Mecka Builds the Data and Deployment Layer for Physical AI

Mecka builds the data, evaluation, and deployment infrastructure that helps robots learn from human activity and operate in commercial environments. Co-founder and CEO Josh Gao and co-founder and CTO Jason Chong lead a company working between Toronto and New York, where the assignment is less cinematic than a humanoid demo and more consequential: turn physical experience into something machines can use.

That matters because physical AI has a data problem hiding inside a hardware story. Language models could learn from decades of digitized text. Robots have no comparable archive of how a hand grips a cup, how pressure changes during a fold, or how a worker recovers when the environment refuses to cooperate. Mecka is building that missing layer, then carrying it through evaluation and deployment so enterprise customers do not need to assemble a robotics organization from spare parts.

The company now sits at an unusually sharp intersection of research, infrastructure, and field operations. Mecka's recent $60M Series B gives the team more capital to scale that stack, but the more useful company story is what the stack reveals about the next robotics market: the valuable layer may belong to whoever can repeatedly convert messy human behavior into reliable machine performance.

About Mecka

Mecka was co-founded by Josh Gao, Jason Chong, Mogen Cheng, and Duy Nguyen. The founding group did not arrive from a conventional robotics pedigree. Gao and Cheng previously built a restaurant-fintech company, while Chong built a crypto exchange that Coinbase acquired. That history helps explain Mecka's operating posture. The company approaches robot learning as an infrastructure and execution problem, not only a model problem.

Mecka designs multi-sensor capture hardware, runs real-world data collection, builds the computer-vision and multimodal systems that reconstruct motion, and provides tools for evaluation. Its commercial deployment work then connects hardware, models, data capture, post-training, integration, and ongoing operations at the customer site.

The distinction matters. Mecka does not claim to build the robot. It aims to be the integrator between robotics hardware, frontier models, and the workplaces where those systems must perform. A polished lab demo can survive one ideal environment. A commercial deployment has to survive the Tuesday afternoon nobody rehearsed.

The Data Engine Behind Physical AI

Mecka's thesis begins with a simple constraint: internet-scale data does not exist for physical interaction. Motion, force, contact, geometry, and recovery behavior must be recorded in the world. Raw video alone is not enough. Robotics teams need synchronized sensor data, annotation, quality controls, task context, model evaluation, and evidence that learning transfers across environments.

That is why Mecka owns more of the collection stack. The company says it designs and manufactures instruments around the signals models need downstream, then operates capture programs in homes and commercial settings. Its internal research lab turns noisy recordings into structured signals through motion tracking, 3D reconstruction, and sensor alignment.

EgoVerse shows the research side of the strategy. The collaborative dataset and benchmark brings together Georgia Tech, Stanford, UC San Diego, ETH Zurich, Meta, Scale AI, and Mecka contributors to study how egocentric human demonstrations transfer into robot learning. The project matters beyond one dataset. It gives researchers a shared environment for testing whether more human experience actually improves robot performance and where diversity, task alignment, and embodiment still break the promise.

From Dataset Access to Deployment

Mecka offers several entry points into the same system. Robotics labs and frontier-model teams can work with existing datasets, commission task-specific collections, use evaluation workflows, or build larger data programs. Enterprises can use Mecka to assess physical workflows and support deployments without first hiring an entire robotics research and operations team.

The deployment loop is the strategic edge. A conventional integrator installs a system and treats delivery as the finish line. Mecka's model captures data at the site, uses that evidence for post-training and evaluation, and keeps improving performance as the system encounters real work. In this market, integration is becoming a learning loop.

The company says it supplies several leading robotics labs and multiple Magnificent Seven companies, though it does not publicly identify those customers. Mecka also reports that it surpassed a $100M run-rate revenue level in June 2026 and projects a $300M run rate by year-end. Those are company-reported figures, not independently audited results, but they indicate that demand for the infrastructure around robot learning is arriving well before general-purpose robots become ordinary workplace equipment.

Leadership, Capital, and Operating Momentum

Mecka's capital base now reflects the breadth of the physical AI stack. Sequoia Capital led the October 2026 Series B, with new backing from NVIDIA, M12, Qualcomm Ventures, and Samsung, plus returning support from Kindred Ventures, Framework Ventures, and Neo. The investor mix spans venture, compute, semiconductors, enterprise software, and electronics. That is less a trophy shelf than a map of the dependencies required to move robotics into production.

The team has also expanded through acquisition. Mecka acquired Docula in early 2026 and brought its three-person engineering team into the company to help process large volumes of motion data. The deal was small in headcount and revealing in function: data collection becomes defensible only when the company can normalize, inspect, and learn from what it captures.

Why Mecka's Hiring Signal Matters

Mecka is hiring across research, hardware, operations, and deployment. Its public roles emphasize technical judgment, truth-seeking, urgency, measurable outcomes, and end-to-end ownership. A current forward-deployment role asks for someone who can calibrate robots, monitor sensors and telemetry, lead data capture, troubleshoot networks, and communicate with customers in the field.

That job description is the company thesis in work boots. Physical AI will not cross from research into the economy through model accuracy alone. It needs people who can keep hardware running, preserve data quality, diagnose edge cases, and feed the result back into the learning system. Mecka's hiring momentum is therefore a market signal: robotics infrastructure is becoming an operating discipline, and the boundary between research engineer and field operator is getting thinner.

What Mecka Signals for the Robotics Market

The physical AI race is often narrated through machines. Mecka is betting that the deeper bottleneck sits underneath them. If many robot builders need real-world demonstrations, evaluation, and deployment support, a shared data and integration layer can compound across hardware platforms instead of being locked inside one machine.

That does not make the market easy. Human activity is difficult to record, privacy and quality controls matter, models still struggle to generalize, and customer environments punish brittle systems quickly. But those constraints strengthen the case for infrastructure built around the full loop rather than a dataset sold once.

Mecka's opportunity is to make robot learning less like a collection of heroic experiments and more like a repeatable industrial process. The company is still early, and many of its biggest performance claims remain company-reported. The signal is nevertheless clear: physical AI is moving from model spectacle to deployment economics, and the businesses that measure reality best may determine which robots can finally work inside it.