Computing With Light: Nanolasers, Polaritons and the Race to Cut AI's Energy Bill

A modern processor spends a surprising amount of its energy not computing, but moving data from one place to another. As AI models grow, that cost is becoming a serious problem for data centres. Two 2026 results point to the same solution: use light instead of electricity to carry information inside computers^[1][2].

Why moving data costs energy

Inside a chip, data moves as electrical signals along tiny copper wires. Every wire has resistance, so some of the signal's energy is transferred to the wire as heat (P = I²R). Faster signals make it worse, and all that heat must then be removed by fans and cooling systems, which use even more energy. In the words of DTU's team, using electricity for data transmission “limits speed and generates heat”^[1].

Light already runs the internet

You use optical data transmission every day: fibre-optic broadband and the links between data centres carry information as pulses of light, using total internal reflection to keep the light inside the glass fibre. Light can carry enormous amounts of data with very little loss. The challenge is shrinking the light source and switches down to chip scale.

Result 1: a nanolaser that beats the size limit

Ordinary laser cavities cannot easily be made much smaller than the wavelength of the light they produce. In February 2026 Jesper Mørk's group at the Technical University of Denmark (DTU) reported in Science Advances a nanolaser whose nanocavity concentrates photons and electrons in an extremely small region; Mørk says it “breaks the traditional limit on how small lasers can be”^[1]. It works at room temperature with very little power, and thousands could, in principle, fit on a single chip^[1]. Mørk estimates nanolasers could halve computers' energy consumption^[1].

Result 2: half light, half matter

Light is great for carrying data, but photons barely interact with each other, which makes it hard to use light to switch other light, the basis of any logic. In May 2026 Bo Zhen's group at the University of Pennsylvania reported in Physical Review Letters a way around this using exciton-polaritons: hybrid particles formed when photons couple strongly with electrons in an atomically thin semiconductor^[2]. They combine light's speed with matter's ability to interact, allowing optical signals to be switched without repeatedly converting to electrical signals and back^[2]. The switching energy was about 4 quadrillionths of a joule (4 × 10⁻¹⁵ J)^[2].

Data transmission, processors and energy efficiency appear across GCSE and A-Level computer science. For the networking side, read Wi-Fi vs Ethernet; for why efficiency matters in software too, see Big O notation explained.