Serve Robotics launches Physical AI Services to power other robot makers

From delivery to infrastructure

Serve Robotics is expanding its business model beyond its own fleet of self-driving sidewalk delivery robots. The company, which went public in 2024, announced a new division called Physical AI Services. This move is intended to turn the company into a technology supplier for other firms building autonomous machines.

The company originally spun out of Uber’s acquisition of Postmates in 2021. While it continues to operate its own delivery robots, it now wants to sell the underlying technology it developed to other industries. This includes robots designed for farms, warehouses, and other uncontrolled environments.

CEO and co-founder Ali Kashani compared this shift to the 'AWS moment' for robotics. He explained that the company is taking the technology it built for its own fleet and opening it up for other people to build on. This transition follows the company's growth in the sidewalk delivery space with partners like DoorDash and Wonder.

The three pillars of Physical AI

The new Physical AI Services line consists of three specific products designed to solve common hurdles in robotics. One of these is Serve Link, which manages the connection between robots in the field and AI systems located in data centers. This allows companies to use powerful AI models without needing to put all that heavy computing power directly on the robot.

Kashani noted that most robotic applications need to connect to some form of offline intelligence to collect data or handle exceptions. The challenge is maintaining a high-performance connection over unreliable networks. Serve Robotics claims to solve this specific connectivity problem for other developers.

The second product is a data curator tool. This tool helps companies sift through the massive amounts of real-world data that robots collect to find specific 'edge cases.' These are rare or unusual situations that are critical for training and improving the performance of AI models.

A catalog of human environments

The third product offered by Serve Robotics is a real-world, foundation-model data catalog. This catalog was built using data from Serve’s own sidewalk robots. It is designed to teach AI models how to understand and navigate complex human environments.

Kashani stated that he learned this playbook from Amazon. He believes that when technology is still in its early stages, a company must build the end-to-end application first to understand what needs to be created. Once the technology matures, the goal is to open that technology so others can build on the platform.

By providing a catalog of data from real-world sidewalk interactions, Serve is offering a shortcut for other companies. Instead of starting from scratch, new developers can use Serve's existing data to help their robots learn how to behave in the messy, unpredictable world of human spaces.

Partnerships and customers

Serve Robotics is already beginning to see interest in these new services from other players in the robotics space. Bedrock Robotics, a startup that develops self-driving systems for construction equipment, has already been identified as a customer for these tools.

The company is also working with T-Mobile to bring AI computing power closer to robots. This partnership aims to improve the speed and reliability of the systems by reducing the distance between the robot and the processing power it requires. This infrastructure support is a key component of their new service offering.

While these new services are a major focus, the company's existing robot businesses are not disappearing. Serve continues to operate its googly-eyed sidewalk delivery robots and its hospital-based robots, which it acquired through the purchase of Diligent Robotics. The company is simply diversifying how it generates revenue from its core technology.

The path toward a platform

The move toward a platform model suggests that Serve Robotics wants to own the 'pipes' of the robotics industry. By providing the connectivity, data curation, and training catalogs, they position themselves as a necessary middleman for any company wanting to deploy robots in public or complex spaces.

This strategy allows Serve to profit from the success of its competitors as well as its own fleet. If a farm robot company or a warehouse automation firm needs to navigate complex environments, they may have to pay Serve for the data and connectivity tools required to make their machines functional.

As the robotics industry matures, the winners will likely be those who can provide the most reliable infrastructure. Serve Robotics is betting that the technology it perfected for sidewalk delivery is the blueprint for the future of all physical AI applications.