Maven Robotics Exits Stealth With $100M to Scale Industrial Robots

In 2024, Maven Robotics had no product. Hamza Derbas, the company's chief executive, called the state of the business "a cartoon of a robot and a team of people." A large consumer goods company was touring automation vendors that week. Derbas talked his way into the meeting. Then he asked to walk the factories.

Two years later, the cartoon has hands.

Dealroom ranks the Series A in the 99th percentile of all US rounds on record, a sample of 27,020 deals. Earlier backers had put roughly $18 million into the company across a seed round and accelerator programs.

The deployments matter more than the cheque. Hardware teams raise nine figures on a demo every quarter. A robot that survives paid production shifts is a different claim, and it is the one that will produce a real number.

What eight robots actually prove

Mixed palletizing is dull work. Pallets of boxed goods arrive from several factories. A distribution center rebuilds them into store-specific mixes. The order can change the moment product reaches a shelf.

"Within 48 hours of them putting the stuff on the shelves, they want to change the mix based on real-time demand," Derbas told TechCrunch. "Here's an order with different mixed products going to that retail store; please build it out."

The hardware is deliberately plain. Each robot rides a wheeled base that can move at 10 miles per hour, with two vacuum-gripper arms that lift up to 30 kilograms. No legs. Derbas argues bipedal machines are "very complex, unreliable, and add unnecessary cost" on a warehouse floor. "ROI is the name of the game here."

What the fleet does have is a data loop borrowed from self-driving cars. It pulls information back from operating robots within minutes or hours, retrains models, runs ablation studies, then redeploys. Derbas spent nine years in Apple's special projects group, widely reported to have housed the company's self-driving car effort before it shut down in 2024. His brother Khalid, a co-founder and the chief financial officer, came from private equity. The wider team includes veterans of Tesla and Rivian.

The number that will define the company is smaller than the round. It expects its fleet to pass 100,000 hours of autonomous operation by the end of 2026, and 1 million by the end of 2027. Uptime has run at 99% or better across those 16-hour days.

The operator's case for nine figures

RoboStrategy's Jack Pearson framed the investment around operations rather than research. There is "a huge gap between a robot that demos well and one that survives three production shifts a day, seven days a week," he said. The team, in his account, had shipped products before and built for enterprise requirements from day one.

"We're not in the race for models. We're in the race to solve industrial labor and make this work possible at the scale the world needs."— Hamza Derbas, chief executive and co-founder, Maven Robotics

Among robot makers, the company sits closer to Agility, going public this fall in a $2.4 billion special purpose acquisition company (SPAC) deal, than to the general-purpose humanoid labs. Both chase industrial workflows. It simply refuses to stand up.

From palletizing to material handling

The system is billed as general-purpose. It gets there one task at a time. "We're grounded in solving one customer problem at a time," Derbas said. "If you focus on solving problems and you pick sizeable problems, each problem is a multi-billion-dollar market."

The first task is mixed palletizing, which the company sizes at an $80 billion market. Material handling and assembly, the next layer, is pegged above $1 trillion. To move up, it is collecting more factory data, and it has even built pincer-like gloves that let humans mimic the gripper form factor it wants to train.

As we wrote in September, the industry's proof metric has shifted from the demo to the deployment. Its eight robots are the newest entry in a race now measured in surviving shifts, not launch videos.

The bet against one big model

The counter-case writes itself. A frontier lab could ship a physical AI model that handles mixed palletizing out of the box, turning its task-specific edge into a commodity. Derbas answers that the moat is not architecture. It is the data that comes off real machines, in real facilities, on real shifts. Models need that data. It is collecting that data now, customer by customer, and each task it masters adds another dataset the labs cannot buy.

That argument holds only while the tasks stay narrow enough to sell. Palletizing is narrow. Material handling is wider. Fabrication is wider still, and the manipulation needed there does not exist yet. The plan is to fund the gap with revenue from the simple work.

How fast can industrial robots clear a million operating hours?

Development scenarios