Edge AI Examples: Real City Deployments

Edge AI processes data locally at the device or network edge instead of sending it to a centralized cloud. In deployed city systems, that means real-time decisions in milliseconds: Derq cut crashes 90% at one Sarasota intersection, MiTAC delivered city-scale detection across Taipei, Vicente López's IRIS+ earned global recognition, and LTTS cut utility losses 22% across 11 million smart meters.

What Is Edge AI, Exactly?

Edge AI is the practice of running artificial intelligence inference directly on a device or at the network edge, rather than sending data to a centralized data center for processing. The result is real-time analytics and decision-making with minimal latency, no cloud round-trip required, no connectivity dependency, no waiting.

That distinction matters enormously in city environments. A camera detecting a pedestrian stepping off a curb can't wait 100 to 200 milliseconds for a cloud response. An edge processor embedded in a smart pole, a roadside unit, or a utility meter makes that call locally, in milliseconds.

The best way to understand edge AI is through what it actually does in the field. The four deployments below each carry public outcome data. Together they cover traffic safety, public safety, city operations, and energy utilities: the core systems of any functioning city.

Why Cities Can't Wait for the Cloud?

Cloud-only AI introduces three problems that city operators cannot work around: latency, sovereignty, and resilience. Traffic signals, emergency dispatch, and grid protection all require decisions in tens of milliseconds, timescales that a round-trip to a remote data center structurally cannot meet.

Data sovereignty compounds the problem. City governments in the European Union, Latin America, and across Asia-Pacific operate under legal frameworks that restrict where citizen data can be processed and stored. Sending video feeds or sensor streams to a third-party cloud is often legally prohibited, not just operationally inconvenient.

Resilience is the third constraint. Networks fail during exactly the moments when city systems are under the most stress: severe weather, major incidents, coordinated cyberattacks. An edge inference architecture keeps critical services running when connectivity is degraded. A cloud-dependent one goes dark.

Derq + Sarasota: 90% Crash Reduction

Derq's vehicle-detection platform deployed in Sarasota, Florida, produced a 90% reduction in crashes at one intersection, one of the most cited public-outcome numbers in intelligent transportation. The system runs on Intel edge hardware, processing sensor and camera data at the intersection rather than offloading to a remote server.

Real-time vehicle detection is the mechanism. Derq's platform identifies conflict events: near-misses, wrong-way movements, pedestrian encroachments, as they develop, then signals connected infrastructure in time to intervene. That intervention window disappears entirely if inference runs in the cloud.

The Sarasota result is exactly the kind of evidence cities should demand before committing to a deployment. It's a public outcome, from a named city, on a named platform, with a specific numeric result. See how deployments like this integrate into AI-powered smart city applications to understand the full operational picture.

MiTAC + Taipei: Detection at City Scale

MiTAC's intelligent CCTV monitoring system (MiCMS) in Taipei raised the detection rate to 99% and cut the incident rate by 11.83% across a deployment of 18,000 cameras. Field personnel carry license-plate-recognition industrial PCs powered by Intel® Core™ processors that run video analysis at the individual-unit level, while Intel® Xeon® Scalable servers coordinate the citywide system. That layered architecture puts analysis at the point of need instead of leaving every feed to manual central review.

Cubic's GRIDSMART system offers the scale benchmark: real-time vision analytics at nearly 10,000 intersections on Intel silicon. The Taipei results are not lab figures: they are field measurements across a live urban network.

MiTAC's deployment demonstrates what the Sarasota case also shows. Edge AI at city scale does not require a single monolithic system. It requires a proven hardware foundation, Intel edge silicon in both cases, and software that can coordinate inference across thousands of distributed nodes without a central processing dependency.

Vicente López IRIS+: Operator Workload

Vicente López's Irisity IRIS+ platform in Argentina covers public safety and city operations, not just traffic. The deployment reduced operator workload measurably while earning global smart-city recognition: a combination that signals the platform is delivering real operational value, not just passing a lab benchmark.

