GPU Colocation in Edinburgh
GPU colocation in Edinburgh delivers PUE of 1.15–1.30 thanks to a cool climate (~10°C annual average), Scottish renewable-heavy grid, and transatlantic subsea-cable adjacency (FLAG, AEC). Best for long-term training where PUE compounds, Scottish public-sector AI, and transatlantic-replicated systems. Latency to London: ~12ms. Operators: Pulsant, Iomart, Brightsolid.
Why does Edinburgh have a real claim on AI infrastructure?
Edinburgh is small relative to London, Slough, or Dublin, but it has a structural argument that those markets can't make: the climate. Edinburgh's annual mean temperature sits around 10°C, with cool/cold conditions for most of the year. For AI training workloads, this translates directly into lower mechanical cooling demand and PUE figures that are difficult to match in the M4 corridor. Operators here can run free or near-free cooling for a substantial share of the year, which materially shifts the economics of dense compute.
The second leg of the Edinburgh story is connectivity. Edinburgh hosts the southern landing for several transatlantic and Nordic-route subsea cables (FLAG, AEC, North Atlantic Loop). For multi-region AI deployments wanting low-latency replication to North America or to Iceland/Faroe Islands compute, Edinburgh sits naturally on the path.
Which operators offer GPU colocation in Edinburgh?
- Pulsant Edinburgh — the largest established operator in the city, with Tier III capacity and strong public-sector tenant base (Scottish Government, NHS Scotland).
- Iomart Edinburgh — managed colo serving regulated UK tenants and Scottish-anchored commercial buyers.
- Brightsolid — Dundee-anchored but serves the wider Scottish market and is part of DC Thomson.
- Glasgow corridor — Glasgow sits 50 miles west and has additional regional capacity (Pulsant, Iomart, smaller operators) that Edinburgh-targeted buyers should consider as part of the same effective market.
- Hyperscale campuses in Aberdeen and Highland — outside the Edinburgh metropolitan area but worth flagging: Scotland is becoming an attractive destination for purpose-built AI campuses leveraging the climate. We track several emerging hyperscale sites.
How much does Edinburgh's free-cooling save in real money?
Industry data on PUE in cool-climate UK locations puts Edinburgh-area facilities in the 1.15–1.30 band, against 1.30–1.50 typical for retrofit M4-corridor capacity. For a 1MW deployment over 36 months, that PUE delta translates into real money — six-figure savings range, depending on contract structure and PPA. We can model this for specific deployment specs.
The caveat: extracting the full free-cooling benefit requires the facility to be designed (or thoroughly retrofitted) to use ambient air or evaporative cooling. Older Edinburgh facilities running standard mechanical cooling don't capture the full advantage. The newer purpose-built capacity does.
When is Edinburgh the right place for GPU colocation?
- Long-term training clusters where PUE compounds — 36-month or 60-month commits where the cooling-cost saving offsets the latency penalty for non-real-time workloads.
- Scottish Government and NHS Scotland AI workloads — devolved data residency, public-sector procurement frameworks.
- Transatlantic-replicated AI systems — model serving with US east-coast replication where Edinburgh's subsea-cable adjacency matters.
- University-anchored research — Edinburgh University's School of Informatics is one of the strongest AI research clusters in Europe; commercial AI tied to those collaborations benefits from local compute.
- Sustainability-anchored deployments — Scottish grid mix is heavily renewable (wind dominant); operators routinely sign 100% renewable-matched contracts.
What should you ask Edinburgh operators about GPU colocation?
- What share of annual cooling load is free or evaporative vs mechanical?
- Demonstrated PUE — independently verified, not marketing
- Density ceiling at the rack and at the row (often the binding constraint at older sites)
- Carrier diversity and subsea-cable presence (varies sharply across operators)
- Cleared-personnel availability if your workload involves UK Government data
What are Edinburgh's latency numbers to London and elsewhere?
Edinburgh sits roughly 12ms RTT from London Docklands over standard Tier-1 carriers. For real-time UK inference this typically rules Edinburgh out for a primary deployment — but it's a fine choice for replicated inference tiers serving Scottish or northern-English end-users, or for training nodes feeding a London-anchored serving tier.
To Iceland (atNorth, Verne Global): 18–25ms via undersea routes — relevant for buyers operating mixed Iceland/UK compute pools. To US east-coast: ~70ms typical, lower for premium subsea routes.
Find GPU colocation in Edinburgh
Tell us your spec and timeline. If Edinburgh's free-cooling advantage compounds for your workload, we'll shortlist; if not, we'll suggest a better-fit UK or European location.
Get Edinburgh Colocation Quotes