Graphene Light Chips Move From Lab Benches Toward Real Data Centres

Graphene silicon photonics converts electrical signals into light inside the chip itself, moving data without the heat of copper. The market grows from $289 million in 2026 to $837 million in 2032, with North America at 44% and optical interconnects at 50%.

There is a physical problem that has been heating up data centres for years: copper. Processors, memory and storage have to exchange enormous amounts of information, and as speeds rise, copper links lose signal, need extra correction and give off heat. The solution gaining ground is to put light where electricity used to be. And for that, a material one atom thick is quietly making its way into chip factories.

Graphene silicon photonics combines a one-atom-thick sheet of carbon with silicon wafers that guide light. Graphene moves signals fast; silicon provides the manufacturable foundation of modern computing. Together they allow a chip to convert electrical signals into light, carry it a short distance and convert it back. That avoids much of the heat and congestion that slow down copper wiring.

## A $218 million market aiming for 837

The industry's numbers, compiled in a report by **Markntel Advisor**, put the graphene-based silicon photonics market at **$218 million** in 2025. The projection is **$289 million in 2026** and **$837 million in 2032**, with a compound annual growth rate of **19.39%** between 2026 and 2032. These are the figures of a small segment within the enormous semiconductor business, but the pace of expansion says a lot about where chip fabs are looking.

**North America** accounted for **44%** of the market in 2026. It is no coincidence: that is where hyperscale data centres, chip manufacturing plants and research teams with silicon photonics expertise are concentrated. That mix matters because graphene devices still demand careful handling and testing, and places with clean rooms and packaging lines can move faster. Buyers in the region are also the first to feel the squeeze when old links run out of headroom.

## Modulators carry the early workload

At the heart of this work are modulators, the components that imprint information onto a beam of light. Graphene-based versions took **45%** of the market in 2026. Their appeal is simple: they are small, switch fast and consume little compared with the alternatives. In a data centre, where thousands of links run in parallel, that combination adds up.

The engineering challenge is not small: every device has to behave the same as the one next to it, batch after batch. That is where the technology's industrial credibility is decided. Devices have reached **67 GHz of bandwidth** with **80 Gb/s modulation** on silicon, which puts the technique beyond theory. The open question is whether it can be manufactured cheaply and repeatably for everyday use.

## Graphene transfer, the bottleneck

Making a device in a laboratory is one thing. Making millions on a wafer is another, and that is where graphene photonics faces its toughest test. Graphene has to be grown elsewhere and then transferred onto the silicon wafer, and that transfer can leave wrinkles, cracks or particles. Any of those defects scatters light or makes one device behave differently from its neighbour.

The industry's answer has been to treat manufacturing as a research problem. Teams are working on cleaner transfer methods, protective layers that shield graphene during processing, and foundry steps that fit the material into chip fabrication flows. Advances are being reported, including wafer-scale processes that cover large areas with usable graphene. Until yields rise and costs fall, graphene photonics will remain concentrated where performance matters more than price: high-end computing.

## Sensors and optical computing as a second outlet

Data centres dominate the conversation, but they are not the only place where light chips earn their keep. Optical sensing uses the same building blocks to detect tiny changes in light, which can reveal chemical traces, physical strain or biological signals. Optical computing, still at an earlier stage, attempts to perform calculations with light instead of electricity, with promises of estimulante ilegal for certain kinds of tasks.

That breadth spreads the risk. If one application stalls over cost or performance, another can advance and keep fabrication lines and research funding alive. Different buyers tolerate price and reliability differently, so a technology with several homes is less exposed than one tied to a single customer. What is learned in sensors or computing flows back into data centre designs, and vice versa.

The horizon reads better as a story of steady engineering than as a sudden revolution. Demand for artificial intelligence workloads and cloud services keeps pushing bandwidth upward, and public support in Europe, including work under the **European Chips Act**, is helping pilot lines. Data centre power consumption is under scrutiny, which rewards technology that moves information with less energy. If wafer-scale manufacturing matures, graphene silicon photonics could become routine. As with any industrial bet, it is wise to look at actual yields before taking the forecast for granted.