Sequoia leads talks to value Mecka AI at $500M after its $60M raise

Why it matters

The proposed deal prices human-motion data as scarce robotics infrastructure. Mecka AI must prove its collection network can produce durable model gains instead of simply accumulating more video.

Josh Gao, Mogen Cheng, Jason Chong and Duy Nguyen are nearing a financing led by Sequoia Capital that would value Mecka AI at about $500M, according to TechCrunch's September 11th report. The round size remains unknown, and the terms could change before the deal closes. For now, $500M is a proposed price rather than a completed financing.

The talks come three months after Mecka AI announced $60M across two financings led by Framework Ventures. That compressed timeline shows how quickly investors are moving to secure a position in the supply chain for robotics training data, before a clear category winner has emerged.

The four founders arrived at the problem without conventional robotics careers. Gao and Cheng previously built restaurant payments technology and sold that venture in 2023. Jason Chong had worked as a Microsoft software engineer and co-founded Utopia Labs, an onchain payments operation acquired by Coinbase. Nguyen had made money reselling sneakers and later ran Diamond Agency. Their backgrounds gave Mecka AI an operating thesis rather than a research pedigree: general-purpose robots would need a much larger supply of demonstrations showing how people handle ordinary physical work.

Four outsiders pick the bottleneck

The founders spent months reading robotics papers, visiting laboratories and contacting roboticists before settling on human demonstrations as Mecka AI's core input. Teleoperation, where a person directly controls a robot through a task, produces information tied closely to a machine's joints and actions. Collection is constrained by the number of robots, operators and hours available.

Mecka AI asks people to perform the work themselves. Contributors record tasks such as making coffee, moving objects or repairing cars with smartphones and wearable sensors. Mecka AI then processes the recordings into data intended for robotics models.

The approach treats billions of people as potential teachers for robots. It also creates a difficult translation problem. A human demonstration captures hands, objects, movement and context, while a robot may need force readings, joint states and action labels that are specific to its hardware.

Mecka AI has consequently expanded its pitch beyond collection. Mecka AI's platform covers data processing, evaluation and support for commercial deployments, placing the operation between robotics labs, hardware makers and workplaces that want robots performing defined tasks. Gao told Fortune in June that "useful robots can be deployed today," framing integration work as part of the product rather than an eventual add-on.

A valuation moving faster than the disclosure

The previously announced $60M consisted of a $25M Series A that closed in November 2025 and a $35M follow-on investment, according to Fortune's June 1st report. Framework Ventures led both, with Menlo Ventures, SV Angel, Kindred Ventures and angel investor Ted Xiao participating. Mecka AI had about 40 employees at the time.

Gao also told Fortune that Mecka AI expected to reach a $100M annual revenue run rate by the end of 2026, based on signed contracts. The figure is a projection rather than realized annual revenue. Customer identities, pricing and the amount already collected sit outside that metric, leaving the proposed $500M valuation anchored partly to contracts that have yet to appear as a full year of reported sales.

The fundraising pace is still notable. A Sequoia-led deal would give Mecka AI another pool of capital only months after Framework Ventures supplied $60M for data infrastructure, new verticals and robot deployments. It would also bring a large generalist investor onto a cap table that had been led by a fund historically associated with crypto, although Framework Ventures has been expanding its focus into robotics and AI.

Human demonstrations meet robot reality

Mecka AI has begun putting its technical premise into public research. The EgoVerse paper lists Gao and Chong as industry partners and acknowledges Meta Reality Labs Research, Mecka AI and Scale AI for data contributions, engineering support, infrastructure assistance and collaboration. The paper reports 1,362 hours of human demonstrations across about 80,000 episodes, 1,965 tasks and 2,087 demonstrators. The EgoVerse project homepage currently lists a larger snapshot of 4,003 hours, with the same approximate episode, task and demonstrator counts.

The paper offers support and a constraint on Mecka AI's thesis. Researchers found that co-training with human data generally improved robot-policy performance, while effective scaling depended on alignment between the human data and the robot's learning objective. The results also showed benefits from diversity among demonstrators and scenes. Collection quality, dataset composition and transfer methods therefore determine whether additional hours produce useful robot behavior.

That caveat explains why Mecka AI is building collection hardware, annotation tools and deployment services around the recordings. A defensible operation will require control over the full path from a person performing a task to a robot reproducing it reliably in a workplace. A library of videos alone would be easier for customers or competitors to replace.

Investors rush toward the data suppliers

Mecka AI is one of several well-funded attempts to become the data layer beneath general-purpose robotics. XDOF entered talks for a financing led by 8VC at about a $1.2B valuation, TechCrunch reported on September 4th, three months after XDOF disclosed a $70M Series A. Those terms were also unfinished. XDOF told TechCrunch it worked with 20 customers and was approaching $50M in annualized revenue.

The distinction between the two operations is one of emphasis. Mecka AI centers data captured from people. TechCrunch reported in June that XDOF's founders had worked on GELLO, a low-cost teleoperation system, and that the company combined remote robot teleoperation with human collectors wearing sensors to capture egocentric data. Scale AI, Micro1 and other human-data providers are also moving toward physical AI as robotics developers search for datasets that cannot be scraped from the public internet.

For Gao and his co-founders, the reported Sequoia talks validate an outsider bet made after months spent learning a field they had never worked in. The proposed valuation also raises the execution threshold. Mecka AI must convert human activity into data that improves robot performance, then preserve enough control over evaluation and deployment to avoid becoming a labor-intensive collection contractor.

Mecka AI's founders are selling investors a position between the people who know how physical work gets done and the machines being trained to copy them. The rush to price that position at $500M has arrived before the financing terms, customer list and year-end revenue projection can provide a complete picture. The founders now have to make the technical transfer work as quickly as the valuation has moved.