AI Needs Faster Connections. Light Could Be the Answer.

26 September 2026 | Lightcone Advisors Research

Imagine building a bigger city without widening its roads. More people arrive, but traffic prevents them from moving efficiently.

AI data centres face a similar problem. Adding powerful chips is only part of the job. Those chips also need to exchange enormous amounts of information. When connections cannot keep up, expensive computing equipment spends time waiting.

That is bringing greater attention to photonics: technology that uses light to carry information.

Why copper is coming under pressure

Copper carries data through electrical signals. It remains useful for short connections, but maintaining reliable signals becomes harder as speeds and distances increase. The electronics needed to compensate can consume additional power.

Optical connections carry information using light through fibre. They can move large amounts of data over longer distances, potentially using less energy per unit of information. The benefit is more data moving efficiently, not simply that light “travels faster” than an electrical signal.

The change is happening closer to the chip

Data centres already use optical fibre extensively. The emerging shift is bringing the conversion from electricity to light closer to the processors and networking chips.

Think of it as bringing a high-capacity road closer to every building, reducing the distance traffic must travel through narrow local streets.

One approach, called co-packaged optics, places optical components alongside the chip they serve. The aim is to shorten the demanding electrical connection before data moves onto fibre.

What changes?	Why it matters
More data can move between chips	Less waiting during demanding AI tasks
Optical links can span greater distances	More flexibility in connecting equipment
Connections can use less energy per unit of data	Less power spent moving information

The actual benefits depend on the design and workload; they are not guaranteed improvements for every data centre.

What Lightmatter and Ayar Labs are developing

Lightmatter’s Passage technology focuses on optical connections for AI infrastructure. In March 2026, the company announced that it had demonstrated a chiplet carrying 1.6 terabits per second per fibre. This is a component-level milestone, not a claim that an entire data centre becomes faster by the same amount.

Ayar Labs combines its TeraPHY optical engine with an external light source called SuperNova. Its approach is designed to connect groups of AI processors using optical links. The company’s September update described ongoing work to expand manufacturing and support commercial deployment.

These are examples of an industry transition, rather than evidence that one supplier or architecture will dominate.

Will copper disappear?

A gradual shift is more likely than a complete replacement.

Copper remains practical for many short connections and continues to carry electrical power. Optical systems also require lasers, packaging, testing and maintenance. Their benefits must justify those costs.

The likely direction is more optics where distance, data volume and energy use make it worthwhile, with copper continuing elsewhere.

As AI infrastructure expands, the challenge is becoming as much about connecting computers as building them. Photonics could help turn more of that expensive computing capacity into useful work.

Disclaimer: General technology and sector commentary for educational purposes. Company references describe public announcements and do not constitute investment recommendations.