Table of Contents
AI Superintelligence Labs in 2026: What Changed
Meta’s Superintelligence Labs has evolved beyond recruitment into active AGI development, with their first proprietary models achieving 95% accuracy on complex reasoning benchmarks. The lab now operates three specialized divisions: cognitive architecture, multimodal intelligence, and safety alignment. Following successful poaching campaigns, MSL expanded to 180 researchers across facilities in Menlo Park, London, and Tel Aviv. The division’s $12 billion annual budget rivals entire national AI programs, while their recent breakthrough in neural architecture search has accelerated development timelines by an estimated 18 months.
Key Metrics
| Total MSL Researchers | 180 (up from 20 in launch) |
| Annual R&D Budget | $12.4 billion |
| Compute Infrastructure | 2.1 million H100 GPUs equivalent |
| Patent Applications Filed | 847 (AGI-related, 2025-2026) |
| Average Researcher Compensation | $8.7 million annually |
| Model Training Runs Completed | 23 large-scale experiments |
| Projected AGI Timeline | 2028-2030 (internal estimates) |
Why This Matters in the AI Era
MSL represents the first corporate division explicitly targeting superintelligence rather than incremental AI improvements. This shift signals that major tech companies are moving beyond product-focused AI toward foundational intelligence research. For businesses, MSL’s trajectory suggests that AGI capabilities may arrive faster than anticipated, requiring accelerated digital transformation strategies. The lab’s talent concentration also demonstrates how AI development is consolidating among well-funded players, potentially creating significant competitive advantages for Meta’s ecosystem partners and strategic challenges for rivals lacking similar research depth.
Meta has launched Meta the intelligence factory race between AI labs — s-chatgpt-co-creator-shengjia-zhao-to-lead-superintelligence-labs/”>Superintelligence Labs (MSL), a new AI division aimed at developing artificial general intelligence and superintelligence. Led by former Scale AI CEO Alexandr Wang and former GitHub CEO Nat Friedman, the lab has recruited at least 20 top AI researchers from competitors including OpenAI, Google DeepMind, and Anthropic, offering unprecedented compensation packages reportedly ranging from $100 million to $450 million over four years.
Leadership Structure
Top Leadership
- Alexandr Wang – Chief AI Officer, overall lead of MSL
- 28-year-old former CEO and co-founder of Scale AI
- Meta invested $14.3 billion for 49% stake in Scale AI as part of his recruitment
- Described by Zuckerberg as “the most impressive founder of his generation”
- Nat Friedman – Co-lead, heading AI products and applied research
- Former GitHub CEO (Microsoft)
- Previously ran one of the leading AI investment firms
- Served on Meta Advisory Group for a year before joining
- Daniel Gross – AI Products
- Former CEO of Safe Superintelligence (co-founded with Ilya Sutskever)
- Business partner of Nat Friedman
- Joined after Meta’s failed attempt to acquire Safe Superintelligence
Complete List of Confirmed Hires
From OpenAI (11 researchers)
- Trapit Bansal – Pioneered RL on chain of thought and co-creator of o-series models
- Shuchao Bi – Co-creator of GPT-4o voice mode and o4-mini; led multimodal post-training
- Huiwen Chang – Co-creator of GPT-4o’s image generation; invented MaskGIT and Muse architectures
- Ji Lin – Helped build o3/o4-mini, GPT-4o, GPT-4.1, GPT-4.5, 4o-imagegen, and Operator reasoning stack
- Hongyu Ren – Co-creator of GPT-4o, 4o-mini, o1-mini, o3-mini, o3 and o4-mini; led post-training group
- Jiahui Yu – Co-creator of o3, o4-mini, GPT-4.1 and GPT-4o; led perception team at OpenAI
- Shengjia Zhao – Co-creator of ChatGPT, GPT-4, all mini models; led synthetic data at OpenAI
- Lucas Beyer – From OpenAI Zurich office (previously Google DeepMind)
- Alexander Kolesnikov – From OpenAI Zurich office (previously Google DeepMind)
- Xiaohua Zhai – From OpenAI Zurich office (previously Google DeepMind)
- Additional unnamed researcher
From Google/DeepMind (3 researchers)
- Jack Rae – Pre-training tech lead for Gemini and reasoning for Gemini 2.5; led Gopher and Chinchilla
- Pei Sun – Post-training, coding, and reasoning for Gemini; created last two generations of Waymo’s perception models
- Additional researcher (not yet publicly named)
From Anthropic (2 researchers)
- Joel Pobar – Worked on inference at Anthropic; previously at Meta for 11 years
- Shengjia Zhao – Also counted in OpenAI alumni (moved from OpenAI to Anthropic to Meta)
From Other Companies
- Johan Schalkwyk – Former Google Fellow, early contributor to Sesame, technical lead for Maya
Organizational Structure
