Largan Precision, the Taiwanese company best known for its smartphone camera lenses, has partnered with TSMC to work on co-packaged optics (CPO) technology. The two firms are pursuing AI opportunities in that area, a move that takes Largan well beyond its traditional phone market.
What CPO brings to the AI data center
CPO bundles optical components with the compute chip in the same package, shortening the distance data has to travel and cutting the power it takes to move that data. In an AI data center, where enormous volumes of information shuttle between processors and memory, these savings matter. Largan and TSMC are betting that co-packaged optics will become a key piece of that infrastructure.
The technology isn't new in principle, but it's only recently become practical for large-scale use. The partnership aims to push CPO from the lab into real deployments. Largan brings its optical know-how, which it has spent decades building for cameras. TSMC brings its chipmaking and advanced packaging. That combination could make CPO a mainstream tool for AI systems.
Largan's move beyond the lens
Largan's business has been tied to smartphones for years. The company's lens modules are in many of the world's flagship devices. But the smartphone market has become crowded, and Largan has been looking for other revenue streams. The CPO partnership with TSMC is a deliberate step toward the AI infrastructure sector, a market that's expanding quickly.
The deal also gives Largan a way to leverage its precision optics experience in a completely different field. Instead of shrinking lenses for phones, it's now developing components for data center switches. That's a strategic shift for a company that's been known for a single, mature product line.
What's left to settle
So far, the companies haven't disclosed a timeline for when CPO products from this collaboration will reach data centers. The partnership is in its early stage, and the technology will take time to integrate into existing systems. The next move is to see how quickly the two can translate their combined know-how into a workable solution for AI operators.
