GPU capacity in Frankfurt.
Frankfurt is Europe's interconnection capital and the F in FLAP-D. It is the right answer for latency and frequently the wrong one for training economics.
What decides a deployment in this metro.
Why buyers come here, and what they run into.
Frankfurt's value is the network. One of the world's largest internet exchanges sits here, with carrier and cloud on-ramp density that no other continental European city matches. For anything serving European users where latency is the metric, that is worth paying for.
What it does not have is spare power. Connection queues are real, German energy efficiency requirements set expectations that not every legacy building can meet at AI densities, and the qualifying list for a current-generation GPU rack is considerably shorter than the market's overall capacity suggests.
The design that actually works.
The design we run repeatedly: inference in Frankfurt where the network is, sustained training placed where power is cheap and available, connected privately. Within the EU that means no import complications on either leg.
Where a buyer insists on training in Frankfurt, we say plainly what it will cost and how long the connection will take, rather than discovering it together in month four. Wider context sits on the Germany desk.
Start from what is verified.
Current verified lines with quantities, lead times and indicative pricing are public on the live inventory. Anything not listed becomes a sourcing requirement with a first pass inside one business day.
Colocation in Frankfurt, through the same desk.
The desk sources colocation, racks and power in this metro as well as the compute that fills them. Frankfurt is the F in FLAP-D and power there is the binding constraint, which pushes new AI-density halls to the fringe of the metro. Requirements come to us as three numbers: megawatts at day one, kilowatts per rack, and the date. The space and power desk takes it from there, and above 40 kW per rack we will tell you honestly how short the local shortlist is.
Asked first, answered straight.
Yes. Space, power and cooling in this metro route through the same desk as the compute, priced against your density and date. Above 40 kW per rack the local shortlist narrows quickly, and we say so rather than tour you through halls that cannot take the load.
For interconnection and pan-European latency, yes. For sustained training, rarely. The workload does not benefit from exchange proximity and the power premium is significant.
Many cannot. Cooling and power per rack are the gating questions, and the qualifying list is shorter than headline capacity implies. We qualify on that before discussing hardware.
The Nordics, Spain and capacity outside the metro all price better on power. Keep inference in Frankfurt and connect privately; it is the pattern most European AI operations converge on.
Not for EU-located hardware, which is a large part of our verified inventory. Clearance comes off the critical path entirely.
Working through this on a real requirement?
Twenty minutes with the desk, no pitch and no quote at the end of it. Tell us roughly what you need and we will come back within one business day.
Sent. We are on it.
Your enquiry has landed with the desk. Acknowledged within one hour.