Metro data centers: the AI boom’s urban challenge – and how switching can help

AI workloads aren’t just bigger, they behave differently. They generate dense, high-intensity traffic between general processing units (GPUs), creating thermal hotspots and stressing power and cooling systems beyond typical enterprise IT. For metro data centers, which often face strict planning regulations and limited ability to expand, this shift presents a major challenge.

Exploring the energy crunch

Urban data centers are already grappling with energy constraints. In cities like London, Dublin, Frankfurt and Amsterdam, grid capacity is at a premium. AI is exacerbating this, with Europe’s data center electricity demand expected to rise 20% by 2030. And, when you’re operating in a metro footprint where every kilowatt counts, efficiency becomes a crucial imperative.

Yet much of the conversation has focused on compute power and cooling. A less discussed, but equally vital, component is the network switch. In AI-heavy environments, data is constantly moving between GPUs, often requiring conversion between optical and electronic signals at each switch. This optoelectronic process consumes significant energy and generates heat, multiplying the infrastructure burden.

The photonics revolution at the edge

One of the most promising advances to address this is fully optical switching. By keeping data in the optical domain throughout the routing process, photonic switches eliminate the energy-hungry optical-electronic conversions that dominate traditional switching. For metro and edge data centers, this innovation is particularly valuable. Optical switching can significantly reduce both energy consumption and latency, two critical factors in dense urban deployments.

UK-based startup Finchetto is pioneering a packet-based optical switch that’s already turning heads. Unlike traditional optical circuit switches that are limited to backbone networks, Finchetto’s technology can be deployed at rack level, making it highly relevant for metro sites. Their design reduces energy use by up to 53 times and achieves latencies as low as 40 nanoseconds, freeing up precious power and cooling headroom.

Heat reuse: from burden into benefit

Metro data centers have an added incentive to innovate. Their proximity to urban infrastructure opens opportunities for heat reuse. AI-intensive workloads generate significant heat which, if captured and redirected, can supply energy to nearby housing developments, schools, or hospitals. This not only improves overall site efficiency but also builds stronger ties with local communities – a growing factor in planning approvals.

There are already use cases of heat reuse schemes across Europe, from Helsinki’s data centers warming residential districts to London’s ambitions to tap data center heat for social housing. But to make these schemes viable, data center operators need to reduce inefficiencies elsewhere – for example in networking – so that waste heat can be reliably captured and channelled.

A holistic urban strategy

Switching technology is just one piece of the puzzle. Metro data centers need an integrated approach that aligns power, cooling, and networking in a tightly choreographed topology. Advanced liquid cooling methods like direct-to-chip and immersion are being combined with modular builds that can flex density within tight urban footprints.

Optical switching is unlocking new design possibilities, allowing operators to rethink rack configurations and cooling layouts. By reducing the power and thermal load of switching, photonic technology gives metro operators the flexibility to pack more compute into limited spaces – without tipping the balance on cooling or risking grid overload.

A bright future ahead

As AI reshapes the digital landscape, metro data centers will play a pivotal role in bringing AI services closer to users, enabling low-latency applications from autonomous vehicles to real-time language processing. But with city grids under strain and planning approvals getting tougher, these facilities must become smarter, leaner, and more integrated.

Investing early in technologies like photonic switching won’t just reduce operational costs, it will future-proof metro data centers to handle the surging tide of AI. For operators in the world’s busiest cities, that could mean the difference between scaling successfully and being left behind.

Urban infrastructure is evolving, and data centers need to evolve with it. The winners will be those that see efficiency not just as an operational target but as a competitive advantage in the AI era.

About the author:

Daren Watkins, chief revenue officer at VIRTUS Data Centres began his career as a graduate Military Officer in the RAF before moving into the commercial sector. He brings over 20 years experience in telecommunications and managed services gained at BT, MFS Worldcom, Level3 Communications, Attenda and COLT. He joins the VIRTUS team from euNetworks where he was Head of Sales for the UK, leading market changing deals with a number of large financial institutions and media agencies, and growing the company’s expertise in low latency trading. Darren is responsible for sales, service management and strategic account management of all VIRTUS customers. 

Article Topics

AI  |  AI infrastructure  |  EDGE Data Centers  |  metro data centers  |  optical switching  |  VIRTUS Data Centres