Serve Robotics Launches Physical AI Services for Robot Fleet Connectivity and Training
By Greg Bock · · 3 min read
Insider Brief
- Serve Robotics has launched Physical AI Services, a platform offering robot connectivity, data management and AI training technology developed for its autonomous delivery fleet to other robotics companies.
- The platform includes Serve Link for connecting robots to remote computing systems, Serve Data Curator for identifying valuable training data and a catalog of real-world data collected from sidewalk delivery operations.
- Serve said its technology already supports thousands of vehicles across roughly a dozen industries, with Bedrock Robotics among its launch customers, using Serve Link to connect autonomous construction equipment through Wi-Fi, cellular and satellite networks.
Serve Robotics has launched a new business offering the connectivity, data management and AI training technology developed for its autonomous delivery robots to other companies operating robotic fleets.
According to the San Francisco-based company, its Physical AI Services platform will help robot developers connect machines to remote computing systems, identify useful training data and improve AI models using information collected during real-world operations.
Serve noted it has deployed more than 2,000 robots across more than 40 cities and 27 hospitals, developing the software and infrastructure needed to keep its machines connected and operating in different environments.
The company is now making those tools available to other robotics developers, including businesses working in construction, mining and autonomous trucking. Bedrock Robotics, which develops autonomous systems for heavy construction equipment, is among the launch customers.
Three Products for Physical AI Development
The new platform includes three services designed to support robot operations and AI training:
- Serve Link: Connects robots in the field to remote AI models, monitoring systems and computing infrastructure, allowing machines to exchange information with systems outside the robot.
- Serve Data Curator: Identifies useful training examples from large volumes of information collected by robotic fleets, reducing the amount of data engineers must review manually.
- Real-world data catalog: Provides training data collected from Serve’s sidewalk delivery operations, including information about terrain and infrastructure, to help AI models understand complex urban environments.
According to Serve, the services already support thousands of vehicles across roughly a dozen industries, including its own autonomous delivery operations.
Bedrock Robotics uses Serve Link to connect its construction machines through Wi-Fi, LTE, 5G and satellite networks. The system provides a communications link between equipment in the field and remote computing and monitoring systems.
The data curation service addresses another challenge for companies operating large robotic fleets. As robots collect information during everyday operations, developers must identify the examples most useful for improving their AI models. Serve said its software helps select those examples, allowing developers to concentrate on relevant information rather than reviewing entire datasets.
“Building a robot is only the beginning. Keeping thousands of them connected, learning and improving requires another layer of technology,” Serve co-founder and CEO Ali Kashani said in the announcement. “We’ve put that technology to work in our own fleet. Now we’re opening it to other companies so they can focus on what their robots do, rather than rebuilding the systems around them.”
On Thursday, Serve also reported third-quarter revenue of $2.8 million, up 313% year over year, and raised its full-year 2026 revenue forecast to $12 million from a previous range of $9 million to $10 million. The company said daily deliveries among new platform partners tripled during the quarter, while recurring revenue accounted for 69% of total sales. Serve ended September with $198 million in cash and marketable securities and lowered its full-year non-GAAP operating expense guidance to $150 million.
