$100M/year Scale AI for folding laundry
Jan-Erik Asplund
TL;DR: Mecka pays people to record themselves folding laundry, making coffee & doing other physical work, then sells those demonstrations to companies teaching robots to perform the same tasks. Sacra estimates Mecka hit $100M in annualized revenue in June 2026, with the company expanding from supplying training data to installing & operating robots as data providers, gig-work platforms & world model developers compete to teach machines about the physical world. For more, check out our full report and dataset on Mecka.
We've covered the data labeling market starting with our January 2023 report on Scale AI at $290M in revenue, then following up with coverage of Invisible Technologies ($134M) and an interview with Prolific COO Jemma White on human-in-the-loop data labeling.
In robotics, we’ve interviewed founders including Foundation CEO Sankaet Pathak (March 2025) on why humanoids win, Cytronic CEO Kevin Gibbon (August 2026) on physical AI for ecommerce, and Anvil Robotics CEO Mike Xia (July 2026) on humanoid vs. non-humanoid robots. Most recently, we looked at Runway’s expansion from video generation into world models that simulate physical environments for robot training.
Now we’re looking at Mecka AI, which supplied household-motion data for 1X’s world model & raised a $60M Series B in October led by Sequoia, with NVIDIA, Microsoft’s M12, Qualcomm & Samsung participating.
Key points via Sacra AI:
- Where LLMs could train on ~1,000TB of multimodal internet data, robotics is heavily bottlenecked on the quality & availability of real-world video & teleoperation data, creating the opening for Mecka (2024) to pay mechanics, cooks, electricians and baristas to wear special headgear & gloves and record their daily work with an iPhone app, capturing video, 3D position, force, and pressure data. Similar to how Scale AI paid Filipino contractors to draw bounding boxes around cars & pedestrians in LiDAR scans, Mecka recruits individual workers to its consumer app (Atlas Capture) via gig marketplaces, paying them per hour to record household chores on a head-mounted iPhone, while partnering with bakeries, repair shops & warehouses in the Philippines through its B2B arm, selling the datasets to robotics labs like 1X Technologies ($126M raised, OpenAI Startup Fund).
- While robotics labs themselves already collect data by hiring gig workers to do tasks on camera (Figure) or control their robots via VR rigs (Tesla Optimus), labs buy novel or edge-case data from Mecka, then fine-tune their Vision-Language-Action (VLA) models on it, supplement with teleoperation, and run thousands of physical tests before deploying. Already supplying training data to frontier AI labs, Mercor, Scale AI, Turing & micro1 are expanding into robotics, bringing existing customer relationships, global worker networks & software for assigning tasks, reviewing work & delivering datasets into competition with specialists like Mecka.
- With $47B raised globally for companies working on physical AI in H1'26, Sacra estimates Mecka hit $100M in annualized revenue in June 2026, valued at $500M as of its $60M October Series B (Sequoia) for a ~4x revenue multiple. Compare to direct robotic data competitor XDOF "nearing" $50M in annualized revenue as of September 2026, in talks to raise at a $1.2B valuation for a 24x multiple, and to LLM-centric data labelers moving into robotics like micro1 at $1B in annualized revenue in October 2026 (+1,686% YoY in 2025) valued at $500M in September 2025 for an ~8x multiple & Mercor at $2B in annualized revenue in June 2026 (+2,096% YoY in 2025), valued at $10B in October 2025 for a ~22x multiple.
For more, check out this other research from our platform:
- Mecka AI (dataset)
- XDOF
- micro1 (dataset)
- Mercor (dataset)
- Scale AI (dataset)
- Turing (dataset)
- Snorkel AI (dataset)
- Handshake (dataset)
- Scale: the $290M/year Mechanical Turk of machine learning
- Invisible at $134M in revenue
- Jemma White, COO of Prolific, on why humans ensure AI safety
- $1.1B/year Indeed for data labelers
- LinkedIn for data labelers
- Sankaet Pathak, CEO of Foundation, on why humanoids win in robotics
- Kevin Gibbon, CEO of Cytronic, on physical AI for ecommerce
- Mike Xia, CEO of Anvil Robotics, on humanoid vs. non-humanoid robots
