Serve Robotics opens its connectivity and data tools to other robot builders

Serve Robotics is opening the technology behind its robot fleets to other companies building machines that work in the physical world. On November 5, 2026, it launched Physical AI Services, a business combining robot connectivity, training-data tools and real-world datasets. Serve named autonomous construction-equipment developer Bedrock Robotics among its launch customers.

Connecting the machine to intelligence elsewhere

The first product, Serve Link, connects robots in the field to AI models, monitoring systems and training systems outside the machine. That connection could let developers use more powerful models in data centers without putting all the required computing hardware onboard. It is a link to remote intelligence, rather than a new robot or a new AI model.

The difficult part is keeping that connection useful when networks are unreliable. CEO and co-founder Ali Kashani told TechCrunch that robots need outside systems for tasks including collecting data and handling exceptions. Serve’s pitch is that it has already built the connectivity technology needed to support those exchanges, rather than leaving each robotics company to solve the problem itself.

Bedrock provides a concrete example beyond sidewalk delivery: Serve says the construction-autonomy company uses Serve Link across Wi-Fi, LTE, 5G and satellite networks. Serve also said it is working with T-Mobile to bring AI computing closer to robots. That collaboration is a stated effort, not a demonstrated performance result for the newly launched business.

Finding the useful moments in robot data

The second product, Serve Data Curator, addresses what happens after a robot collects information. Serve says a fleet generates more real-world data than an engineering team can review. Its tool identifies rare situations—often called edge cases—that developers can feed back into model training. The aim is to select useful examples from the volume, not simply gather more recordings.

The third offering supplies data rather than a tool for sorting it. Serve’s real-world foundation-model data catalog draws from its own sidewalk deployments, including terrain and infrastructure information. The company positions it as training material for models learning to understand complex human environments, such as city sidewalks. Unlike the curator, which helps companies find valuable examples in robot-collected data, the catalog makes Serve’s own field experience part of the product.

Building a robot is only the beginning. Keeping thousands of them connected, learning and improving requires another layer of technology.

Ali Kashani, Serve Robotics co-founder and CEO, in the launch announcement

A supplier business built from operating robots

Serve’s argument for selling this technology rests on having used it itself. The company says it has deployed more than 2,000 robots across more than 40 cities and 27 hospitals. It also says Physical AI Services already supports thousands of vehicles across a dozen industries, including construction, mining and autonomous trucking, alongside its own sidewalk delivery fleet.

Kashani compared the strategy to Amazon’s AWS playbook in his TechCrunch interview: develop technology for an internal operation, then let other businesses build on it. He said building Serve’s own fleet was necessary to reach this point. The comparison describes the company’s ambition, not evidence that its new supplier business has achieved anything like AWS’s position.

This is not a withdrawal from operating robots. Kashani said Serve’s sidewalk delivery and hospital robot businesses will continue. The new line instead offers technology developed for those operations to builders targeting farms, warehouses and other uncontrolled environments. His proposed division of labor is straightforward: customers concentrate on their robots’ jobs while Serve supplies some of the systems around them.