Every second, the data behind billions of emails, TikTok videos and AI queries travels around the world as pulses of light through fiber-optic networks. Along the way, these signals pass through tiny components that act as channels for light: photonic chips. These devices don't just carry signals—they direct and combine them, ensuring information moves efficiently across complex networks.
But photonic chips still have limits. They struggle to perform certain key light-processing operations. Tasks such as signal conversion and amplification still rely on additional components—components that are bulky, consume energy and generate heat.
Today, the energy footprint of these components remains relatively small, accounting for only a few percent of a data center's total electricity use. But generative AI is already starting to change that equation.
Unlike a simple search query, generative AI systems depend on constant back-and-forth exchanges between processors. Each exchange increases the number of times signals must be converted and reshaped. What was once a minor cost is becoming a structural challenge—and one that could limit how far AI systems can scale.
Without changes, that trend could drive a rapid—and potentially unsustainable—rise in the energy use of digital infrastructure, which already represents about 2% of global electricity consumption.
A team led by engineering physics professor Stéphane Kéna-Cohen at Polytechnique Montréal believes it may have found a way forward. Their results appear in Science Advances.
To read more, click here.