Orion Telekom plans to establish advanced artificial intelligence computing capacity at Serbia’s National Data Center, with an initial power requirement of 4 MW and the potential to expand the installation to 20 MW. The deployment would introduce a new category of electricity-intensive digital infrastructure into the Serbian market, linking telecommunications, computing services and energy procurement.
- AI Computing Creates Additional Electricity Demand
- Grid Capacity and Connection Costs Shape Expansion
- Renewable Power Procurement Offers Commercial Opportunities
- Battery Storage Could Support Reliability and Flexibility
- Customer Contracts and Hardware Depreciation Determine Returns
- Liquid Cooling Adds Engineering and Operational Requirements
- Serbia Targets Opportunities in Regional Digital Services
- Regional Capacity Opportunities Remain Uncommitted
- AI Data Centres Require Distinct Financing Structures
- Electricity Prices Affect the Economics of 20 MW Capacity
The project involves NVIDIA GB300 NVL72 systems, designed for demanding AI workloads, with operations expected to begin in the first half of 2027. The installation will include liquid-cooling infrastructure, while an initial customer commitment from Serbia’s incumbent telecommunications operator provides a starting point for commercial utilization.
Neither Orion Telekom nor NVIDIA disclosed the capital expenditure required for the Serbian installation. The project’s financial prospects will depend on the relationship between announced computing capacity, contracted demand, construction financing and the costs of operating and maintaining the equipment. Orion has also identified potential locations for additional AI data centres across southeastern Europe, representing approximately 120 MW of possible future capacity. These locations constitute expansion opportunities rather than approved or financed projects.
AI Computing Creates Additional Electricity Demand
Advanced AI data centres have substantially different power requirements from conventional telecommunications facilities. High-density computing systems require extensive electrical distribution, cooling equipment and backup infrastructure to maintain operational reliability. The planned 4 MW initial installation represents a specialized computing deployment in Serbia, while expansion to 20 MW would significantly increase its potential electricity requirements.
At continuous operation and full stated capacity, a 4 MW facility would consume approximately 35 GWh of electricity annually. At 20 MW, annual consumption would reach approximately 175 GWh under the same assumptions. These figures represent theoretical upper bounds rather than forecasts of actual consumption.
Realized demand will depend on equipment utilization, system configuration and whether the announced power capacity includes cooling and other facility loads. Installed capacity and continuous electricity consumption are therefore distinct measures. Nevertheless, an operational AI data centre could establish a relatively stable source of industrial electricity demand. The implications for Serbia’s electricity market would become more substantial if Orion achieves its planned 20 MW scale.
Grid Capacity and Connection Costs Shape Expansion
Reliable electricity supply is a fundamental requirement for high-density computing infrastructure. Data centres need stable power quality, redundant electrical systems and sufficient grid connection capacity, as interruptions can affect computing operations, customer commitments and equipment availability.
Securing electricity capacity can consequently become as important to project development as procuring the computing hardware. A 20 MW installation could require substantial connection infrastructure, depending on the existing capacity of the National Data Center and the configuration of its electrical systems.
The announcement does not establish whether additional transmission or distribution upgrades will be necessary. This will depend on the host facility’s technical capabilities, the actual electricity load profile and the requirements of the relevant network operator. If the project expands beyond its initial installation, grid connection and associated infrastructure could represent material components of total investment. Lenders will therefore need to assess available network capacity and connection timelines as part of their technical due diligence.
Renewable Power Procurement Offers Commercial Opportunities
The planned computing capacity could create an additional customer segment for Serbian renewable-energy producers. Data-centre operators seek predictable electricity costs and credible arrangements for procuring lower-carbon power, potentially creating opportunities for long-term power purchase agreements (PPAs).
For renewable-energy developers, a customer with reliable demand and sufficient creditworthiness can improve revenue visibility and potentially support project financing. Electricity procurement must account for the difference between renewable generation profiles and the continuous demand associated with data-centre operations.
Solar generation is concentrated during daylight hours, while wind output varies with weather conditions. AI computing facilities may require electricity around the clock, meaning that renewable PPAs alone do not eliminate the need for balancing, residual grid supply or other flexibility arrangements. For Serbia, the commercial opportunity could involve combining renewable electricity contracts with firm supply, balancing services and transparent consumption measurement. This could also support demand for structured electricity procurement products from utilities and independent suppliers.
Battery Storage Could Support Reliability and Flexibility
Battery energy storage systems could contribute to AI data-centre operations, although their economic value would depend on technical requirements and electricity-market conditions.
Batteries can provide short-duration backup, support power-quality management and deliver selected flexibility services. They may also help manage electricity demand during periods of high market prices. Storage cannot be treated as a direct substitute for the comprehensive backup and redundancy systems required by mission-critical computing infrastructure. The appropriate configuration depends on outage requirements, grid connections, backup generation and the commercial value of flexibility.
Investors would need to compare storage capital expenditure with its operational benefits. A data centre could benefit from battery systems without achieving sufficient electricity-price savings to justify a large standalone installation, making the business case specific to each project.
