The global AI race is often framed as a contest to build the fastest, most advanced processors. On that front, Europe is behind.
But the next phase of AI may hinge on something far more fundamental than chip design: the ability to move colossal volumes of data quickly and efficiently without melting the power grid.
This, according to Sébastien Bigo, a Fellow at Nokia Bell Labs, offers Europe a second chance, and where photonics, the technology that uses light to carry information, becomes a strategic imperative.
AI systems are no longer constrained by compute alone. Their performance increasingly depends on the connectivity fabric linking processors, memory, storage and distributed computing resources.
As AI models grow and workloads spread across larger clusters, advanced connectivity determines how effectively those resources can operate as a single system. Electrical interconnects (the copper traces linking chips) were never designed for anything on the current scale. They impose physical limits on how densely processors can be packed in a data centre, generate heat, and consume huge amounts of energy.
Recent analysis from Axis Intelligence's Data Centre Electricity Demand Tracker shows global data centre electricity demand is projected to exceed 1,000 TWh by the end of 2026, roughly Japan's annual consumption, driven primarily by AI focused facilities.
The industry has already reached the point where adding more GPUs doesn't automatically deliver greater capability.
"There is a lot of focus on AI chips and electrical chips, and the other piece of the puzzle, which is how we interconnect them, is being overlooked," said Bigo.
The ability to connect those processors efficiently increasingly determines the performance, scale, and economics of the overall AI system. If Europe continues to focus solely on catching up in microelectronics manufacturing without prioritising photonics, it risks losing twice: failing to match the scale of US and Asian chip ecosystems while also missing the emerging battleground where the next generation of AI infrastructure will be won.
"We therefore strongly welcome the EU's broader strategy and praise its recognition of chip design as a top strategic priority," added Bigo.
The consequences are not abstract: without more efficient connectivity to move data, expanding AI infrastructure will place additional pressure on electricity grids and increase power and cooling costs. It could also make it harder for Europe to deploy AI broadly across sectors such as industry, healthcare and transport.
Building advanced connectivity capabilities would help Europe strengthen the resilience and competitiveness of its AI infrastructure.
Europe's strategic advantage
Photonics already powers the world's long-distance communications. It is now moving into the data centre, replacing electrical links with optical ones that move data faster and farther, with dramatically lower energy consumption.
That shift is already underway, from the optical links now standard between data-centre racks to co-packaged optics that bring light directly next to the processor, and ultimately to photonic-chip integration, in which compute, memory, and optical input-output sit side by side.
Bigo describes the significance: "There are ideas that are even becoming realities, where you directly connect photonics to the chips themselves. You co-package the computing, potentially the memories, with the optics themselves. At the output, you have fibre."
The more AI infrastructure relies on light rather than electricity, the more scalable and energy-efficient it becomes. "The general trend is that the more photonics you add to your data centre, the more energy-efficient you're going to be," Bigo explained. This is where Europe has a genuine competitive edge.
Why Europe can lead
Bigo is frank about Europe's position in chips: "We would like to lead in AI, but let's be honest, we didn't get there first."
Catching the US and Asia in advanced AI chip manufacturing will be difficult. Europe's photonics capability, however, presents a different opportunity: world-class research institutes and deep expertise in optical communications, strong telecom operators, specialist component manufacturers, and decades of industrial capability in semiconductor-adjacent technologies.
But he is equally clear about the opportunity: "If we had to isolate technologies where we are well positioned to lead, I would say photonics is among them. We are lucky enough to own technology blocks that are essential to AI, and they are building the future."
This is the strategic logic behind the AI Grand Challenge - Europe should look to double down on its own strengths. Photonics is becoming a critical infrastructure layer in determining how efficiently AI can scale.
Europe has another advantage, hiding in plain sight. AI inference is increasingly happening closer to users, where latency matters. Telecom operators already have thousands of distributed sites across the continent, many with compute infrastructure in place. Bigo sees this as a structural opportunity: "If you expand those sites with AI, you have a very good opportunity to grow the infrastructure in a much more distributed fashion. Instead of relying on a few tens of data centres across Europe, you could distribute AI infrastructure across thousands of sites."
This distributed model could also ease pressure on electricity supply by breaking up the concentrated power demand of mega data centres and locating smaller facilities closer to available energy sources. Europe's advanced optical networks could connect these distributed computing resources, allowing AI capabilities to be deployed closer to users while still operating as part of a coordinated infrastructure.
For Bigo, Europe now needs to connect its strengths in photonics, semiconductors, packaging and networks to support development from individual components through to complete AI systems. That will require coordinated investment in research, pilot lines, and manufacturing at a pace that matches AI's rapid development.
The next era of AI will depend not only on advances in computing, but on the connectivity that allows computing resources to operate at scale. Europe already has significant strengths in photonics and networks – technologies that can make AI more powerful, distributed and energy-efficient.
The question is whether it will invest at sufficient scale and move quickly enough to turn those strengths into a globally competitive position in AI infrastructure.