Orion Telekom expands Serbia AI project to 20 MW at National Data Center

Orion Telekom has outlined plans to expand its artificial intelligence computing project at Serbia’s National Data Center, starting with a 4 MW installation and potentially scaling it to 20 MW. The company said it intends to deploy NVIDIA GB300 NVL72 systems, with operations expected to begin in the first half of 2027.

The announcement puts Serbia on the map of a growing regional market for AI infrastructure. It also highlights a new type of electricity demand, one that is shaped by high-powered computing, cooling needs and the availability of reliable grid capacity.

The project includes liquid-cooling infrastructure and an initial customer commitment from Serbia’s incumbent telecommunications operator. Orion also said it has identified possible additional AI data-centre locations in southeastern Europe with around 120 MW of future capacity opportunities, although those sites are not yet approved or financed projects.

Neither Orion nor NVIDIA disclosed the capital expenditure for the Serbian installation. For investors, the key questions remain the same: how much power will be secured, how firm the customer demand is, and what financing structure will support the buildout.

AI computing brings a different power profile

Advanced AI data centres are not comparable to standard telecom facilities. They require dense computing hardware, more intensive cooling and stable electricity supply, all of which increase the complexity of the project.

The planned 4 MW phase is modest by global standards, but it is still a meaningful specialised deployment for Serbia. If the site reaches 20 MW, its electricity needs would rise sharply, creating a larger and more predictable industrial load for the local power system.

At full capacity, a 4 MW facility would theoretically consume about 35 GWh a year, while 20 MW would correspond to roughly 175 GWh. These are upper-bound calculations based on uninterrupted operation at stated capacity, not forecasts of actual use.

That distinction matters because installed capacity is not the same as continuous consumption. Actual demand will depend on utilisation, technical configuration and the role of cooling and other facility systems.

Grid access and financing are central

The project also shows how grid access can become as important as the computing equipment itself. AI infrastructure requires reliable power quality, redundancy and enough connection capacity to support operations without interruptions.

Whether extra transmission or distribution upgrades will be needed has not been disclosed. That will depend on the technical conditions at the National Data Center and the requirements of the relevant network operator.

For lenders, grid timing and connection costs are likely to be part of due diligence. If the project expands beyond the first phase, electricity infrastructure could become a significant part of total investment.

Power pricing may also shape the economics. For a theoretical 20 MW facility using 175 GWh annually, every €10/MWh change in delivered electricity price would alter annual electricity costs by about €1.75 million.

Potential role for renewables and storage

The project could also create opportunities for Serbian renewable-energy producers. Data-centre operators often look for predictable electricity costs and low-carbon supply arrangements, which can support long-term power purchase agreements.

Even so, renewables alone do not solve the problem of around-the-clock demand. Solar output is daytime-based, while wind varies with weather, so the facility would still need balancing, grid supply and other flexibility tools.

Battery storage could help with backup and short-duration flexibility, but it is not a full substitute for the redundancy systems required by critical computing infrastructure. The final setup will depend on outage requirements, grid conditions and the commercial value of flexibility.

Business case depends on utilisation

The main commercial question is not only how much electricity the site will consume, but whether the hardware can generate enough revenue over its useful life. High-end GPU systems are expensive and can lose value quickly as newer technology appears.

Orion’s initial customer commitment is therefore important, but the company has not disclosed the contract value, duration or pricing. Those details would be needed to assess how much demand is already secured.

For banks, the relevant issues are contracted revenue, customer concentration, depreciation and the residual value of the equipment. AI computing infrastructure combines elements of telecoms, energy and technology investment, which makes financing more complex than for traditional projects.

The announcement also does not provide enough detail to assess water use, cooling efficiency or possible environmental constraints. Those factors will matter once engineering and permitting move forward.

Serbia’s wider opportunity

If the project advances, it could support Serbia’s role as a provider of advanced digital services rather than only a location for connectivity infrastructure. AI computing capacity can be used remotely, which gives it export potential if the pricing, reliability and security standards are competitive.

Orion’s mention of around 120 MW of possible future locations across southeastern Europe suggests broader ambitions, but those plans remain at the opportunity stage. Separate financing, grid access and customer contracts would still be required.

For Serbia, the project links three areas at once: digital infrastructure, energy supply and technology financing. The next milestone will be the conversion of the announcement into detailed contracts, confirmed power arrangements and a financed construction plan.

For readers following Serbia’s business and infrastructure pipeline, this is another sign that the country’s energy system and digital economy are becoming more closely connected. More updates on such projects are available in our latest news coverage.