What’s next for the future of data center networking? Optics.

Every time an artificial intelligence (AI) system automates a task or writes code, data races between processors inside a data center. And the rate of data is exploding.

”The sheer amount of data needed to process these models has ballooned to a point where no single processor can get the job done,” said Michael Lu, optical networking and signal-chain business lead at TI.

So engineers are distributing the workload across many processors: on a single board or rack; across multiple rows of racks; and across entire data center facilities, all working as one system.

Each scaling level spans longer distances and demands more bandwidth per interconnect than the last. For engineers, the challenge is moving data fast enough to keep processors running at full capacity. When the links between them lag, performance suffers. Response times stall, AI model training slows, and expensive computing resources sit idle waiting on data.

”For a long time, engineers used copper and electrical networking in data centers,” said Brian Ginsburg, TI Fellow and director of radio frequency and photonics at TI’s Kilby Labs.

Copper remains cheap and reliable, but it hits physical limits at long distances, with signals degrading before they reach their destination.

That’s why engineers are now combining electrical networks with the next frontier of AI data center networking: optics. In addition to moving electrons through copper, optics transmit data as pulses of light, and can do so across longer distances.

Optical networking has become a fast-growing area of investment in AI infrastructure, with analysts projecting the global sales of AI-related optical transceivers to reach $26 billion in 2026, growing about 60% year over year.

The time to prepare for optics is now, but this shift is bringing new challenges and considerations for engineers. We sat down with Michael and Brian to understand the scale of adopting optics and what the challenges are.

Q: From a design engineer’s perspective, how did optics move from an option to a requirement?

A: Brian: Data rates are skyrocketing. To put that into perspective, optical transceivers – which convert electrical signals to optical signals – are quadrupling from 400G designs to 1.6T transceivers. This means that every second, a transceiver moves 1.6 trillion bits of information. In the future, transceivers are scaling towards 12.8T, which is enough to stream 512,000 ultra-HD movies simultaneously.

As data rates grow, an inverse relationship forms between the amount of bandwidth being pushed and the distance the bandwidth can travel electrically. As we’re moving faster and reaching hundreds of gigabits, only the shortest communication distances using copper can survive without losses, and these copper-capable distances are shrinking with each generation.

But optics are effectively lossless, since light travels through fiber-optic cables.

Michael: Engineers aren’t just concerned with sending data at faster rates. They also want to squeeze more data into a fixed area so that they can send more data from one place to another. Copper cables are notoriously thick compared to optical cables. Fiber-optic cables can be grouped more closely together than copper cables.

Q: If optics are becoming essential, what technology is making optical connectivity scalable?

A: Michael: Light can navigate through silicon like a highway, similar to how electrons move through copper. To a photon at the wavelengths used in data centers (1310nm and 1550nm), silicon is essentially transparent. The implementation of optics via silicon is called silicon photonics. With silicon photonics, engineers design photonics ICs so that light is confined in a silicon channel called a waveguide. Engineers steer it, split it, and modulate it with electrical data. By the time the transmit signal leaves the chip, it has gone from electrical to optical.

A similar phenomenon happens in reverse for the receive direction. Optical input data is routed via silicon waveguide to a photodiode in the silicon photonics chip that converts photons to electrical current.

Brian: Optics have always been expensive. Building an optical link has traditionally meant relying on exotic materials that are difficult to process, which kept costs high. Silicon photonics changes that.

Silicon photonics uses the same silicon wafers already used for ordinary chips, unlike alternative optical materials such as indium phosphide, which requires smaller wafers. Because the wafers are so much bigger, each one yields far more chips.

Q: How is TI helping engineers scale their optical systems?

Brian: In data center networking, system bandwidth is doubling every four years. Engineers must scale it without using more power or money. TI makes the components that surround the optical signal path and keep it running: power delivery, microcontrollers, and the analog front ends that stabilize each optical module. What also sets us apart is flexibility. Every optical technology needs a different control scheme, and we can build those front ends quickly, tailored to each scheme. Whatever technologies and materials the industry adopts, TI has the backbone to support them.

Q: Finally, optics is here. What does that mean for the future of networking?

Michael: In the future, data rates are only going to increase. That doesn't mean copper disappears. But more and more, optics is a must. And these two will coexist.

Brian: The future is silicon photonics. It can support incredible capacity demands at low cost as optical interconnects are just exploding throughout data centers. I can’t wait to see how networking evolves because of it.