Customer Contracts and Hardware Depreciation Determine Returns
The principal commercial challenge for AI infrastructure is generating sufficient revenue from expensive computing equipment throughout its useful economic life. Financial performance depends on customer contracts, computing utilization, service pricing, operating expenditure and future hardware requirements.
Facilities with high utilization and long-term customer commitments may have materially different financial profiles from installations that rely on short-term or uncertain computing demand.
Orion’s reported initial customer commitment is therefore relevant to the project’s commercial prospects. The announcement does not disclose the contract’s value, duration, pricing structure or minimum utilization obligations. Without these details, the extent to which contracted demand supports project financing remains unclear.
Banks will need to examine contracted revenue, customer concentration, equipment depreciation and the residual value of computing hardware. Unlike conventional property and infrastructure assets, advanced AI processors can lose economic value rapidly as more efficient technologies become available. The resulting financing profile combines infrastructure investment with technological obsolescence risk, requiring careful assessment of both the facility’s long-lived physical assets and its shorter-lived computing equipment.
Liquid Cooling Adds Engineering and Operational Requirements
The planned deployment of NVIDIA GB300 NVL72 systems requires specialized thermal management to remove the heat generated by high-density computing equipment and maintain operating performance. Liquid cooling can offer advantages over conventional air-based systems for dense AI installations, but it also introduces additional engineering, maintenance and operational requirements. Cooling systems must be coordinated with electrical distribution, building infrastructure and equipment layouts.
Design deficiencies could affect reliability or require costly modifications after installation. Independent engineering reviews for project financiers should therefore examine the integration of computing hardware, cooling equipment, electrical systems and operational redundancy. The assessment should also cover the water, cooling and environmental infrastructure available at the host facility. The announcement does not provide sufficient information to quantify water consumption or cooling-system efficiency.
Serbia Targets Opportunities in Regional Digital Services
The installation could support Serbia’s development as a provider of advanced computing services to regional and international customers. AI processing can be delivered remotely through digital networks, with competitiveness depending on computing performance, service reliability, connectivity, data security and pricing. Serbia already has an established information technology and software-services sector. Additional domestic AI computing capacity could provide resources for software companies, enterprise customers and institutions requiring specialized processing capabilities.
Installing advanced computing equipment does not automatically establish an internationally competitive AI services business. Commercial performance will depend on customer availability, technical expertise and the ability to operate the infrastructure effectively. The project could connect data-centre operators, software developers, telecommunications companies and energy suppliers within a broader digital-services ecosystem.
Regional Capacity Opportunities Remain Uncommitted
Orion’s identification of approximately 120 MW of potential AI infrastructure locations across southeastern Europe indicates that its development plans extend beyond Serbia. At continuous operation and full stated capacity, 120 MW would correspond to approximately 1.05 TWh of annual electricity consumption.
This calculation is theoretical and should not be interpreted as a forecast of actual consumption or a confirmed investment programme. Individual developments would require separate financing, land or facility arrangements, grid connections and customer commitments.
The potential scale nevertheless highlights the importance of electricity infrastructure in AI investment decisions. Countries offering reliable power, competitive tariffs, digital connectivity and predictable permitting conditions may be better positioned to attract such developments. Serbia’s ability to compete for future installations will consequently depend on electricity supply and infrastructure alongside its technology-sector capabilities.
AI Data Centres Require Distinct Financing Structures
AI infrastructure combines elements of telecommunications, industrial electricity consumption and specialized technology investment, creating financing requirements that differ from those of conventional infrastructure projects. Buildings, electrical installations and cooling systems may remain operational for relatively long periods, while GPU equipment can require replacement considerably sooner. Financing structures must distinguish between durable physical assets and computing hardware exposed to rapid technological change.
Lenders may seek contracted customer revenue, sponsor equity, equipment-financing arrangements and conservative assumptions about utilization. Electricity-price exposure also requires attention.
A facility selling computing services under fixed-price contracts while purchasing electricity at variable market prices could experience margin volatility. Long-term electricity procurement and appropriate pricing mechanisms could help reduce that exposure. For Serbian banks and international financiers, Orion’s planned deployment could provide an early test of approaches to valuing, financing and monitoring AI computing infrastructure.
Electricity Prices Affect the Economics of 20 MW Capacity
The commercial performance of AI computing depends on hardware costs, customer demand, utilization, financing and electricity expenditure. Although power is not necessarily the largest operating cost for high-end AI systems, electricity prices can materially influence margins. For a hypothetical facility consuming 175 GWh annually, a €10/MWh change in the delivered electricity price would alter annual electricity expenditure by approximately €1.75 million.
The calculation illustrates the financial significance of power procurement at the proposed 20 MW scale. Actual expenditure would depend on consumption, network charges, balancing costs, taxes and contractual arrangements. For Serbian electricity suppliers, AI infrastructure could create demand for longer-term supply products that incorporate price stability and reliability. Predictable power costs could also help data-centre operators plan expenditure and support pricing commitments to computing customers.