Meta Superintelligence Labs (MSL) encompasses:
- Foundation Models Teams
- Llama model development teams
- Teams working on Llama 4.1 and 4.2
- Infrastructure supporting over 1 billion monthly active users
- FAIR (Fundamental AI Research)
- Meta’s long-standing AI research division
- Now consolidated under MSL umbrella
- AI Products Teams
- New Frontier Lab
- Focused on developing next-generation models
- Small, talent-dense team working on achieving AGI/superintelligence
- Parallel research track to existing Llama development
Technical Focus Areas
Based on the expertise of hired researchers, MSL is focusing on:
- Reasoning and Chain-of-Thought Models
- Multimodal AI
- Voice, image, and video understanding
- Several hires with GPT-4o multimodal experience
- Synthetic Data Generation
- Critical for training next-generation models
- Led by former OpenAI synthetic data lead
- Perception and Computer Vision
- Inference Optimization
- Making models faster and more efficient
- Critical for widespread deployment
Compensation and Recruitment Strategy
- Signing Bonuses: Reportedly up to $100 million
- Total Packages: Some researchers offered up to $450 million over 4 years
- Typical Offer: Around $200 million over 4 years for top talent
- Recruitment Method: Zuckerberg personally involved, meeting candidates at his homes in Lake Tahoe and Palo Alto
- “The List”: Zuckerberg compiled a list of most-cited AI researchers to target
Physical Organization
- Location: Meta’s Menlo Park headquarters
- Seating: Reorganized so new superintelligence team members sit near Zuckerberg
- Size: Approximately 50-person core team, with plans to expand
Strategic Goals
- Personal Superintelligence for Everyone: Deliver AI assistants that exceed human capabilities across Meta’s platforms
- Compute Advantage: Leverage Meta’s massive infrastructure investment (up to $65 billion in 2025)
- Product Integration: Deep integration across Meta’s 3+ billion user ecosystem
- Open Philosophy: Continue open-source approach with Llama while pursuing AGI
Timeline and Milestones
- June 2025: Initial recruitment begins, Zuckerberg starts personal outreach
- June 30, 2025: Official announcement of Meta Superintelligence Labs
- July 1, 2025: Alexandr Wang officially joins as Chief AI Officer
- Coming weeks: Additional hires to be announced
- Next 12-18 months: Target to reach AI frontier with next-generation models
Competitive Context
Meta’s MSL represents a significant escalation in the AI arms race, directly challenging:
- OpenAI’s lead in reasoning models
- Google DeepMind’s research capabilities
- Anthropic’s safety-focused approach
- Growing competition from Chinese labs like DeepSeek
The formation of MSL signals Meta’s transformation from a social media company with AI features to a dedicated AI research powerhouse competing for superintelligence leadership.
How AI Is Changing This
Meta’s Superintelligence Labs is leveraging AI to fundamentally transform how artificial general intelligence (AGI) systems learn and reason. A concrete example of this transformation is their development of advanced multimodal AI models that can simultaneously process and understand text, images, audio, and video in real-time conversations. Unlike traditional AI systems that handle one data type at a time, Meta’s researchers are creating unified architectures where AI can seamlessly transition between discussing a photo’s contents, generating relevant audio responses, and maintaining contextual understanding across all modalities within a single interaction. This approach represents a significant shift from compartmentalized AI tools toward more human-like cognitive processing. The lab’s focus on building these integrated systems is accelerating progress toward superintelligence by enabling AI to develop more sophisticated reasoning capabilities that mirror how humans naturally combine multiple senses and information sources to understand and interact with the world.
For deeper analysis: The Business Engineer — AI Strategy Intelligence
Notable Observations
- Chinese Talent: 7 of 11 publicly named technical hires are graduates of prestigious Chinese universities (Tsinghua, Peking University, etc.)
- OpenAI Dominance: The majority of hires (11) came from OpenAI, suggesting targeted poaching
- Failed Acquisitions: Meta attempted to acquire Safe Superintelligence but was rebuffed by Ilya Sutskever
- Cultural Concerns: OpenAI leadership has expressed that Meta’s approach could create “deep cultural problems”
This comprehensive restructuring and aggressive hiring campaign represents one of the most significant organizational shifts in the AI industry, with implications for the development and deployment of artificial general intelligence in the coming years.