Croatia has approved the power supply blueprint for a proposed one gigawatt AI campus, connecting one of Europe’s largest prospective computing projects to an extensive reconstruction of the country’s electricity network.

Pantheon AI said national transmission operator HOPS had confirmed the technical and regulatory feasibility of supplying its planned site in Topusko. The approved design covers more than €500 million of infrastructure, including 280 kilometres of transmission and fibre lines, new roads, and a 400kV substation.

The developer intends to provide 800MW of usable IT load across a 310 acre campus and says construction remains scheduled to begin in early 2027. Pantheon would build the associated grid infrastructure before transferring it to the Croatian state for permanent ownership and operation.

Approval of the blueprint moves the proposal beyond its earliest conceptual stage, although it does not complete the financing, procurement, customer contracting, grid connection, or construction needed to create an operating datacentre. Projects of comparable scale can change substantially between an initial technical decision and the arrival of the first computing equipment.

The datacentre becomes a grid programme

Pantheon says the network work could open access for more than five gigawatts of renewable generation currently constrained by limited transmission capacity. If those lines and substations are delivered as described, infrastructure built initially around a large electricity consumer could also strengthen Croatia’s wider power system.

Realising that benefit will depend on the final ownership terms, the location and timing of renewable projects, and how the network manages demand when generation falls. A datacentre operating continuously cannot rely solely on the annual volume of renewable electricity produced if the system lacks sufficient storage, dispatchable power, or interconnection during periods of lower output.

Claims of fully renewable operation therefore require a more precise account of how electricity will be matched. A campus can buy enough renewable power over a year while still drawing from the general grid during individual hours, whereas continuous matching requires generation and storage capable of following the facility’s actual load.

The proposed scale also demonstrates why AI infrastructure planning increasingly begins with electricity rather than servers. A one gigawatt campus is a power system undertaking whose feasibility depends on generation, transmission, cooling, land, water, roads, and long term agreements with network operators.

Croatia offers land and potential power capacity outside Europe’s most congested datacentre markets, where connection queues and planning limits have delayed construction. However, locating a campus away from established clusters also requires the developer to establish resilient fibre routes, equipment supply chains, maintenance capacity, security, and access to engineers familiar with high density computing.

Execution sits behind the headline number

Pantheon describes the campus as being designed around Nvidia’s gigawatt scale AI factory specifications, with Croatian engineering businesses participating in transmission and substation work. Local involvement could extend the economic effect beyond construction if domestic suppliers retain lasting capability in grid engineering, cooling, operations, and digital infrastructure.

Employment claims will need to distinguish between construction and permanent roles because large sites require substantial labour during development but can operate with comparatively small teams once commissioned. Municipal services, housing, transport, and local supply chains can still face considerable pressure during a build lasting several years.

Customer demand presents another uncertainty. Cloud providers, model developers, and infrastructure companies are reserving large blocks of AI capacity, but processor efficiency and model design continue to change. A campus designed for current assumptions about rack density and cooling must remain useful across several generations of hardware.

Financing has a similar relationship with contracted demand. Large infrastructure investors generally require credible long term customers before committing the full cost of a campus, while prospective customers want assurance that power, equipment, and completion dates are secure before signing.

The project’s regional effects will also depend on who pays for the grid reinforcement and how the resulting capacity is allocated. Transferring privately built assets into state ownership can support national infrastructure, but public authorities must avoid allowing one development to absorb network investment needed by industry, housing, or other economic activity.

Central and eastern European countries have an opportunity to capture a larger share of Europe’s computing investment as pressure grows in Frankfurt, Dublin, Amsterdam, London, and Paris. Lower land costs alone will not be enough, because customers place a premium on power reliability, political stability, connectivity, and the ability to deliver complex facilities on time.

HOPS’ approval advances the Topusko project by addressing a constraint that has halted numerous European datacentre plans. Pantheon must now convert a technically feasible grid design into funded infrastructure, committed customers, and a site capable of operating at a scale rarely attempted in the region.