The hardware race behind artificial intelligence may be entering a new phase. According to South Korea's Maeil Business Newspaper (매일경제), the AI "battleground" is shifting away from graphics processing units, or GPUs, toward optical processing.
For the past several years, GPUs have been the workhorses of the AI boom. Originally built to render video-game graphics, these chips turned out to be exceptionally good at the massive parallel math that training and running AI models require. That is why demand for them has been so intense across the industry.
Optical processing points to a different approach. Instead of moving information purely as electrical signals through silicon, optical, or photonic, systems use light to carry and process data. The appeal, in broad terms, is speed and energy efficiency: light-based components can move information quickly and may sidestep some of the heat and power limits that electronic chips run into as they are pushed harder.
The Maeil Business Newspaper item, surfaced via Google News, frames this as a move in where the competition is heading rather than a finished transition. The available source is a brief report, so specifics on companies, timelines, and performance figures are not detailed here.
The broader takeaway is about direction. GPUs remain central to AI today, but if computation begins migrating toward optical methods, it could reshape which firms hold the advantage, how much energy data centers consume, and what the physical machinery of AI looks like.
Why it matters: whoever controls the underlying hardware controls the pace and cost of AI, so a shift from GPUs to light-based processing could redraw the map of an industry currently defined by chips.