European Cities Are Testing AI Agents to Manage Roads, Ports and Public Buildings

From a French department's 500 public buildings to a Belgian oil port, local governments across EMEA are running pilots that let software watch, reason and act on city data, without sending that data to the cloud.

AI2Day NewsdeskAI-assistedPublished Editor: Lee Brown4 min read
Illustration: a modern European city at dusk, streets lit with warm amber streetlights
Illustration made with AI. Not a photograph of the events described.
Share

Key points

  • Côte-d'Or department in France cut document-search time from weeks to minutes and reduced operating costs by 10 percent, targeting 30 percent at scale, across more than 500 public buildings.
  • Ten software firms, including Honeywell, Milestone and Galene AI, are building city AI tools on Nvidia's Cosmos and Nemotron open models.
  • Infrastructure partners Cisco, Dell Technologies, HPE and Supermicro are packaging Nvidia chips into locally run AI factories so cities keep control of their own data.
  • Projects are live or in pilot across Belgium, Finland, France, Italy, the Netherlands, Poland, Sweden and the UK.

For most city workers, a building compliance report or a traffic-safety alert still means hours in a filing system. A cluster of pilots across Europe is trying to change that, and the results are starting to look concrete.

Nvidia announced the programme on its blog, describing how ten software partners are using two of its open AI model families to build what the industry calls AI agents: software that carries out multi-step tasks on its own rather than just answering a single question. Cosmos is built to interpret video footage, spotting traffic jams or safety hazards. Nemotron is designed to read documents and data, then respond in plain language.

What are these agents actually doing in cities?

France offers the sharpest example. The Department of Côte-d'Or, working with Akila and ThinkDeep, built agents that pull together sensor readings and infrastructure data across more than 500 public buildings. The system runs inside what the partners call a sovereign digital twin (a live virtual copy of the department's buildings held on local servers rather than an outside cloud), and staff can now search documents in minutes instead of weeks.

In Italy, the Milan-area municipality of Cinisello Balsamo is running an AI factory, a locally operated server cluster built around Nvidia chips, that offers real-time translation, chatbots for citizen queries and procurement assistance. Four neighbouring municipalities are planned next.

At the Port of Antwerp-Bruges, Milestone and Dell Technologies are testing vision AI (software that watches camera feeds) to give early warnings of oil spills. Because real spills are rare, the team trains the system on synthetic data, computer-generated images of what a spill would look like, so the model's ready before an incident happens.

In Sweden, a digital twin of Gothenburg is being built as part of the EU's Citiverse project. It'll let residents preview a venue before they arrive, checking seating and crowd levels, with a focus on accessibility.

The UK entries are still at the exploration stage. K2K is working with Liverpool City Region, the Island of Jersey and Innovation City Belfast on separate pilots covering mobility and event management.

Should residents worry about surveillance?

Data sovereignty is the stated priority of every project here, and it's worth taking seriously as a design choice, not just a marketing claim. Every AI factory described runs on local hardware, meaning city data doesn't leave the municipality's control. That's a direct response to longstanding European concerns about public data flowing to American or Chinese cloud servers. We covered Nvidia's Nemotron model family when it first appeared on 16 July 2026, and our September story on Nemotron 3 Diarization showed how the same open-model approach is already in use for crowd-level audio. Whether these city pilots hold to their local-data commitment once they scale is a fair question to keep asking.

The pilots will be presented publicly at Smart City Expo World Congress in Barcelona, running 3 to 5 November 2026.

The honest read on this announcement: the sovereign-AI framing is genuinely meaningful in a European regulatory context, but most of what's described is still pilot-stage. Watch whether Cinisello Balsamo's four neighbouring municipalities actually come online, and whether Côte-d'Or's 30 percent cost-reduction target survives contact with full-scale deployment.

Common questions

Does this mean cities are watching residents on camera?

Some pilots, including the Port of Antwerp-Bruges and K2K's UK projects, do use camera footage. The stated purpose is safety and operations monitoring, and the data is processed on local government servers rather than shared externally.

What is a digital twin, in plain terms?

A digital twin is a live computer model of a real place, updated continuously with data from sensors and cameras. City planners use it to test changes before making them in the real world.

Who pays for all this?

The announcements don't specify public funding amounts. Several projects mention EU initiatives such as Citiverse, which suggests at least partial EU grant funding, but contract values aren't disclosed.

© 2026 AI2Day