Meta's superintelligence unit
"Image synthesis assisted by Seedream 5.0 Pro, an AI partner within the Global Future Nexus ecosystem."
From a $145 billion annual investment and the wholesale acquisition of a rival's leadership to a fundamental restructuring of its entire AI stack, Meta has launched an all-out assault on the frontier of artificial intelligence—a bid to leapfrog the competition and deliver what it calls "personal superintelligence" to every user.
The New Contender
The AI landscape of 2026 is defined by a surprising new frontrunner. While much of the industry's attention has been fixed on the rivalry between OpenAI, Google, and Anthropic, a fourth contender has been quietly building a war chest and assembling a team of unprecedented scale. Meta, the parent company of Facebook, has re-emerged as a formidable force in the AI race through the creation of its Meta Superintelligence Labs (MSL) . After a year of aggressive restructuring and record-breaking capital expenditure, the unit is now projected to overtake Google in frontier AI within the next six months.
MSL represents a complete overhaul of Meta's AI strategy. The tech giant has rebuilt its AI infrastructure "from the ground up", consolidating its core teams—including the Llama team and FAIR's foundational research—into a single, centralised powerhouse. This is not an incremental improvement; it is a fundamental restructuring designed to pursue a singular, ambitious goal: bringing superintelligence to everyone.
A Trillion-Dollar Bet
To achieve this, Meta has deployed a scale of capital that is reshaping the industry. The company has raised its 2026 capital expenditure guidance to a staggering $125 billion to $145 billion, a sum nearly double its 2025 spending of $72.2 billion. This financial firepower, fuelled by a robust advertising business, is being directed towards a massive build-out of physical and computational infrastructure.
The scale of this ambition is almost incomprehensible. Meta plans to deploy seven gigawatts of computing power in 2026, doubling that capacity to 14 gigawatts in 2027. This will be achieved through the simultaneous construction of multiple "Titan" giga-scale data centre clusters, positioning Meta to surpass OpenAI and Anthropic in total AI compute by the end of the year.
To support this hardware expansion and reduce costs, Meta is also developing its own AI chip, codenamed "Iris", with mass production scheduled to begin in September 2026.
The Talent and Data Advantage
Capital alone does not build intelligence. Recognising this, Meta has engaged in one of the most aggressive talent acquisition campaigns in tech history. The company reportedly offered sign-on bonuses as high as $100 million to attract top AI engineers.
The centrepiece of this strategy was a $14.3 billion investment in Scale AI, which led to the recruitment of its founder, Alexandr Wang, as Meta's Chief AI Officer. Wang now leads MSL, which includes a core team of 44 people, 50% of whom were recruited from China. This superteam, capable of converting massive computational resources into frontier-level AI, was further bolstered by the return of former executive Hugo Barra.
Complementing this talent is a unique data strategy. As public data becomes increasingly scarce, Meta has created a proprietary data pipeline by tracking its own internal workflows, redeploying 3,000 engineers to build a large-scale reinforcement learning environment factory. This internal supply chain provides MSL with a sophisticated, and entirely unique, source of training data for next-generation AI agents.
The First Step: Muse Spark
In April 2026, MSL unveiled its first creation: Muse Spark. This natively multimodal reasoning model is a tangible demonstration of the unit's new direction. Unlike simple chatbots, Muse Spark is built for agentic tasks, with support for tool use, visual chain of thought, and multi-agent orchestration. This represents a shift from models that generate content to models that can perform complex, multi-step actions.
A key feature is its "Contemplating mode," which orchestrates multiple agents to reason in parallel, allowing it to compete with the extreme reasoning capabilities of rival frontier models. The model scored 58% on Humanity's Last Exam and 38% on FrontierScience Research. A more advanced version, Muse Spark 1.1, was released in July 2026 as a paid API, directly competing with similar offerings from OpenAI and Anthropic.
The Vision: Personal Superintelligence
Meta's strategy is distinct from competitors who view superintelligence as a centralised, monolithic entity that could automate all valuable work. Instead, Zuckerberg's vision is to "bring personal superintelligence to everyone", giving users a powerful, personal AI tool that deeply understands their world and helps them achieve their goals. This means making superintelligence a utility that amplifies individual human agency rather than one that replaces it.
The GFN Context
For Global Future Nexus, Meta's Superintelligence Labs embody the core tensions of the AGI era. The unit's unprecedented scale and speed of development highlight the "velocity gap" between technological progress and institutional adaptation. The centralisation of data, talent, and compute power raises fundamental questions about the distribution of power and the importance of borderless, accountable governance. As the lines between human and machine agency blur, the work of GFN—in building frameworks for coexistence, trust, and sustainability—becomes not just relevant, but essential.
Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)
Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)
