zuckerberg made a decision that reshaped Meta: the company would aim not just for better AI assistants but for “superintelligence” — AI systems that outperform humans across many tasks. Since then Meta has hired aggressively, spent heavily and released new models. Here is what the bet involves.

Why Meta changed course

Meta had been an AI pioneer. Its research lab, founded in 2013, produced important work, and its Llama family of open-weight models, launched in 2023, was widely used by developers. But by 2025 many in the industry felt Meta had fallen behind rivals such as OpenAI and Google in the race for the most capable models.

Zuckerberg responded by reorganising the company’s AI work.

Meta Superintelligence Labs

In mid-2025 Meta created Meta Superintelligence Labs (MSL). To lead it, Meta made a large investment in the data-labelling company Scale AI and hired its founder, Alexandr Wang, as chief AI officer. Former GitHub chief executive Nat Friedman joined to lead product work.

Meta also recruited researchers from other leading AI labs with very large pay packages, which drew wide attention across Silicon Valley.

“Personal superintelligence”

In a public letter in 2025, Zuckerberg set out his vision of “personal superintelligence” — AI that helps each person achieve their own goals, rather than only automating work. He argued that devices such as smart glasses, which can see and hear what the user does, will be the main way people use such AI.

On Meta’s results call for the third quarter of 2026, he repeated that the company intends to build superintelligence through scaling — training ever larger models on ever larger computing clusters.

New models in 2026

In October 2026 MSL released new models under the Muse name:

• Muse Wonder, released on 12 October;
• Muse Spark Open, an open-weight model, released on 15 October.

They power the Meta AI assistant across Facebook, Instagram, WhatsApp and Meta’s glasses.

The data-centre race

Superintelligence needs enormous computing power. Meta is building some of the largest data centres in the world in the United States, including the Prometheus project in Ohio and Hyperion in Louisiana, and has raised its capital spending sharply. Investors watch these costs closely, and analysts disagree about when the investment will pay off.

Hardware: glasses

Meta’s AI push is closely linked to its hardware. Ray-Ban Meta smart glasses, made with EssilorLuxottica, have become a surprise hit, and in 2025 Meta introduced glasses with a built-in display. Zuckerberg believes glasses will eventually replace phones for many tasks.

Open or closed?

Meta has long championed open-weight AI models, arguing that openness makes AI safer and spreads its benefits. With the most powerful new systems, Meta has said it will be more careful about what it releases openly. The release of Muse Spark Open shows the company still wants a role in open AI.

Concerns

Critics raise questions about:

• safety — how very powerful AI should be tested and controlled;
• privacy — what data from Meta’s apps and glasses is used;
• cost — whether the spending is sustainable;
• talent wars — the effect of huge pay offers on the wider research community.

What it means

Zuckerberg has bet big before: on mobile in 2012, which succeeded, and on the metaverse from 2021, which has been slower to pay off. The superintelligence push is his biggest bet yet. Its outcome will help decide whether Meta remains one of the leaders of the technology industry in the age of AI.