In just three months, Mecka AI has vaulted from a $60 million funding round to the cusp of a deal that would value the startup at roughly $500 million — an extraordinary leap that underscores the insatiable appetite for physical-world data in the race to build general-purpose robots. The new round is being led by Sequoia Capital, according to two people familiar with the negotiations, positioning Mecka as one of the fastest-rising companies in the burgeoning field of human motion data collection for robotics and AI. While the precise size of the round remains undisclosed and terms could still shift, the valuation alone signals that major venture capital firms see human-generated data — not just synthetic or simulation-based training — as the critical accelerant for humanoid robots and other autonomous systems.
Mecka AI’s business is deceptively simple: it pays people to record themselves performing everyday tasks — making coffee, repairing a vehicle, folding laundry — using body-worn sensors and smartphones. That footage, captured from a first-person or “egocentric” perspective, is then processed, labeled, and sold to robotics companies and AI labs hungry for realistic, real-world interaction data. The company’s name, derived from “mecha” — the fictional giant robots piloted by humans in anime and science fiction — reflects its core philosophy: machines learn best by watching humans do. And in a field where data scarcity has become the single most stubborn bottleneck, Mecka is positioning itself as the bridge between human actions and robotic capabilities.
The startup’s trajectory has been startlingly rapid. It was founded only in 2024 by four entrepreneurs who had no prior background in robotics. Three months ago, Mecka announced a $60 million round led by Framework Ventures, with participation from Menlo Ventures, SV Angel, and Kindred Ventures. Now, with Sequoia Capital leading what appears to be a much larger infusion, the company’s valuation has more than octupled. Industry observers are drawing parallels to the explosive growth of Scale AI, Micro1, and other human-data platforms that transformed the training of large language models — except Mecka is tackling a vastly more complex domain: the messy, unpredictable, and sensor-rich physical world.
What Is Mecka AI? A Data Pipeline for the Physical World
Mecka AI is a data collection and annotation company specialized in human motion and interaction data used to train humanoid robots and other autonomous machines. The company pays contractors to wear body sensors and use smartphone cameras to record themselves performing routine tasks in natural environments — from kitchen chores to industrial repairs. Those recordings are then processed into structured datasets that teach robots how to grasp objects, navigate spaces, and interact with tools and people in ways that simulation alone cannot replicate. The core insight behind Mecka is that the most effective training data for general-purpose robots comes from observing actual humans doing actual tasks, not from software-generated synthetic data or controlled lab recordings.
The Data Bottleneck That Could Hold Back Humanoid Robots
For years, the robotics community has wrestled with a fundamental problem: there is no abundant, high-quality dataset of everyday human physical interactions. Language models benefited from the vast text corpus of the internet. Vision models trained on billions of labeled images. But for robots — especially humanoid robots that must operate in human environments — the training data must capture the full physical context: how a hand wraps around a coffee mug, how weight shifts when lifting a box, how an eye tracks movement through a cluttered room. Simulation can generate enormous amounts of synthetic data, but simulation rarely captures the friction, variability, and unpredictability of the real world. Mecka’s founders, Josh Gao, Mogen Cheng, Jason Chong, and Duy Nguyen, recognized this gap and acted on it. None of them came from robotics; their backgrounds spanned restaurant fintech, crypto exchange acquisitions, and operations. But they saw the same bottleneck that has frustrated established robotics companies for years — and they built a company to unlock it.
How Mecka AI Collects and Processes Human Motion Data
Mecka’s data pipeline relies on a combination of wearable body sensors and smartphone cameras. Contractors are equipped with motion-capture devices that track joint angles, acceleration, and spatial position, while their smartphones record video from a first-person or third-person perspective. The recordings are then synced, cleaned, and annotated by human labelers — and increasingly by automated tools — to produce training datasets that include not just the visual stream but also the underlying physical dynamics. This “egocentric” approach is critical: robots need to understand not only what a human action looks like, but also the forces, torques, and sequences that make it successful. The resulting data is used by robotics companies and AI labs to train models that can then generalize to novel tasks, reducing the need for tedious teleoperation or hand-coded routines. As of early June, Mecka projected that it would end 2026 at an annual run rate of $100 million — a figure that suggests strong customer demand even though the company has not publicly named its clients.
Meet the Four Founders Who Bet on Physical Data
Mecka AI was co-founded in 2024 by four individuals with complementary but unexpected skill sets. Josh Gao and Mogen Cheng, both Canadians, previously built a restaurant fintech startup, giving them experience in operational scaling and payment systems. Jason Chong joined Coinbase after it acquired his crypto exchange, bringing deep technical and product expertise from a high-growth environment. Duy Nguyen, the only non-Canadian on the team, focuses on operations — a role that is critical for a company that must manage a distributed workforce of data collectors, process sensor data at scale, and deliver reliable datasets to demanding customers. The team’s lack of direct robotics experience initially raised eyebrows, but the founders have argued that their outsider perspective allowed them to see the data gap more clearly than companies steeped in hardware and simulation. Their rapid success — from a $60 million round to a potential $500 million valuation in three months — suggests the market agrees.
