As AI puts growing pressure on power infrastructure, Mark Rushworth, founder and CEO of Finchetto, explains why the UK has an opportunity to become a major player in the emerging photonics market – but must act before that advantage moves elsewhere.

Data centres now consume roughly 2.5% of UK electricity, with the grid queue to connect new facilities growing by 460% in the first half of 2025 alone. Globally, electricity demand from AI-focused data centres surged 50% in a single year and is projected to triple by 2030, according to the IEA. Britain has promised to become the fastest adopter of AI in the G7, but the reality is that the grid is strained and billions in announced infrastructure investment appear illusory.

One lever for addressing this critical issue, while creating a distinctive industrial opportunity for the UK, is photonics. The process of using light rather than electrons to carry and process data is beginning to move into parts of the data centre that electronics has always owned. It will not solve the energy problem by itself, and it is not the only lever operators have. But it is one of the few that has the potential to address cost and carbon at the same time.

What photonics can do, and when

Electrons have long carried data inside computing devices, but they come with limits: heat generation, power inefficiency and bandwidth bottlenecks. Every time a signal switches between optical and electronic domains, latency and energy costs increase. Photonics replaces these electrical pathways with optical ones, allowing data to travel at the speed of light while reducing energy consumption.

But photonics cannot simply rip and replace electronics. In electronics, silicon is the universal platform, and decades of investment have created capabilities in processing, manufacturing scale and integration. It is therefore natural that silicon has become the first major focus for photonics as well. But photonics is different: no single photonic material covers every need, so the technology arrives as a range of approaches, some widely deployed, some nearly there and some still further away.

Fibre and pluggable optical transceivers are widely deployed layers that already move data between and within facilities. Co-packaged optics is now emerging, bringing optical transceivers onto the same package as the compute chip to cut the distance data has to travel as an electrical signal. Further on the horizon are optical interposers, all-optical switching and, eventually, optical computing, where light handles the processing itself.

The greatest near-term impact is in moving data between chips, racks and servers in AI and high-performance networks, rather than replacing switching or computing wholesale. This is the layer where the energy savings are clearest and the integration path is shortest.

The barriers to scaling

The first barrier is integration. Photonics must work inside the electronic systems that already run data centres, rather than replacing them overnight. Any credible technology here must work with existing standards and protocols such as Ethernet, as well as the spine-and-leaf networks operators already run. The realistic path is a phased one: proof-of-concept trials, non-disruptive use cases and greenfield sites where the savings can be designed in from the start. Operators describe it as an evolution rather than a rip-and-replace, and that is likely to remain the practical route to adoption.

The second is cost. Emerging technologies carry a premium in their early years, and photonics is no exception. The capital needed to move from pilot to volume is real, and manufacturers working on newer material platforms are still raising substantial sums simply to scale production.

The third is manufacturing itself. The diversity of materials that makes photonics flexible, such as silicon nitride, indium phosphide and thin-film lithium niobate, also means there is no single dominant process platform, which complicates standardisation, supply chains and volume scaling. Packaging and testing, rather than the optical circuits themselves, are where much of the cost and difficulty still lie. And while the UK has real research and pilot strength, it does not yet have mature, high-volume domestic photonic integrated circuit (PIC) manufacturing.

The manufacturing opportunity hiding in plain sight

Advanced electronic semiconductor fabrication is a mature industry dominated by large overseas players. For the UK, the most realistic strategic position in electronics is to remain strong in design while playing a smaller, specialised role in global manufacturing. By contrast, PICs are a fast-growing emerging market in which the UK could hold a larger stake.

Manufacturing plays an important part. Unlike electronic chips, which are pushing towards 2-5 nanometre nodes and require vast, capital-intensive mega-fabs, PICs typically operate at feature sizes of around 200 nanometres today, requiring less complex manufacturing and lower capital investment.

Photonics manufacturing prioritises materials science and integration over extreme miniaturisation, playing to areas where the UK already holds considerable strengths. Cornerstone in Southampton, the CSA Catapult in Newport and the wider photonics ecosystem in Wales, alongside emerging R&D infrastructure in Glasgow and Sheffield, provide a foundation from which manufacturing capacity could be scaled.

This is a high-value, low-volume opportunity for the UK, rather than a bid to out-build Taiwan. But it is dependent on closing the pilot-to-volume gap.

The photonics power shift is already underway

Leading players in the industry are already moving to embrace this shift. Nvidia has built co-packaged optical interconnects that bring light closer to silicon to reduce power requirements and increase bandwidth. It has also recently invested heavily in photonics suppliers, signalling the growing importance of optical technology to future infrastructure. Other hyperscalers are also redesigning networks around light-based platforms as they look to address similar power and bandwidth constraints.

The direction of travel reinforces the case. JLL expects inference to overtake training as the dominant AI workload in 2027 and AI to account for around half of all data centre workloads by 2030. Because inference is latency-sensitive and needs to sit close to users, it puts even more weight on how efficiently data moves, which is precisely where photonics is strongest in the near term.

An open opportunity with limited time

The UK Council for Science and Technology warned in March that this window is significant but time-limited, with the US, China, Taiwan and Singapore already investing heavily in national photonics capabilities. Meanwhile, the European Commission approved €211 million in state aid for photonic chip development in Italy alone in April 2026, while the EU Chips Act was revised in June to bring photonic chips into core research programmes.

The UK needs a national strategy in which industry, government and academia collaborate from prototype to production. The government’s recently announced £1.1 billion AI Hardware Plan is aimed at building sovereign capability in ‘emerging paradigms such as photonics and edge AI inference’, while at the same time ‘not seeking to replicate the global supply chain’. This differentiation reflects the reality that the UK has particular strengths that could support jobs and wider economic value if they can be translated into manufacturing capacity.

Photonics will be one of the physical foundations of the next digital economy, with AI, quantum computing and cloud networking all depending on the ability to move and process data faster, cooler and more efficiently. The foundations in the UK are real, but so are the barriers: the manufacturing base, the integration work, the capital to scale and the time needed to develop the ecosystem. The question is whether the country builds that base to capture the near-term interconnect opportunity before the advantage settles somewhere else.