Meta buys a humanoid brain

PLUS: 1X's huge new NEO factory

Good morning, robotics enthusiasts. Meta just acquired Assured Robot Intelligence, a San Diego startup building foundation models for humanoids — and folded its two elite founders directly into Superintelligence Labs.

The move comes days after Meta committed up to $145B to AI infrastructure this year. What, exactly, it plans to build is still murky — but the humanoid race just got a new heavyweight.

In today’s robotics rundown:

  • Meta snaps up robotics startup ARI

  • 1X opens massive NEO humanoid factory

  • Figure claims it can build a robot every hour

  • Uber to turn its drivers into an AV sensor grid

  • Quick hits on other robotics news

LATEST DEVELOPMENTS

META

Image source: Reve / The Rundown

The Rundown: Meta just acquired humanoid startup Assured Robot Intelligence for undisclosed terms, bringing two elite roboticists into its Superintelligence Labs to build foundation models for whole-body humanoid control.

The details:

  • Meta bought San Diego-based ARI, a 20-person startup that focuses on foundation models enabling humanoids to handle household tasks.

  • The founders: Lerrel Pinto, an NYU professor who co-founded Fauna Robotics (acquired by Amazon), and Xiaolong Wang, a former Nvidia researcher.

  • The deal folds ARI into Meta’s Superintelligence Labs division and comes days after Meta raised its 2026 AI infra capex to $125–145B.

  • A leaked 2025 internal memo revealed Meta is developing consumer humanoid hardware, though the company has not confirmed the plan yet.

Why it matters: Meta’s acquisition positions it to compete with Tesla, Figure AI, and Boston Dynamics in commercializing humanoids — if it wants to. But regardless, many AI researchers believe that achieving AGI requires training models through physical interaction, making embodied AI a strategy beyond large language models.

1X

Image source: 1X

The Rundown: Humanoid startup 1X opened its NEO factory in Hayward, California. The 58K-square-foot facility, billed as the most vertically integrated humanoid factory in the U.S., has commenced full-scale production of NEO.

The details:

  • The plant has the capacity to produce 10K NEO units in its first year, with a target of 100K units annually by 2027.

  • 1X is using in-house manufacturing for key components like motors, batteries, and transmissions to reduce reliance on Chinese suppliers.

  • NEO runs on NVIDIA’s Jetson Thor platform, with behaviors trained in simulation, and is pitched as a mobile assistant for light household tasks.

  • The robots are expected to cost around $20K or be available via subscription, with customer shipments expected to begin in 2026.

Why it matters: If 1X can ship at scale, it transforms humanoids from demo reel to actual product — and the data flywheel could matter more than the hardware. With a second San Carlos plant set to come online next year, 1X is betting that vertical integration outpaces Tesla’s factory muscle and Figure’s deep pockets.

FIGURE

Image source: Figure / YouTube

The Rundown: Humanoid startup Figure said its San Jose, California-based BotQ factory has hit a 24x production jump in under four months — and it’s using every robot it makes to train the next generation of AI.

The details:

  • Figure has ramped BotQ from one Figure 03 per day to one per hour — a 24x throughput jump in under 120 days — with 350-plus units already shipped.

  • Each robot reportedly clears 80 functional tests before leaving the floor, with end-of-line yield topping 80%; the battery line runs at 99.3%.

  • BotQ’s 150 networked workstations have produced 9K actuators across 10 component types, targeting 12K robots per year at full capacity.

  • Every unit that ships is also a data engine: real-world fleet signal feeds Helix, Figure's AI model, enabling OTA updates across the entire fleet.

Why it matters: Hard production numbers for humanoids come with a lot of caveats, of course. Tesla missed its 10K-unit 2025 target badly, and 1X’s new Hayward factory hasn’t shipped to customers yet. Figure’s 350 units are mostly internal too — but it’s the only one publishing yield rates, throughput curves, and a credible production ramp.

UBER

Image source: Images 2.0 / The Rundown

The Rundown: Uber is pitching its human drivers as roaming data collectors, using their everyday trips to feed sensor data into an “AV cloud” that self-driving companies can tap to train and test their autonomous systems, TechCrunch reports.

The details:

  • Uber CTO Praveen Neppalli Naga says the company aims to equip human drivers’ cars with sensor kits, so routine trips feed a shared AV data pool.

  • He says AV development is now bottlenecked by high-quality driving data, and that Uber can fill the gap cheaper than AV startups deploying their own fleets.

  • Uber is already running an “AV cloud,” a library of labeled sensor data that some 25 autonomous-vehicle partners can query.

  • Companies can even use it in “shadow mode” to see how their models would have handled real Uber rides without putting a robotaxi on the road.

Why it matters: This could turn Uber’s global fleet into the gatekeeper for real-world driving data, giving it leverage over any AV player that wants to train or deploy on its network. That shift would let Uber profit from autonomy whether or not it owns robotaxis, while tightening its grip on the data layer that future AV competitors need.

QUICK HITS

Chinese robotics firm Unitree launched a low-cost, modular upper-body-only humanoid starting at about $4,290, targeting researchers and developers.

A Waymo robotaxi drove off with a San Jose passenger’s luggage at the airport, and customer service reportedly said the car couldn’t be turned around.

Uber is partnering with Hertz’s new Oro Mobility unit to handle charging, cleaning, maintenance, and depot staffing for its upcoming robotaxi fleet in the Bay Area.

German auto supplier Schaeffler plans to roll out at least 1K Hexagon/AEON humanoids across its global factories by 2032 after a successful 2025 pilot program.

Dutch startup VNYX raised €1M ($1.1M) to scale its AI‑powered robotics systems that automate fashion resale of returns, overstock, and second‑hand garments.

MIT developed microscopic magnetic hydrogel robots — smaller than a grain of sand — that can perform complex maneuvers when controlled by an external magnet.

A humanoid named Bebop caused a Southwest flight delay after airline staff discovered its lithium battery exceeded size limits and confiscated it before departure.

COMMUNITY

That's it for today!

Before you go we'd love to know what you thought of today's newsletter to help us improve The Rundown experience for you.

Login or Subscribe to participate in polls.

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — The Rundown’s editorial team