As artificial intelligence continues to expand rapidly, it is transforming everything from scientific discovery to industrial automation
However, this transformation comes with its own challenges, including the rising rates of energy demands of the data powering modern AI systems.
As Europe moves toward establishing large-scale AI factories with more than 100,000 advanced chips, researchers are racing to develop new technologies to keep pace with both performance demands and sustainability goals.
Horizon Europe’s Digital programme, managed by the European Health and Digital Executive Agency (HaDEA), is supporting a series of cutting-edge photonics projects to address these challenges.
By advancing optical communication components and innovative network architectures, these projects aim to significantly cut energy consumption while enabling the massive scalability required by future data-intensive computing.
Rising energy pressure in the AI era
AI training and inference rely on thousands of interconnected GPUs spread across hundreds of clusters. While processing power has grown exponentially, today’s data-centre networks still rely heavily on traditional electrical switches. These systems are not optimised for the extreme bandwidth, speed, and density now demanded by large-scale AI workloads.
As Europe pursues AI gigafactories and expands its computational capacity, energy use is projected to increase sharply unless more efficient communication technologies are adopted. Optical technologies, particularly those built on photonic integrated circuits, are emerging as a critical solution.
ADOPTION: Advancing co-packaged optics for hyperscale environments
One of the most promising hardware innovations comes from ADOPTION, a project developing next-generation co-packaged photonic components.
This approach brings optical interfaces directly after switching silicon, minimising the distance that electrical signals must travel. The result brings faster data movement, lower power consumption, and reduced latency across server clusters.
For AI training, digital twins, and other data-intensive applications, these efficiencies could translate into high cost and energy savings. ADOPTION also aims to reinforce Europe’s semiconductor ecosystem by covering every stage from chip fabrication to system integration and deployment in cloud data centres.
DYNAMOS: Dynamic, reconfigurable optical networks
Scaling data-centre communication requires not only new hardware but also more innovative network architectures. The DYNAMOS project is addressing this need by developing a modular, dynamically reconfigurable network built around energy-efficient photonic components. Its Dynamic In-line Photonic System cards allow the network to adapt quickly to shifting data-flow demands.
By cutting power consumption by roughly an order of magnitude and dramatically reducing data exchange times between compute clusters, the DYNAMOS architecture is designed to accelerate distributed machine learning and high-performance computing.
OCTAPUS: Photonics at the network edge
The OCTAPUS project is pushing optical innovation beyond the data-centre core and into the edge of the network. Its low-cost, energy-efficient PIC framework enables decentralised service components that can be reconfigured as needed. This flexibility supports emerging 5G, IoT, and Industry 4.0 applications, while enhancing data-centre interconnects with high-capacity, software-defined optical functions.
PUNCH: Next-generation optical switching
Reliable, low-latency communication is crucial for industrial 5G and modern data centres. The PUNCH project is developing a new optical switching paradigm designed to reduce congestion, minimise data loss, and lower interface power consumption. By combining advanced photonic components with real packaging processes, PUNCH aims to deliver industrial-grade prototypes that will be tested in real data-centre and 5G environments.
These photonics research initiatives align with Europe’s plan to improve semiconductor manufacturing, boost cloud and edge capabilities, and support low-latency, high-bandwidth communication infrastructure.