The Competitive Landscape: XDOF, Scale AI, and Micro1
Mecka is not alone in chasing the physical data opportunity. Last week, TechCrunch reported that XDOF, a startup that also collects real-world data for robot training, was nearing a new round at a $1.2 billion valuation — more than double Mecka’s prospective figure. While XDOF focuses on a different technical approach, the two companies are effectively competing for the same customers: robotics firms and AI labs that need training data for manipulation, locomotion, and interaction. Meanwhile, established human-data platforms like Scale AI and Micro1, which built their businesses on labeling text and images for language models, are expanding into physical data collection. Scale AI has been a dominant force in the AI data market, but cracks have appeared in its relationship with Meta and other partners; Micro1 recently raised funds at a $500 million valuation. This convergence suggests that the physical data market is growing fast enough to support multiple players — but that valuations are also inflating rapidly as investors race to back the winners.
Why Sequoia’s Involvement Signals a Strategic Shift
Sequoia Capital’s decision to lead Mecka’s new round is significant. The legendary venture firm has been one of the most influential backers in AI, from early investments in OpenAI to later bets on infrastructure and application layers. By placing its weight behind a company that collects human motion data for robots, Sequoia is signaling that the next frontier of AI — the physical world — requires a new kind of data pipeline. Sequoia’s involvement also lends credence to the thesis that humanoid robots and general-purpose machines are approaching a tipping point where data, not hardware, will determine which companies lead. The firm’s due diligence likely included a deep examination of Mecka’s data quality, customer traction, and scalability — and its decision to lead a round at a $500 million valuation suggests the findings were compelling.
The Economics of Physical Data: Projected $100 Million ARR by 2026
Mecka’s financial projections, shared by co-founder Josh Gao with Fortune in early June, indicate a company that is growing with unusual speed. The forecast of $100 million in annual run rate by the end of 2026 — roughly two and a half years from now — implies that Mecka expects to sign major contracts with leading robotics companies and AI labs, and that its data collection and processing operations will scale accordingly. For comparison, Scale AI took several years to reach a similar run rate, and it benefited from a much larger market for language model data. Mecka’s trajectory suggests that demand for physical-world data is accelerating faster than many analysts anticipated. However, the company’s customer list remains undisclosed, making it difficult to verify the projection. Investors will be watching closely to see whether Mecka can convert its data pipeline into long-term, high-margin revenue — or whether competition and customer turnover will compress margins.
What Are the Risks and Challenges for Mecka AI?
Despite the excitement, Mecka faces several significant risks. First, the quality and consistency of human-sensor data can vary widely depending on the contractor’s environment, compliance, and equipment. Maintaining a large, reliable workforce of data collectors is expensive and operationally complex. Second, the company is entering a competitive landscape where deep-pocketed rivals like XDOF, Scale AI, and Micro1 are also racing to capture the same customers. If robotics companies develop their own internal data collection capabilities — or if synthetic data from simulation improves dramatically — Mecka’s value proposition could erode. Third, the valuation of $500 million is based on a single round led by a single investor, and the terms may include protections that could dilute earlier investors if the company fails to hit growth milestones. Finally, the broader robotics market remains nascent; humanoid robots are still largely experimental, and the number of customers that can pay top dollar for training data is limited. If the hype cycle peaks before the technology matures, Mecka could face a demand freeze.
What Is the Broader Significance for Robotics and AI?
Mecka’s rise is part of a larger shift in how the AI industry thinks about data. For years, the dominant paradigm was “scale is all you need” — feed a large model more text, more images, more compute, and emergent capabilities follow. But physical robots require a different kind of scale: not just more data, but data that captures the full complexity of real-world interaction. Without high-quality human motion data, humanoid robots will remain brittle and limited to controlled environments. Mecka and its competitors are essentially building the data infrastructure for the era of physical AI. If they succeed, they will have created the foundational layer upon which general-purpose robots can be trained — a role analogous to what ImageNet did for computer vision. If they fail, the bottleneck will persist, and the dream of household robots, warehouse automation, and humanoid assistants will remain tantalizingly out of reach.
The deal with Sequoia is not yet closed, and terms could change. But the signal is unmistakable: the human data market, which transformed language AI, is now being applied to the physical world, and investors are betting that the same playbook will unlock machines that can move, manipulate, and work alongside people. For Mecka, the next few months will determine whether it can sustain its momentum, build a durable business, and deliver on the promise of training robots by watching humans. For the rest of the industry, the company’s trajectory offers a glimpse into how the next generation of AI will learn — not from books or simulations, but from the messy, glorious reality of human life.