Operator workload reduction is a concrete outcome that matters for budget-constrained city governments. When edge inference handles routine monitoring, classification, and alerting automatically, human operators focus on decisions that require judgment rather than pattern matching. That shift changes the economics of city operations.

IRIS+ also shows that edge AI's value extends well beyond mobility. Public safety, environmental monitoring, and service coordination all run on the same underlying Intel edge platform, which means a city investing in edge infrastructure for traffic gets a foundation it can extend to every other domain without starting over.

LTTS AMI: 22% Loss Reduction at Scale

LTTS's Advanced Metering Infrastructure (AMI) network cut utility losses by 22% across 11 million smart meters, the largest-scale edge AI deployment in this group and the clearest proof that edge inference extends from mobility and safety into energy and utilities. AMI is the network of connected meters and sensors that gives utilities real-time visibility into consumption, theft, and grid anomalies.

Processing meter data at the edge rather than in a central data center enables anomaly detection and loss attribution at the speed the grid operates. A 22% reduction in losses across 11 million endpoints represents significant recovered revenue and a direct reduction in the carbon cost of over-generation.

The LTTS case addresses scale directly. Eleven million meters is a city-scale deployment in the most demanding sense: distributed, heterogeneous, operating continuously across a live grid. The Intel edge platform underneath it handles that load without a centralized processing bottleneck.

A Live Edge AI Stack in Action

At ITS World Congress 2025, Intel and OnLogic ran a live edge AI demonstration that showed what a deployable stack looks like in practice. Running live traffic feeds in real time, the solution identified incidents, analyzed intersections, and optimized routes, all powered by Intel Core processors, Intel® Arc™ graphics, and SceneScape software running on OnLogic ruggedized edge systems.

The significance of a live demonstration on real traffic feeds is that it removes the lab-conditions objection. Ruggedized OnLogic hardware is designed for the physical constraints of roadside deployment: temperature variation, vibration, limited power budgets, and intermittent connectivity. Intel Core and Arc silicon provides the compute headroom for multi-model inference without requiring data center power.

SceneScape, Intel's spatial intelligence software, coordinates scene understanding across multiple camera feeds, turning raw video into structured situational awareness that downstream systems can act on. That full stack, silicon plus ruggedized system plus spatial software, is what cities need to evaluate, not individual components in isolation.

What Problems Does Edge AI Solve?

Edge AI directly addresses four problem classes that cloud-only architectures cannot resolve: latency, data sovereignty, resilience, and operational cost. The case studies above map onto all four. Derq and MiTAC address latency through real-time detection. IRIS+ addresses sovereignty via local processing. LTTS addresses operational cost through grid-edge anomaly detection at scale.

The pattern extends beyond these four cases. Intel edge platforms run in more than 100,000 production deployments, and ISS operates safety systems in over 50 countries on Intel architecture. Those are deployed systems, not projections, and they are consistent with what the four city deployments here are producing.

The resilience problem deserves specific emphasis. Cities that have moved critical inference to the edge continue operating during network outages, cyberattacks, and natural disasters, exactly the conditions under which centralized systems fail. That operational continuity is not a feature that can be added later. It has to be in the architecture from the start.

Outcomes Beat Promises Every Time

Four city deployments, four public numeric outcomes, one common hardware foundation. Derq's 90% crash reduction at one Sarasota intersection, MiTAC's city-scale detection across Taipei, IRIS+'s operator workload reduction in Vicente López, and LTTS's 22% utility loss reduction across 11 million meters all run on Intel edge platforms. That's not coincidence: it's what an open, proven, city-scale edge infrastructure produces when the deployment is designed for real operational conditions, not controlled benchmarks.

The gap between a proof of concept and a deployed system at city-wide scale is closed by infrastructure that has already proven itself under load, in the field, with public data to show for it. TOPS counts and cloud integration promises don't cross that gap. Verified outcomes do. Cities that insist on public outcome data before committing are the ones that actually move from pilots to city-wide scale. It is also what the shift from smart city to cognitive city looks like in production.

