Serve Robotics is selling the systems it built to run its own robots as services for other companies developing machines.
Built from running robot fleets
Spun out of Uber in 2021, Serve had 2,000 sidewalk-delivery robots by the end of 2025. In 2026, it expanded into hospital robotics by acquiring Diligent Robotics. CEO Ali Kashani has said those businesses will continue; the new services offer another way to sell technology developed while operating them.
Tools for other robot makers
Serve Link connects robots in the field to computing, monitoring and training systems. Data Curator selects unusual or useful moments from fleet data for engineers to review. Serve’s Data Catalog offers real-world information from its sidewalk deployments, including terrain and infrastructure.
Bedrock Robotics, which develops autonomous systems for heavy construction equipment, is an early customer using Link across Wi-Fi, cellular and satellite networks. Serve is also working with T-Mobile to bring computing closer to robots.
A new source of revenue
The tools could let developers spend less time building network and data systems and more time on a robot’s specific job. Serve could earn recurring software revenue without building each customer’s robot or operating every fleet. On its third-quarter earnings call, Kashani said Link would generally be sold on a recurring per-device or per-seat basis, with software-like margins. He cited interest and a sales pipeline in construction, mining and defense, but gave no pipeline size or timetable for contracts.
In August, Serve cut its 2026 revenue outlook to $9 million–$10 million, citing lower-than-expected Uber Eats delivery volume. Three months later, it raised the outlook to $12 million as demand and revenue from a more diversified mix improved. Q3 revenue was $2.8 million, and the company reported a $59.2 million GAAP net loss. Recurring revenue was 69% of the quarter’s total, but Serve has not said how much came from the new services.
Questions Serve still needs to answer
Serve has not explained who owns customer recordings handled by Data Curator, whether they can be reused, or what retention and sharing rules apply. It has also not announced an open technical standard across robot makers. Using Serve could save customers time but make them dependent on a vendor whose robots compete in some markets. A shared infrastructure layer could help fleets work together, while closed platforms could divide the market; neither outcome is established.
The next test is whether Bedrock and other developers move from early use to sustained, paid deployments, and whether Serve can show reliability across different machines and environments. Sidewalk data may help models understand crowded urban spaces, but it will not automatically teach an excavator to work at a job site. Wider adoption will depend on each tool’s value, clear data terms and confidence that robots can keep working when networks fail.