Serve Robotics opens its fleet tech to other robot makers with Physical AI Services

Serve Link is the connection between machines in the field and the models, monitoring and training systems that increasingly run in data centres[1]. Unite.AI reports it bonds 4G, 5G, Wi-Fi, satellite and private cellular links and handles task management, remote assistance and data collection[3].

Serve Data Curator turns the petabytes a fleet records into searchable data, so engineers can surface rare edge cases worth training on instead of reviewing every recording[1][3].

The third product is a real-world foundation-model data catalogue built from Serve's sidewalk deployments, covering terrain and infrastructure data meant to teach models how busy human environments work[1].

Bedrock Robotics, which builds autonomy for heavy construction equipment, uses Serve Link to connect its machines over Wi-Fi, LTE, 5G and satellite[1]. Serve says the services already run on thousands of vehicles across about a dozen industries, including mining and autonomous trucking[1]. It is also working with T-Mobile to push AI compute closer to robots[2].

Serve says it has more than 2,000 robots in over 40 cities and 27 hospitals[1]. In its Q3 results the same day, it reported $2.8 million in revenue, up 313% year on year, lifted 2026 guidance to $12 million and noted its first international city launched in Australia in October[4].

Most robotics startups rebuild the same unglamorous plumbing: staying online on patchy networks and finding the few clips worth training on. If Serve can sell that layer, it gets a second revenue stream with higher margins than delivery, and smaller fleets get infrastructure tested at scale[2][4].