Cognitive Cities connects the silicon, systems, and software that make deployments like Sarasota, Taipei, Vicente López, and the LTTS grid possible: open enough to preserve choice, local enough to protect trust, and proven enough to scale with confidence across every system that makes a city work.

FAQ

What is an example of edge AI in a real city?

Derq's vehicle-detection platform in Sarasota, Florida is one of the most documented examples. It runs AI inference locally on Intel edge hardware at intersections, processes sensor and camera data in real time, and produced a 90% reduction in crashes at one instrumented intersection. That is a public outcome from a live city deployment, not a lab benchmark.

How does edge AI differ from cloud AI?

Edge AI processes data locally on a device or at the network edge, eliminating the round-trip to a remote data center. Cloud AI sends data off-site for processing, introducing latency measured in hundreds of milliseconds. For city systems like traffic signals, grid protection, and emergency response, that latency gap is operationally unacceptable.

What outcomes has edge AI produced in traffic management?

Derq's Sarasota deployment cut crashes by 90% at one intersection. MiTAC's Taipei intelligent CCTV system sustained city-scale detection, with a 99% detection rate and an 11.83% incident-rate reduction across a live urban network. Cubic's GRIDSMART runs real-time vision analytics at nearly 10,000 intersections on Intel silicon.

Can edge AI work in utilities as well as transportation?

LTTS's AMI network demonstrates that edge AI scales directly into utilities. Running on Intel edge platforms across 11 million smart meters, the system cut utility losses by 22% by enabling real-time anomaly detection and loss attribution at the grid edge, without routing all meter data through a central server.

Why does data sovereignty require local edge inference?

Many jurisdictions legally restrict where citizen data can be processed and stored. Sending video feeds, sensor streams, or health data to a third-party cloud may violate EU GDPR, Argentine data protection law, or equivalent frameworks. Local edge inference keeps data within the jurisdiction, satisfying sovereignty requirements without sacrificing real-time performance.

What hardware runs deployed edge AI city systems?

All four city deployments cited, Derq in Sarasota, MiTAC in Taipei, IRIS+ in Vicente López, and LTTS across 11 million meters, run on Intel edge silicon. The ITS World Congress 2025 demonstration used Intel Core processors and Intel Arc graphics on OnLogic ruggedized edge systems running Intel's SceneScape software.

What is the ITS World Congress edge AI demonstration?

At ITS World Congress 2025, Intel and OnLogic ran a live edge AI stack on real traffic feeds, identifying incidents, analyzing intersections, and optimizing routes in real time. It used Intel Core processors, Intel Arc graphics, and SceneScape spatial intelligence software on OnLogic ruggedized hardware designed for roadside deployment conditions.

How does edge AI improve emergency response times?

Local inference enables immediate anomaly detection and alert dispatch without cloud latency, and systems stay operational during network outages, sustaining response capability at exactly the moments when centralized systems are most likely to fail. At the Kumbh Mela, LTTS and Intel edge systems helped manage 30 million pilgrims in a single day while resolving more than 80,000 citizen grievances.

What is the IRIS+ platform in Vicente López?

IRIS+ is a public-safety and city-operations platform deployed in Vicente López, Argentina. It runs edge AI inference locally to automate monitoring, classification, and alerting, reducing operator workload and freeing human staff for higher-judgment decisions. The deployment earned global smart-city recognition for delivering measurable operational outcomes under real governance constraints.

What should cities demand before committing to an edge AI deployment?

Cities should require public outcome data from named deployments in comparable environments, not vendor benchmarks or controlled lab results. Specific numeric outcomes like crash reduction percentages, detection rates, and loss reduction figures from live city deployments are the only reliable signal that a platform will perform at city-wide operational scale.

Notices and Disclaimers:

Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex

Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available updates. See backup for configuration details. No product or component can be absolutely secure. Your costs and results may vary.