Data Centers Waste Energy — Not Because of AI, but Because of Electronics
How photonics can structurally reduce the energy footprint of digital infrastructure
The rapid expansion of AI has placed data centers at the center of the global energy debate. Governments, utilities and technology companies warn of grid congestion, rising electricity demand and the limits of “green power”. The underlying assumption is clear: AI is inherently energy-hungry.
That assumption is misleading.
The real issue lies not in intelligence itself, but in the physical foundations of the infrastructure that carries it. Modern data centers consume vast amounts of energy not because they compute too much, but because they lose energy to heat, resistance and cooling. Photonics offers a structural alternative to that inefficiency.
The hidden energy cost of digital infrastructure
A modern data center’s energy use can be divided into three layers:
- Computation — CPUs, GPUs and accelerators
- Data movement — chip-to-chip, server-to-server, rack-to-rack
- Cooling — required to dissipate the heat generated by the first two
In electronic systems, these layers are inseparable. Every bit that moves through copper encounters resistance. Resistance produces heat. Heat requires cooling. Cooling consumes additional energy.
As a result, a significant share of a data center’s electricity is not spent on computation at all, but on mitigating the physical side effects of electronics.
Why AI makes the problem visible
Traditional IT workloads generated relatively limited data traffic. AI workloads do not. Modern AI systems rely on:
- massive parallelism
- continuous parameter exchange
- extremely high bandwidth between processors
This shifts energy consumption away from computation toward communication and cooling. In large AI clusters, data movement and thermal management can consume as much — or more — energy than the calculations themselves.
AI does not create a new problem. It amplifies an existing physical bottleneck.
What photonics changes at a fundamental level
Photonics does not optimize electronics. It replaces the underlying carrier. Instead of electrons moving through copper, information is carried by light through glass.
That distinction has profound consequences.
1. Eliminating resistance at the source
Light propagates through optical media without electrical resistance. This means:
- minimal heat generation during transport
- no voltage drop over distance
- sharply reduced cooling demand
Every watt not converted into heat is a watt that does not need to be removed.
2. Near-zero-marginal energy for data transport
In photonic interconnects, data is:
- optically modulated
- transmitted via fiber or on-chip waveguides
- converted back to electronics only at the edge
For data centers, this makes chip-to-chip and server-to-server communication, historically one of the largest energy sinks, energetically flat. This is especially critical for AI systems, where communication is often the limiting factor.
3. Computation without proportional energy scaling
For specific classes of tasks — such as matrix operations and pattern recognition — photonic systems can exploit the physics of light itself. Interference and diffraction perform the computation.
Crucially, these operations require:
- no clock cycles
- no sequential instructions
- no additional energy per computational step
The energy input is largely fixed, independent of speed or scale.
This is not incremental efficiency. It is a different scaling law.
What photonics does not do
Photonics is not a universal replacement for electronics. It:
- does not eliminate electronic logic entirely
- does not make data storage optical overnight
- does not render data centers energy-neutral
But that is precisely why it is credible. Photonics targets the most energy-intensive layers of the system — communication and specific compute kernels — where electronic inefficiency is structural rather than incidental.
It lowers the energy curve instead of merely compensating for it with renewable supply.
From computing facilities to infrastructure policy
As heat and energy losses are reduced, the role of data centers begins to change:
- higher compute density per site
- lower marginal cost per AI workload
- reduced dependence on large-scale grid expansion
- tighter integration with national energy strategies
Data centers cease to be purely technical assets and become strategic infrastructure.
Control over photonic core technologies increasingly translates into control over energy efficiency, scalability and long-term digital sovereignty.
The real policy question
Much of today’s debate focuses on how to power AI with cleaner electricity. That framing avoids the harder question. As long as the underlying infrastructure remains electronically constrained, efficiency gains remain limited.
The more fundamental question is:
Which physical principles should underpin our digital future?
Conclusion
Electronics consumes energy through resistance and heat.
Photonics avoids those losses by moving — and in some cases processing — information with light.
This does not merely make data centers more efficient. It makes large-scale AI physically and economically manageable within real-world energy limits.
Not by cooling harder.
Not by generating more power.
But by rethinking the foundations.
Illustration credit:
Conceptual pencil sketch of a future photonic data center, visualising low-energy computing infrastructure based on light rather than electronics.
About the forthcoming book
A related electronic book will be released shortly.
The book explores how a shift from electronics to light reshapes not only computing, but intelligence itself — connecting physics, AI, infrastructure and responsibility. It examines why the limits of electrons are becoming visible now, how photonics changes the logic of scale, and what this transition means for Europe’s technological autonomy and the future of intelligent systems.
