Caprelion opens talks for 2.5 MW AI node in Slovakia

Caprelion is starting early anchor-operator discussions for a planned roughly 2.5 MW high-density AI deployment in Košice, Slovakia, aimed at 2027 readiness. The move targets GPU operators planning European rollouts, including liquid-cooled NVIDIA GB300-class systems, as data-center capacity across Europe tightens.

Why it matters: - Caprelion is trying to lock in an early anchor customer before the Košice project’s final technical design is frozen. - Early participation could shape power, cooling, network and operating requirements around a future high-density AI deployment instead of forcing an operator to adapt later. - The project targets a market where European data-center space remains tight and AI-related capacity demand is rising.

What happened: - Caprelion opened early discussions for a planned approximately 2.5 MW high-density AI infrastructure deployment in Košice, Slovakia. - The target readiness window is the second half of 2027. - The company is seeking potential anchor GPU infrastructure operators planning European deployments for 2027–2028. - The Košice AI POD is being positioned as dedicated high-density AI infrastructure, not conventional rack-by-rack colocation.

The details: - The planned deployment is designed around an anchor-scale AI requirement. - Caprelion intends to factor the selected anchor operator’s requirements into the final infrastructure configuration. - The interface areas under discussion include GPU platform, rack density, electrical interfaces, direct liquid cooling, network fabric, storage architecture, security requirements, commissioning, acceptance and the operating model. - Caprelion says early involvement is meant to give an operator a chance to influence the infrastructure envelope before design freeze. - An operator joining later may have to adapt its compute platform to a more fixed facility architecture. - The technical qualification process is built to accommodate operators considering GB200, GB300 or other high-density GPU architectures. - The current qualification framework requires the target platform class to be identified before the project’s final technical configuration is set. - NVIDIA describes GB300 NVL72 as a fully liquid-cooled rack-scale platform with 72 Blackwell Ultra GPUs and 36 Grace CPUs in one system. - Caprelion does not present the Košice project as already certified, approved or finally engineered by NVIDIA for GB300. - Any compatibility with a specific GPU configuration would need detailed engineering with the selected anchor operator and relevant technology partners. - The source text cites CBRE’s expectation that vacancy across primary and secondary European markets will fall to around 6% by the end of 2026. - The source text also says new colocation signings tied to AI workloads more than quadrupled in the first half of the year. - JLL has identified pre-leasing as an increasingly important occupier strategy as contiguous blocks of high-density capacity become harder to secure in established European data-center markets. - Caprelion included social links for LinkedIn and YouTube.

Between the lines: - Caprelion is signaling that future AI infrastructure will be shaped as much by site-specific power and cooling design as by access to GPUs. - The emphasis on pre-leasing and anchor participation reflects a broader scramble for scarce high-density European capacity. - The GB300 reference shows how quickly rack-scale liquid-cooled systems are raising the bar for facility design.

What's next: - Caprelion will continue qualification talks with prospective anchor operators. - The final technical configuration will be set only after the target platform class and operator requirements are defined. - Detailed engineering with the chosen operator and technology partners will determine whether a specific GPU stack can be supported at the Košice site.

The bottom line: - Caprelion is marketing Košice as an early-design opportunity for a future European AI buildout, not as a finished colocation product.