Artificial intelligence is reaching a tipping point, not in terms of model capability or adoption, but energy. According to a new report released by Photonics21, AI’s surging appetite for electricity could overwhelm power grids, drive up emissions, and derail Europe’s clean energy transition—unless computing moves from electrons to photons.
Compiled by market intelligence agency TEMATYS, the report finds that photonic technologies, using light instead of electricity to move and process data, offer the only credible pathway to scale up AI computing without exacerbating the climate crisis. As data centre energy use rises and conventional silicon architectures reach their physical limits, the shift to photonics is now presented as a strategic imperative.
AI’s exponential energy burden
AI models are not just growing, they are multiplying in deployment. As training workloads intensify and inference becomes ubiquitous across devices, global electricity consumption from data centres has ballooned. The International Energy Agency estimates that in 2024 alone, data centres consumed 415 terawatt hours (TWh). By 2030, TEMATYS projects this figure could more than double, with AI responsible for the lion’s share.
That trajectory puts the entire digital infrastructure on a collision course with Europe’s climate goals. “Photonics can provide the infrastructure that will determine whether AI becomes cleaner and more competitive or simply costlier and dirtier,” Sébastien Bigo, Nokia Bell Labs Fellow and Photonics21 Work Group Leader for Digital Infrastructure, said. “Europe has the research base to lead; what it lacks is coordinated investment and industrial scale.”
The report highlights that while CPUs and GPUs remain essential, they were never designed to handle the data intensity of modern AI systems. By integrating photonics into data centres, through technologies such as co-packaged optics or on-chip photonic interconnects, AI workloads can be processed and moved more efficiently, relieving pressure on conventional chips and reducing power consumption.
Photonics moves from backbone to forefront
Light-based technologies already underpin the internet. From fibre optics in telecoms to lasers in industrial systems, photonics is embedded in today’s digital infrastructure. But the report argues that the next stage, embedding photonic functions directly into computing architectures, is both technically feasible and urgently needed.
Recent breakthroughs have made that vision increasingly real. Researchers at MIT demonstrated a working photonic chip in 2024 that can perform neural network computations using light alone. Several European start-ups are now building prototypes that could become the basis for future fully photonic processors, capable of handling AI workloads with a fraction of the energy required today.
However, these advances are not scaling fast enough. TEMATYS warns that even with current industry claims of 3.5x efficiency improvements from photonic components, these gains cannot keep pace with AI’s energy trajectory unless deployment accelerates dramatically. Without significant action, Europe risks becoming dependent on overseas suppliers for AI infrastructure while failing to meet its climate targets.
A test of sovereignty and strategy
The report is blunt in its assessment: Europe has the scientific expertise, the early-stage innovation, and a growing portfolio of companies in the photonics space. What it does not yet have is large-scale manufacturing, strategic investment, or policy alignment. This gap, if left unaddressed, could see a repeat of past industrial losses in semiconductors and solar technology.
To avoid that outcome, Photonics21 calls on European institutions and national governments to treat photonics as strategic infrastructure for the AI age. Among the recommendations are targeted funding for pilot manufacturing lines, integration of photonics into Chips Act and Green Deal programmes, and new skills initiatives to build a photonics-ready workforce.
There is a growing consensus that AI is not just a software or algorithmic revolution, it is a physical one. Its demands ripple outwards, reshaping energy, infrastructure, and supply chains. Whether Europe can lead in this next phase may depend not on who has the best models, but who can build the smartest, cleanest machines to run them.
