Connected vehicles are becoming data powerhouses, generating up to 25 gigabytes of information every hour through sensors and vehicle-to-everything communication. But as cities deploy AI systems to manage traffic and enable autonomous driving, a critical challenge is emerging: training artificial intelligence models on moving vehicles consumes far more energy than engineers initially expected, particularly when cars switch between cellular network towers .

This energy problem is more urgent than it might sound. Vehicle connectivity rates are projected to reach 95% by 2030, meaning billions of connected cars will soon be feeding data into AI systems designed to predict traffic, issue collision warnings, and promote fuel-efficient driving . Yet the infrastructure to handle this data efficiently doesn't fully exist yet. Researchers at Orange Belfort, a telecommunications research facility, have been studying exactly how much energy these AI systems consume in real-world driving conditions, and their findings reveal a significant efficiency gap that the industry needs to address.

Why Is Connected Vehicle AI So Energy-Intensive?

The core issue stems from how modern AI systems work with sensitive data. Traditional machine learning requires sending all vehicle data to a central server for processing, which raises privacy concerns. Federated learning, a newer approach standardized by the 3rd Generation Partnership Project (3GPP), solves this by keeping personal data on the vehicle itself . Instead, each car trains its own AI model locally, then sends only the model updates to a central server, which combines them into a shared global model that benefits all vehicles.

This sounds elegant in theory, but it creates a practical problem: vehicles must maintain constant communication with servers while performing complex computations. Orange researchers conducted a real-world test using connected vehicles on their network in Belfort, France, measuring energy consumption as cars moved through different network conditions. The results were sobering. The energy cost per data transmission increased by an average of 30% when vehicles performed handovers, meaning when they switched from one cellular antenna to another during movement . For a vehicle traveling through a city, these handovers happen frequently, compounding the energy drain.

What Are the Real-World Implications for Smart Cities?

Cities like Barcelona and Columbus are already deploying connected mobility systems that rely on AI to optimize traffic flow. Barcelona's AI-powered traffic management system has achieved a 20% reduction in congestion by dynamically synchronizing traffic lights and predicting traffic patterns . Columbus is testing autonomous vehicles equipped with advanced sensors that communicate with urban infrastructure in real time. These systems work because they process data efficiently, but as vehicle connectivity scales up, energy consumption could become a bottleneck that limits deployment.

The challenge extends beyond individual vehicles. A broader survey of 510 large organizations across Europe and the United States found that more than one-third are already struggling to meet their climate targets as AI adoption accelerates . While digital technology currently accounts for less than 5% of most companies' total emissions, this share is expected to grow significantly as AI systems become more prevalent. The survey, conducted by management consultancy BearingPoint in February 2026, revealed a critical gap: 95% of organizations have committed to science-based climate targets, yet 37% report delays in achieving them, partly due to rising energy demands from digital infrastructure and AI .

How to Reduce Energy Consumption in Connected Vehicle AI

• Selective Client Participation: Instead of having all vehicles participate in federated learning simultaneously, researchers can prioritize vehicles that are least likely to experience network handovers, reducing unnecessary energy-intensive transmissions and focusing computational resources on stable connections.
• Transmission Queuing Mechanisms: Vehicles can wait for their radio conditions to stabilize before sending model updates to the server, rather than transmitting immediately when requested, which reduces energy waste during poor signal conditions.
• Asynchronous Aggregation Methods: Servers can process model updates from vehicles at different times rather than waiting for all vehicles to submit simultaneously, allowing the system to adapt to varying network conditions and reduce energy pressure on individual devices.

Orange Belfort researchers are actively exploring these optimization strategies. The team's proof-of-concept demonstrated that by combining selective client participation with transmission queuing, energy costs could be substantially reduced without sacrificing model performance . The key insight is that not all vehicles need to participate in every training round, and timing matters enormously when network conditions are fluctuating.

Why Are Technology Leaders Missing the Sustainability Connection?

Despite the growing energy demands of AI, most organizations have failed to integrate technology and sustainability strategies. The BearingPoint study found that only 36% of businesses have fully aligned their technology and sustainability strategies with corporate environmental targets . More striking, 40% of Chief Information Officers (CIOs) and Chief Technology Officers (CTOs) are not involved in sustainability decision-making at all, and only 20% co-develop climate goals at the executive level .

"AI is emerging as both a sustainability enabler and a new source of emissions, requiring stronger governance and measurement," stated Matthias Roeser, global leader of Technology at BearingPoint.

Matthias Roeser, Global Leader Technology at BearingPoint

This disconnect is problematic because technology leaders are uniquely positioned to measure and manage AI's energy footprint. The survey also revealed that 50% of organizations lack the right tools to manage sustainability metrics, and just 33% feel they have sufficient data from technology suppliers to commit credibly to emission reduction targets . Without better measurement and governance, organizations cannot optimize AI systems for energy efficiency, even when they want to.

What Does This Mean for the Future of Connected Mobility?

The findings from Orange's research and the broader BearingPoint survey suggest that connected vehicle AI is at an inflection point. The technology is necessary for smart cities and autonomous driving, but deploying it at scale without addressing energy efficiency could undermine corporate climate commitments and strain power grids. The good news is that solutions exist. Model weight compression, optimal client selection, and transmission queuing are all proven techniques that can reduce energy consumption without sacrificing AI performance .

However, implementing these solutions requires closer collaboration between technology leaders and sustainability teams. Organizations need to embed energy efficiency criteria into AI development from the start, measure digital emissions transparently, and use data to optimize both AI models and the infrastructure that supports them. For connected vehicles specifically, this means designing federated learning systems with network conditions in mind, not as an afterthought.

"AI can become one of the most powerful enablers of sustainability, but only if it is deployed responsibly," concluded Rémy Sergent, global leader of People and Strategy at BearingPoint.

Rémy Sergent, Global Leader People and Strategy at BearingPoint

As vehicle connectivity rates approach 95% by 2030, the decisions made today about how to optimize federated learning will ripple across entire transportation systems. Cities that invest in energy-efficient AI infrastructure now will be better positioned to scale autonomous vehicles and smart mobility services without triggering new emissions crises. The research from Orange Belfort shows that this optimization is technically feasible; the challenge is making it a priority before connected vehicles become ubiquitous.