Four acquisitions. Four months. North of $16 billion in total spend. Meta isn't just building AI models anymore - it's buying entire teams and funneling them into a single division with a single mission.
Here's the timeline:
- December 2025: Meta acquired Manus, the autonomous AI agent startup, for a reported $2 to $2.5 billion (the Wall Street Journal pegged it at $2.5B including a $500M retention pool). Manus had hit $100 million in annual recurring revenue within eight months of launching.
- June-August 2025: Meta invested $14.3 billion for a 49% non-voting stake in Scale AI. As part of the deal, Scale AI CEO Alexandr Wang stepped down to become Meta's first-ever Chief AI Officer.
- March 10, 2026: Meta acqui-hired the Moltbook team, creators of a social network where AI agents interact autonomously. Founders Matt Schlicht and Ben Parr joined Meta Superintelligence Labs.
- March 23, 2026: The Dreamer team, an agentic AI platform for non-technical users, joined Meta. Dreamer's co-founders include David Singleton (former Stripe CTO) and Hugo Barra (who previously led Meta's VR unit before leaving in 2021).
Every single one of these teams landed in the same place: Meta Superintelligence Labs, the division Wang now co-leads alongside Nat Friedman, former CEO of GitHub.
The Agent Bet
Look at what Meta actually bought. Manus builds autonomous web agents that handle real-world tasks through a browser. Moltbook studied how AI agents behave when they interact with each other at scale. Dreamer built personalized AI agents for everyday users. Wang spent years at Scale AI building the data infrastructure that makes AI systems work.
This isn't a scattered shopping spree. Meta is assembling every layer of the AI agent stack under one roof: the data pipeline (Scale AI expertise), the autonomous execution layer (Manus), the multi-agent interaction research (Moltbook), and the consumer-facing agent interface (Dreamer).
The Money Behind It
Meta spent $72 billion in capital expenditure in fiscal 2025 and has projected $115 to $135 billion for 2026. The acquisition spending is a fraction of the infrastructure budget, which tells you where the real investment is going: compute.
The talent consolidation serves a specific purpose. Building a superintelligence lab from scratch takes years. Acqui-hiring four established teams in four months compresses that timeline dramatically, especially when the teams bring working products (Manus's agent is already shipping a desktop app as of March 18) rather than just research papers.
What This Means for Everyone Else
Meta is making a clear bet that the next era of AI isn't about chatbots answering questions - it's about agents doing work. The companies building agent platforms just watched their most likely acquirer take four competitors off the board.
For people using AI tools day-to-day, the practical impact is still downstream. But the signal is loud: the company spending more on AI infrastructure than most countries' GDP is all-in on agents that act, not assistants that talk.