Serve Robotics Opens Physical AI Services to Outside Fleets
Serve Robotics is opening the infrastructure behind its robot fleets to other companies. The Nasdaq-listed robotics operator announced Physical AI Services on November 5, 2026, bringing fleet connectivity, training-data curation and real-world data into a commercial offering for teams building machines that work outside the lab.
Bedrock Robotics, which develops autonomous systems for heavy construction equipment, is among the launch customers. According to Serve's launch announcement, Bedrock uses Serve Link to connect machines across Wi-Fi, LTE, 5G and satellite networks. A company associated with sidewalk delivery is now selling a supporting layer to a business operating much heavier machinery.
For operators following tech news, the development puts a practical purchasing decision underneath the physical AI narrative. Robotics companies need connectivity and useful training examples as their fleets grow, and Serve is offering tools developed through its own operations. The commercial question is how much engineering work those customers can source from another fleet operator while continuing to build their own autonomy products.
Serve turns fleet operations into a product offering
Serve's company history begins inside Postmates' robotics division in 2017, followed by its emergence as an independent company in 2021. That distinction matters because the services business draws on operating experience accumulated before and after the spinout. Co-founder and CEO Ali Kashani is taking tools built around Serve's machines and offering them to teams with different equipment and operating environments.
Serve reports more than 2,000 deployed robots across more than 40 cities. The company also says its Physical AI Services support thousands of vehicles across a dozen industries, including construction, mining and autonomous trucking. Those figures describe Serve's reported operating reach; they leave the number of outside paying customers and the revenue attached to this particular offering undisclosed.
There is a commercial opening in work that rarely gets top billing during a robot demonstration. Machines need a connection to the systems that monitor and improve them, while engineers need a way to find useful examples in the data coming back. Packaging those functions gives Serve another product to sell and gives a robotics buyer an alternative to maintaining every supporting tool internally. The decision can reach into staffing, integration plans and the ownership of a company's surrounding software stack.
Three products address different parts of the learning loop
The Physical AI Services offering groups three related functions. Serve Link connects machines in the field to off-board intelligence, including models, monitoring and training systems. Serve Data Curator helps teams identify valuable observations within the much larger volume of data a fleet produces. The real-world catalog makes data from Serve's sidewalk operations, including terrain and infrastructure examples, available for training.
Each component sits at a different point in the relationship between deployment and development. Connectivity provides a route between working equipment and the systems around it. Curation helps decide which recorded experiences deserve engineering attention. The catalog supplies examples from a particular operating environment that may be useful to teams developing models for the physical world.
The product descriptions establish what Serve is offering, rather than a numerical improvement every customer can expect. The launch materials do not publish pricing, comparative latency benchmarks, an uptime commitment or measured reductions in model-training costs. Buyers evaluating these tools will need to connect the advertised functions to their own equipment, network conditions and development workflow. A fleet that generates valuable data also needs a clear account of which data it can use, under what terms, and for which training tasks.
Bedrock brings the announcement onto the construction site
Bedrock Robotics develops autonomy for heavy equipment and describes its current work around excavators. Its inclusion gives Serve's launch a specific external customer and a use case beyond the delivery routes associated with Serve's own brand. Serve identifies Link as the product Bedrock uses, so the announcement should be read at that level of specificity rather than as evidence that Bedrock has adopted the entire suite.
The list of network technologies is useful because it makes the commercial relationship more concrete. Serve says Bedrock connects machines using Wi-Fi, LTE, 5G and satellite networks. That establishes the range of connections named in the announcement; the company has not supplied customer-specific service terms or a performance breakdown across them. Bedrock's own website confirms its heavy-equipment autonomy business, while Serve is the source for the Link relationship.
For an engineering team, buying connectivity from another robotics operator introduces a supplier into an important part of the system. The practical evaluation includes integration effort, support responsibilities and the performance evidence needed for the intended use. For Serve, a customer operating construction equipment also creates a different support conversation from running its own delivery fleet. The usefulness of experience from one environment has to be demonstrated through the work of serving another.
The infrastructure business has its own operating demands
DevCuration's earlier coverage of Bessemer's Physical AI 50 examined a market spanning deployed machines, models and supporting infrastructure. Serve's launch offers a concrete example of how an operator can enter that supporting layer. The same company can run physical assets and sell technology to other teams building autonomous systems.
That combination creates a potential route to customers outside Serve's own fleet expansion. It also creates obligations around documentation, integrations, product support and the commercial terms of sharing data. These are consequences of the proposed business model, rather than results established by the announcement. Service-specific revenue, customer retention and the cost of supporting external deployments will make the economics easier to assess as evidence becomes available.
Serve's official product introduction grounds the offering in tools developed through its own operations. The next useful evidence will come from how those tools perform for outside buyers with their own machines and operating constraints. Kashani has opened the door to that business with a construction-equipment customer already named; the continuing work sits in the connections, integrations and support that make that relationship useful in the field.
Frequently Asked Questions
Which parts of a robotics operation does Physical AI Services support?
Serve Link connects field machines to off-board systems, Serve Data Curator helps teams select valuable training examples, and the data catalog supplies examples from Serve’s sidewalk operations. The offering addresses infrastructure around fleet operation and model development.
What has Bedrock Robotics adopted from Serve?
Serve names Bedrock Robotics as a launch customer using Serve Link across Wi-Fi, LTE, 5G and satellite networks. The announcement does not establish that Bedrock has adopted every product in the suite.
What should robotics buyers evaluate before adopting the tools?
Buyers should examine integration work, connectivity performance, support responsibilities and training-data terms for their own equipment and operating conditions. The launch materials do not disclose pricing or customer-specific performance guarantees.
How does the launch extend Serve Robotics’ business?
Serve is making tools developed for its own robot fleets available to external robotics companies. This creates a potential infrastructure business alongside its fleet operations, while service-specific revenue and contract values remain undisclosed.
What evidence will clarify the commercial opportunity?
Additional identifiable customers, integration results, service revenue and the cost of supporting external fleets would make the opportunity easier to assess. Serve’s deployment figures are company-reported operating metrics rather than disclosed financial results for this service.
Where the Money Moved
The intelligence briefing of the innovation economy. Funding, M&A, debt and fund closes, read as market signal rather than deal announcements.
Subscribe to Where the Money Moved

