How are data centres evolving to keep pace with demand?

Last year was a monumental year for data centres in the UK. Roughly 100 new data centres were planned for construction over the next five years in 2025, marking a new era of digital infrastructure expansion and investment.

It’s no surprise that the principal driving factor behind the increase in buildouts is Artificial Intelligence (AI) and its associated Large Language Models (LLMs), which require a vast amount of computing power, energy, and specialised infrastructure.

So with the UK now firmly established as a data centre hotbed, how will these buildouts handle these power-constrained environments in the year to come? Let’s take a look at some specialised infrastructure we can see adopted in 2026 and beyond.

1.6T transceivers

Transceivers are ultimately the unsung heroes of data centres. While General and Central Processing Units (GPU/CPU) tend to dominate industry news, they wouldn’t be facilitated without the optical transceivers that actually move the data between them.

By converting electrical signals to optical signals, transceivers enable high speed communication over fibre at 100G, 400G, 800G - and now, 1600G. Referred to as ‘1.6T’, these transceivers already had made a huge impact on AI data centres in 2025 by empowering ultra-high-speed data transmission at 1.6 Terrabits per second.

The transceivers transform infrastructure by doubling the capacity of current 800G modules for faster exchange between Central Processing Units (CPUs) and storage in AI training clusters. In turn, this leads to longer inter-switch links up to 2km, extending bandwidth density and evolving existing architectures.

The implementation of these 1.6T transceivers is expected to reach over 10+ million this year, with companies like AddOn Networks, the world’s largest independent manufacturer of optical transceivers and high-speed cabling, expanding their transceiver offering with the addition of 1.6T last month to bolster low insertion loss and strong signal integrity.

Quantum preparations

Even with specialised GPUs and chips, some data centre workloads like power consumption and physical space are extremely resource intensive. With traditional scaling approaches reaching limits, quantum computing is seen as the solution to bridge this gap.

Quantum computing uses the principles of quantum mechanics to solve complex problems far beyond the capabilities of traditional, binary based computers. While fault-tolerant quantum computers are generally expected to be widespread in 5 to 15 years, quantum data centres from the likes of IBM have already appeared in Germany. 

But with quantum computers eventually possessing the capabilities to break traditional encryption methods like Rivest Shamir Adleman (RSA), defence is essential, and comes in the form of post-quantum cryptography (PQC).

PQC algorithms are designed to run on classical computers while resisting from quantum attacks. But with 84% of IT and telecoms operators currently not having any formal PQC plan in place according to the Trusted Computing Group’s State of PQC Readiness report, it will be intriguing to see how data centre operators pair the integration of quantum and PQC in years to come.

Precision, precision, precision

Precision proved everything for large-scale AI deployments in 2025 given the immense scale and tight latency requirements distributed across clusters. This demand for exact data movement pushed optical circuit switching (OCS) to the forefront of buildout requests.

OCS is an all-optical networking technology that directs light signals between fibres without electrical conversion, offering sub-microsecond latency and high Tbps-level bandwidth essential for hyperscale environments.

This means that data centres are no longer constrained by power and bandwidth bottlenecks of traditional electrical switches, with solutions like POLATIS® from HUBER+SUHNER offering the means to reconfigure GPU clusters to enable more efficient AI model training and traffic demands.

Through 2026, expect to see OCS develop beyond simple point-to-point links to become more tightly integrated with AI. This deeper integration will support scaling of up to trillion-parameter model training, making large-scale AI systems more efficient and economically viable to deploy.

Looking ahead

With data centre power capacity accelerating to roughly 250% by 2030, there’s no doubt we can expect to see 1.6T transceivers, quantum architectures, and advanced optical circuit switching hold the fort along with a whole host of other solutions to support this rapid data centre growth.