Mecka
Last reviewed
Jun 3, 2026
Sources
11 citations
Review status
Source-backed
Revision
v1 · 1,591 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 3, 2026
Sources
11 citations
Review status
Source-backed
Revision
v1 · 1,591 words
Add missing citations, update stale details, or suggest a clearer explanation.
Mecka (Mecka AI) is a robotics data company that builds large-scale human-motion datasets used to train robots. The startup records how people move, walk, and manipulate everyday objects, then packages that first-person footage and sensor data into training corpora for robot learning models. [1][2] Mecka positions itself as the data layer for "physical AI," a bet that the humanoid robot boom will need far more human demonstration data than today's labs can collect on their own. [3][4] In June 2026 the company disclosed that it had raised about $60 million in combined Series A financing led by Framework Ventures, on top of an earlier seed round, bringing total funding since its founding to roughly $68 million. [1][3]
Mecka's product is training data for robots, not robots themselves. The core problem it targets is the scarcity of real-world demonstration data for embodied AI. Language models can train on text scraped from the internet, but there is no comparable archive of humans physically doing things in the messy, three-dimensional world. Robots need that kind of data to learn manipulation tasks, and it largely does not exist at scale. [2][4]
To fill the gap, Mecka captures egocentric, or first-person, recordings of people performing ordinary activities: making lattes, cooking, doing mechanical and metal-fabrication work, handling objects in chemistry labs, and completing chores at home. [1][5] The data is gathered through a distributed network of human contributors who wear custom body sensors and use standard smartphones, mostly iPhones running Mecka's capture software. [1][2] The result is multimodal data that records the fine-grained kinematics of hand gestures, walking gaits, and full-body movement from the worker's own point of view. [2][6]
Mecka aggregates this material into a proprietary repository it calls Egoverse, described as a large-scale first-person interaction dataset that maps how humans move through spaces and manipulate objects across many tasks and environments. [6][7] An internal "video understanding lab" processes the raw footage into training-ready demonstrations with the annotations that robot-policy models need. [5] At its public launch the company said it had assembled more than 30,000 hours of such video, which it claimed was the largest collection of its kind. [8] Mecka offers both custom datasets built for individual corporate customers and a gig-economy style app that pays ordinary people to record themselves completing daily tasks, turning data collection into a crowdsourced supply chain. [8]
Mecka's pitch rests on a specific view of where the humanoid and embodied AI field is bottlenecked. The argument is that better motors, cameras, and foundation models are not enough on their own; the limiting factor is high-quality demonstration data that can teach robot policies how real-world tasks actually unfold. [3][4] Vision-language-action models and world models, the architectures most robotics labs are now training, are hungry for exactly the kind of grounded, physical examples that are hard to harvest from the open internet. [2][4]
Human teleoperation, where a person directly puppets a robot to generate demonstrations, is the conventional way to produce this data, but it is slow and expensive because it needs the robot hardware in the loop for every recording. Mecka's wager is that egocentric human data, captured with cheap sensors and phones and decoupled from any particular robot, can be collected far more cheaply and then transferred onto robot embodiments. [1][2] CEO Josh Gao has framed the company's role as supplying the "last mile" of capability for deployment, acknowledging that real-world robotics is messy and that not every task makes sense to automate yet. [4]
Mecka has openly stated its ambition to become "the Scale AI of robotics data," comparing itself to Scale AI, which built much of the human-labeled data infrastructure behind the large language model wave. [4][7] The broader context is a surge of capital into humanoid robotics through 2025 and 2026, with firms such as Figure AI, Tesla, Agility Robotics, and a cluster of Chinese makers all racing to field general-purpose robots. That race has created demand for an independent data layer any robot company can buy from, the market Mecka is trying to define and own. [3][4]
Mecka's commercial dataset is distinct from, but connected to, an open academic project of a similar name. In April 2026 a group of researchers released EgoVerse, described in a paper (arXiv:2604.07607) as a collaborative framework for building and sharing egocentric human demonstration data designed to train robot manipulation policies. [9][10] The release reported 1,362 hours of human demonstrations, roughly 80,000 episodes across 1,965 tasks, 240 scenes, and 2,087 unique demonstrators from around the world, with annotations such as 21-keypoint hand tracking and 6-degree-of-freedom head pose. [9][10]
The EgoVerse effort brought together academic labs and companies, including Georgia Tech, Stanford, UC San Diego, ETH Zurich, MIT, Meta Reality Labs, Scale AI, and Mecka. [9][10] Gao has publicly said that Mecka is "excited to power EgoVerse," framing it as a growing ecosystem for robot learning from human data that has been validated across several research labs, with tools released so anyone can collect, train on, and contribute data. [11] The academic project gives Mecka a research footprint, while its private Egoverse repository remains the proprietary product it sells to customers.
Mecka's founders are not the usual frontier-AI pedigree of Stanford PhDs and ex-lab researchers. The four cofounders previously worked in fintech, crypto, and consumer resale before pivoting to robotics. [1][2] Josh Gao, the CEO, and Mogen Cheng, the chief product officer, had earlier built and sold a startup focused on payments technology for restaurants, which they exited in 2023. [1] Jason Chong, the chief technology officer, sold a prior company to the crypto exchange Coinbase. [1] Duy Nguyen, the chief operating officer, was a longtime sneaker reseller who made millions buying and flipping limited shoe releases before joining the founding team. [1] The early team also included founding engineers Kevin Hu and Anthony Tat and team member Eric Lok. [8]
The group decided to skip another fintech venture and instead aim at where they thought computing was heading, settling on data infrastructure for robots. [1] The company has a cross-border footprint: it is registered and headquartered in New York City, but three of its four founders are Canadian and a large share of its staff sits in Toronto, which led some Canadian coverage to describe Mecka as a Toronto company. [3][4]
Mecka raised money in stages, much of it quietly, before disclosing the full picture in mid-2026. The company first emerged from stealth in August 2025 with an $8 million seed round led by Neo, the startup accelerator and fund run by investor Ali Partovi, with participation from SV Angel, A-Star Capital, Cade, Timeless, and Offline VC. [8] It then closed a $25 million Series A in November 2025 and added a $35 million follow-on in early 2026, neither of which was announced at the time. [1][2]
On June 1, 2026, first reported by Fortune, Mecka revealed the two later rounds together as roughly $60 million in Series A financing. [1] Counting the earlier seed, total capital raised since the company's formation came to about $68 million. [3] The financing was led by Framework Ventures, with backing from Menlo Ventures, SV Angel, Kindred Ventures, and angel investor Ted Xiao, a former Google DeepMind researcher and a founding member of Project Prometheus. [1][2]
| Funding detail | Reported figure |
|---|---|
| Seed round | $8 million, August 2025, led by Neo [8] |
| Series A | $25 million, closed November 2025 [1] |
| Follow-on | $35 million, closed early 2026 [1] |
| Combined Series A disclosed | ~$60 million, June 1, 2026 [1] |
| Total since founding | ~$68 million [3] |
| Lead investor | Framework Ventures [1] |
| Other backers | Menlo Ventures, SV Angel, Kindred Ventures, Ted Xiao [1][2] |
| Employees | ~40 [1][3] |
Alongside the raise, Mecka said it had reached an annual recurring revenue run rate of about $100 million based on already-signed enterprise contracts, though it did not name most of its customers. [1][2] The most prominent disclosed customer is the humanoid robot company 1X Technologies, which Mecka says used household-motion data from its platform to help generate a world model earlier in 2026. [4][7] Mecka said it planned to use the new capital primarily to expand its technical workforce beyond the roughly 40 people it employed at the time of the announcement. [1]