Mecka AI nears $500 million valuation in Sequoia-led round
Mecka AI is nearing a funding round led by Sequoia Capital at a valuation of about $500 million, people familiar with the proposed deal said. The startup collects and analyses human-motion data for training humanoid robots and other robotics systems. The round’s size has not been reported, and its terms are not final.
The prospective financing follows closely behind Mecka’s announced $60 million round, led by Framework Ventures and joined by Menlo Ventures, SV Angel and Kindred Ventures. That transaction was disclosed three months before the reported Sequoia-led deal. Mecka AI did not comment on the new financing, while Sequoia declined to comment.
Building a supply of physical-world data
Founded in 2024, Mecka was created by Josh Gao, Mogen Cheng, Jason Chong and Duy Nguyen. Gao and Cheng had previously built a restaurant fintech company, while Chong joined Coinbase after the acquisition of his crypto exchange. Nguyen leads operations. The founders did not come from robotics backgrounds.
They identified the limited availability of physical-world data as a constraint on general-purpose robots, including humanoid machines. Mecka’s name derives from “mecha”, the fictional giant robots controlled by humans. Its operating premise is to create a human-data pipeline for robotics comparable to the role companies such as Scale AI, Mercor and Surge play in supplying data for large language models.
Mecka pays participants to record themselves doing daily tasks, including making coffee and repairing cars. The recordings use body sensors and smartphones, producing what is commonly described as egocentric data: material captured from the performer’s perspective and tied to their movements during a task.
Robotics data attracts investor attention
When Mecka announced its prior fundraising, Gao told Fortune that the company expected to finish 2026 with an annual run rate of $100 million. Mecka has not publicly named its customers. Robotics companies and AI laboratories use real-world data captured through egocentric collection alongside other methods, including teleoperation, to develop their models.
The category is drawing competition and investment. XDOF, another company collecting real-world data for robot training, was reported to be nearing a round at a $1.2 billion valuation. Established human-data platforms, including Scale AI and Micro1, are also expanding beyond LLM-focused work into physical-world data collection.
For businesses assessing robotics programmes, the development underlines that training-data operations deserve the same planning as models and hardware: task coverage, capture quality and repeatable collection processes can determine what a system can learn to do.