Silicon Photonics for Agentic AI

For years, advancements in AI have focused on intelligence, but today the limiting factor is the invisible infrastructure that allows intelligence to move. AI hardware typically relies on moving electrons through copper wires, a process that is increasingly stifled by its physical limits. As AI demand skyrockets, the pace required for data transmission is increasingly constrained by the resistance of copper, signal degradation, and immense heat generation requiring energy intensive cooling systems.

This is not unusual. Technological breakthroughs rarely eliminate constraints altogether; they tend to move them elsewhere in the system. Railways became a story about steel production. The internet became a story about bandwidth and fibre optic networks. As AI scales, the bottleneck is increasingly shifting away from computation and towards the movement of information.

Photonics: A Solution to the AI Infrastructure Problem

To advance the next generation of AI, systems will need to move far greater volumes of information than today's architectures were designed to handle. This becomes particularly important as AI evolves from generating responses to taking actions. Unlike a chatbot, an agent may need to access memory, interact with software tools, query databases, and coordinate information across multiple areas. Every additional interaction increases the demands placed on the underlying infrastructure. Photonics offers one such solution. Specifically, photonics integrated into electronics, known as Silicon Photonics or Photonic Integrated Circuits (PICs).

By replacing resistive copper traces of traditional interconnects with nanoscale silicon waveguides, we can transmit data as photons, at much lower energy loss and higher speeds compared to electrical signalling. Through co-packaged optics (CPOs), we can integrate these optical architectures directly onto the same substrate as the AI processor (like a GPU). This reduces the loss associated with electrical signalling, where power consumption scales exponentially with distance of signal transmission. By shortening the electrical signal path from inches to a few millimetres, CPOs allow us to move data between compute and memory at terabit-per-second bandwidths, creating the low-latency, high-concurrency systems that agentic AI requires. 

The Transition: Electrons to Photons

This transition is already underway, with optical interconnects and co-packaged optics being increasingly deployed in AI data centres worldwide to handle the massive data requirements of AI compute. Industry leaders are now integrating silicon photonics directly into their hardware, proving that photonics is essential in enabling next generation AI. However, despite the paramount importance of silicon photonics in enabling next generation AI, it is not widely recognised by governments as a critical technology. 

One reason photonics have struggled to attract broader attention is that foundational technologies are often invisible when they work well. Most people encounter AI through applications, interfaces, and autonomous systems. They rarely see the infrastructure underneath. For example, fibre optic cables transformed the internet long before most people understood how they worked. Semiconductor architectures shape the digital economy despite remaining largely invisible to end users. Photonics face a similar challenge.

Long before agentic AI became a policy priority, photonics researchers were working on many of the challenges that are now emerging as constraints on future AI systems. What once appeared to be a highly specialised field is increasingly underpinning advancements in AI, computing and communications. As attention shifts towards bandwidth, energy efficiency and information movement, capabilities that were previously peripheral to the AI debate are becoming much harder to ignore.

Technologies deliver economic and societal value when they become embedded across markets, standards, procurement systems and everyday practice. Technical success alone rarely guarantees that outcome. Investment, coordination, skills, regulation and demand must evolve together before a scientific breakthrough becomes part of everyday infrastructure. For photonics, it’s not just about developing the technology, but creating the conditions for it to be adopted and scaled across the economy.

The AI debate is framed as a race for intelligence. Increasingly, it may become a race for the infrastructure that allows intelligence to move. Long-term advantages are often created at these enabling layers. If this proves to be true, silicon photonics may turn out to be one of the most strategically important opportunities hiding in plain sight.

Authors

Technology and Innovation programme activities

techUK bring members, industry stakeholders, and UK Government together to champion emerging technologies as an integral part of the UK economy. We help to create an environment where innovation can flourish, helping our members to build relationships, showcase their technology, and grow their business. Visit the programme page here.