Summary
- Brinell Compute plans a large AI campus in the Rakovski Industrial Zone north of Plovdiv.
- The developer has set an autumn 2026 construction target across a site of approximately 809 hectares.
- No public MW figure, grid agreement, customer, financing package, principal contractor, or detailed delivery programme has been released.
Brinell Compute says construction of its proposed €3 billion AI data centre campus near Plovdiv will begin in autumn 2026, setting an aggressive start date for one of Central and Eastern Europe’s largest claimed digital-infrastructure investments.
The project is planned for the Rakovski Industrial Zone in Maritsa municipality, on a site reported to cover approximately 2,000 acres, or 809 hectares. Bulgaria’s government approved a memorandum of intent and understanding with the German developer earlier this year.
Brinell has not disclosed the intended IT capacity, total electrical demand, grid connection, customer commitments, financing structure, planning position, principal contractors, or phased construction schedule. The scope described as beginning in autumn could therefore range from early enabling work to the first data hall.
Land provides scale, not capacity
The site is large enough to support a multi-building campus with substations, cooling plant, generators, storage, offices, logistics, and future phases. Its size also expands the environmental and infrastructure work required before computing equipment can be installed.
Road access, drainage, ground conditions, biodiversity, water, noise, security, and high-voltage cable routes will all need to be established across an area considerably larger than most European data centre sites. Even where industrial zoning is in place, individual buildings and utility systems may require further permits and technical approvals.
A €3 billion investment value does not reveal the campus load. The electrical baseline needs to define the first energised phase, the ultimate demand, the connection voltage, substation arrangement, reinforcement programme, and redundancy standard. Without those figures, the project cannot be compared meaningfully with other AI campuses.
Cooling remains equally opaque. AI clusters increasingly use direct-to-chip liquid systems at the rack, but the site still needs a complete route for moving heat to the outside environment. Cooling distribution units, pumping, water treatment, heat exchangers, dry coolers or evaporative plant, and control systems will account for a substantial part of the mechanical programme.
Bulgaria’s climate includes hot summer periods that need to be reflected in the design conditions. A system optimised around favourable annual averages can lose efficiency or capacity during peak ambient temperatures, when both the power network and cooling plant may already be under greater stress.
Power will decide the build rate
Large AI campuses can require several hundred megawatts if built to hyperscale. Securing that demand normally involves transmission studies, new substations, protection changes, land rights for cables or overhead lines, and long-lead transformers.
An autumn start could be credible for grading, roads, drainage, temporary utilities, fencing, or demolition, even if the final grid programme remains under development. Main data hall construction would require a firmer link between the building schedule and the date at which permanent power becomes available.
Financing is likely to follow the phases rather than cover the entire campus from the outset. Developers usually match each tranche of construction with customer demand, power delivery, and equipment procurement, limiting the amount of capital committed to empty buildings.
Brinell has referred to discussions with large US technology companies, although no customer or binding capacity reservation has been announced. A signed anchor lease would support financing and establish the required density, cooling interface, network architecture, and commissioning programme.
Plovdiv offers lower land and labour costs than Europe’s largest data centre markets, while Bulgaria can provide a base for serving South-East European demand. The site’s position outside an established hyperscale cluster places greater weight on fibre development, customer confidence, and access to experienced contractors and operators.
Diverse long-haul connectivity will be necessary for moving training data, replicating storage, distributing models, and linking the campus with other cloud regions. The physical routes need to avoid common ducts, bridges, road crossings, or carrier dependencies that could create a single point of failure.
The local supply chain will also need to expand around the project. A large campus draws on civil contractors, structural steel, electrical and mechanical installers, controls specialists, commissioning engineers, and permanent operations staff, while critical equipment may still be imported from other European or global factories.
Government support can assist with coordination, but it does not replace customer contracts, permits, grid capacity, or project finance. The development will become easier to assess when Brinell identifies the initial works package, the first-phase load, the utility route, and the parties responsible for design and construction.
Until those details emerge, the autumn date marks the developer’s intended start rather than confirmation of a fully financed, energised campus programme. Early site activity may still provide the first visible evidence that the proposal is moving beyond its government memorandum and into physical delivery.
