Meta Acquires Assured Robot Intelligence to Build the Android of Humanoids

The deal

Meta acquired Assured Robot Intelligence (ARI) on May 1, 2026, for an undisclosed sum. The startup's entire team — including co-founders Lerrel Pinto and Xiaolong Wang — joined Meta Superintelligence Labs the same day the acquisition was announced. ARI had raised a seed round from AI-focused firm AIX Ventures, and its small team was split between San Diego and New York.

ARI was building foundation models designed to give humanoid robots the ability to understand, predict, and adapt to human behavior in unstructured real-world environments. The focus was not hardware — it was the intelligence layer: motion planning, real-time decision-making, whole-body coordination, and learning-based control systems that improve through interaction rather than rigid pre-programmed instruction.

The founders' pedigree

The ARI co-founders are not generic startup talent. Xiaolong Wang was a researcher at Nvidia and an associate professor at UC San Diego, where he won the MLSys 2024 Best Paper Award for work on activation-aware weight quantisation — the same technique that underpins several high-value AI hardware acquisitions. Lerrel Pinto previously taught at NYU before co-founding Fauna Robotics, a startup that built Sprout, an approachable bipedal robot roughly the size of a child. Pinto left Fauna in 2025, and Amazon moved quickly — acquiring Fauna in March 2026 to anchor its own humanoid push.

Pinto co-founded ARI shortly after departing Fauna. The fact that both founders then landed at Meta within weeks of that is worth noting: the window between founding and acquisition was extremely short, suggesting Meta was tracking this team closely and moved fast once Pinto was available again.

The Android play

Meta's stated goal for robotics is not to ship a consumer humanoid — at least not primarily. The company's publicly articulated strategy is to replicate what Android did for smartphones and what Qualcomm's chips did for mobile silicon: own the foundational intelligence layer and let hardware manufacturers build on top of it. Meta Robotics Studio, launched in 2025 and led by former Cruise CEO Marc Whitten, has been recruiting roughly 100 engineers to develop both in-house humanoid hardware and the AI models that power it. ARI slots directly into that stack.

What Meta is really assembling is the software stack for an industry. ARI brings whole-body humanoid control models. Meta Superintelligence Labs provides the research infrastructure. Robotics Studio provides the product and hardware integration surface. The ARI acquisition is less about any single product and more about closing a gap in the capability chain — one that would otherwise take years to build from scratch.

Physical AGI and the broader race

The timing is not incidental. The humanoid robotics market has moved from speculative to genuinely competitive in under two years. Tesla is targeting large-scale Optimus V3 production between July and August 2026, with annual capacity targets of one million units and a price point between $20,000 and $30,000. 1X Technologies opened a factory in Hayward, California, selling out its first year of NEO robot preorders within five days. Apptronik raised $520 million at a $5 billion valuation, partnering with Google DeepMind. Amazon made two robotics acquisitions in a single month. Unitree is targeting 20,000 humanoid shipments this year.

ARI co-founder Pinto articulated the directional bet clearly on X after the acquisition: 'We have the potential to transform AI that can think and talk to AI that can do, assisting humans safely and reliably in the physical world.' Wang added that the team's goal is 'physical AGI' — the idea that true general intelligence will require learning from direct interaction in the physical world, not data alone. Many researchers share that view. Training AI purely on internet-scale text and video has limits. Robots that learn by doing in unstructured human environments represent a different and potentially richer training signal for the next generation of AI models.

The market numbers — and the gap between them

The forecasts for humanoid robotics span an implausibly wide range: Goldman Sachs projects a $38 billion market by 2035; Morgan Stanley puts it at $5 trillion by 2050. That spread — more than 130x — reflects genuine uncertainty about adoption timelines, deployment environments, and which technical bottlenecks will prove hardest to crack. The intelligence layer is one of them. Hardware has improved dramatically; the software to make robots reliably useful in human spaces has not kept pace. That gap is precisely where ARI was working.

Meta's capital expenditure for 2026 has been revised upward by $10 billion, bringing total expected spending to between $125 billion and $145 billion, driven by AI data centre expansion and rising component costs. Against that backdrop, an undisclosed-sum acquisition of a small but technically elite robotics team is a rounding error — but one that signals where the company expects the frontier to move next.

Sources