Serbia’s proposed Đerdap 3 pumped-storage hydropower plant, estimated to require approximately €2.6bn, is increasingly being presented as more than an electricity-sector project. The rapid expansion of artificial intelligence infrastructure and energy-intensive data centres has introduced a new argument for the scheme: Serbia could use large-scale storage and grid flexibility to compete for technology investments that many European power systems can no longer accommodate easily. Euronews Serbia
That argument is commercially attractive, but it requires careful qualification. Đerdap 3 would not create new electricity. As a pumped-storage facility, it would consume power when prices and system demand are low, pump water into the upper reservoirs at Pesača and Brodica, and generate electricity when demand and market prices rise. Around 20–25 per cent of the electricity used in each storage cycle would be lost. Its value would therefore come from flexibility, not net generation.
This distinction matters because a new data-centre cluster needs considerably more than a large nominal power plant nearby. It requires dependable electricity production, high-capacity transmission connections, reserve capacity, predictable long-term prices, redundant grid infrastructure, water or alternative cooling solutions, digital connectivity and a credible route to low-carbon electricity procurement. Đerdap 3 could strengthen several of those conditions, but it would need to sit within a much larger investment architecture involving EPS, transmission system operator EMS, renewable-energy developers, technology investors and regional electricity markets.
The project’s strongest economic role would be to absorb electricity during low-price periods and return it during evening peaks, supply shortages and system disturbances. Serbia’s energy strategy envisages approximately 3.6GW of wind power and more than 7GW of solar capacity by 2040. A portfolio of that scale would create increasingly pronounced hourly imbalances. Solar production would compress midday prices, while winter evenings and low-renewable periods would continue to expose the system to expensive imports.
Đerdap 3 could convert part of that volatility into a revenue stream. During sunny or windy hours, it could buy or absorb electricity that might otherwise be curtailed or exported at weak prices. During evening peaks, winter shortages or regional supply disruptions, it could generate when electricity carries a significantly higher market value. The facility would therefore monetise the spread between charging and discharging prices while reducing renewable curtailment and limiting Serbia’s exposure to high-priced imports.
The commercial model cannot rest on energy arbitrage alone. A project with estimated capital expenditure of around €2.63bn will need several complementary revenue sources: wholesale-market trading, balancing energy, frequency regulation, reserve capacity, congestion management, black-start capability and potentially contracted availability payments. The economic value of each service will depend on the final design, usable storage volume, turbine configuration, cycling capacity, grid-connection arrangements and the market rules applying when the plant enters operation.
This creates an important financing question. Pumped-storage plants are long-lived strategic assets, but their revenues can be highly market-dependent. Banks will not treat forecasts of future price volatility as equivalent to contracted cash flow. To make Đerdap 3 financeable without placing the entire risk on Serbia’s public balance sheet, the government would need a credible revenue-stabilisation structure. This could involve a regulated component for system services, long-term capacity remuneration, availability contracts with EPS or EMS, market-based income and potentially dedicated supply or flexibility agreements with large industrial and digital consumers.
Data centres could become part of this contracted demand base, although they would not replace the need for a broader bankable structure. The existing State Data Centre in Kragujevac has a reported capacity of about 14MW. Continuous operation at full load would translate into annual electricity consumption of approximately 123GWh, broadly comparable with the demand of 25,000–30,000 households and around 0.35 per cent of Serbia’s annual electricity consumption.
That is material for a local network but small in relation to Đerdap 3. The Kragujevac facility would represent less than 0.6 per cent of the proposed pumped-storage plant’s maximum capacity. The project is therefore not economically justified by Serbia’s existing data-centre market. Its digital-investment logic depends on attracting a substantially larger class of facilities.
A 300MW data-centre campus, operating continuously, could consume about 2.6TWh annually. That would be more than 20 times the estimated consumption of the Kragujevac facility and equivalent to roughly 7.5 per cent of Serbia’s present annual electricity demand. A gigawatt-scale artificial-intelligence campus would alter the national load curve far more dramatically, requiring dedicated generation, multiple high-voltage connections and extensive reserve capacity.
International developments indicate where demand could be heading. Large artificial-intelligence campuses under development in the United States are moving from tens of megawatts towards hundreds of megawatts and, in some cases, several gigawatts. This has turned electricity availability from a secondary site-selection factor into one of the principal constraints on digital investment. Land, tax incentives and engineering talent remain relevant, but they cannot compensate for the absence of firm power and grid access.
Europe is already confronting that limitation. In Ireland, data centres have reached approximately 23 per cent of national electricity consumption, exceeding the combined demand of urban households. Grid congestion around Dublin has led to severe restrictions on new connections, while additional projects are increasingly expected to secure dedicated generation, storage and new renewable supply. The lesson for Serbia is not simply that data centres create an investment opportunity. It is that poorly sequenced digital expansion can overwhelm transmission capacity, raise system costs and transfer infrastructure expenditure to other consumers.
Serbia would consequently need to avoid treating spare nominal generation capacity as equivalent to available data-centre capacity. A pumped-storage facility can supply power only after electricity has first been stored. The country would still need sufficient primary generation to cover both ordinary demand and the new digital load. In practical terms, this means connecting Đerdap 3 with accelerated development of wind and solar projects, reinforcement of thermal and hydro reliability during the transition, expansion of the transmission network and potentially regional power-purchase arrangements.
The data-centre proposition would be strongest as an integrated package. A technology investor could contract electricity from a portfolio of new Serbian wind and solar plants, use Đerdap 3 and battery storage to manage hourly mismatches, and retain grid-backed reserve supply. Such a structure could offer more credible carbon accounting than a conventional annual renewable-energy certificate arrangement because production and consumption could be matched at a much more granular level.
For Serbia, this could also create a higher-value use for renewable electricity. Instead of exporting excess solar production at depressed midday prices, the system could store it and use it to support domestic digital infrastructure, industrial electrification or higher-priced evening exports. The gain would appear not only in power-market revenue but also in the value added retained through data processing, cloud services and artificial-intelligence infrastructure.
The opportunity should not obscure the project’s contractual and environmental exposure. Đerdap 3 is planned on a cross-border section of the Danube, making technical coordination with Romania indispensable. A recently established bilateral working structure may advance that process, but questions concerning water regimes, cross-border effects, operating rules and responsibility for downstream impacts will need formal resolution.
The proposed location also intersects with the Đerdap National Park, creating a demanding environmental and spatial-planning process. Pumped storage does not consume water in the conventional sense because water circulates between reservoirs, but this does not eliminate biodiversity, landscape, sediment, hydrological and construction risks. Upper reservoirs, tunnels, access roads, spoil-disposal areas and underground works can produce significant local impacts even where operational water consumption is limited.
The preparatory phase has begun to acquire financial weight. Serbia’s Ministry of Mining and Energy has launched procurement for planning and part of the technical documentation, with an estimated value of RSD625mn, or roughly €5.3mn. An earlier invitation for expressions of interest attracted six companies, while the project is intended to become the first major undertaking covered by the Serbia–United States intergovernmental energy cooperation framework.
US engineering group Bechtel has frequently been associated with the development, partly because it financed earlier feasibility work and already has a major regional presence through the Morava Corridor, constructed with Turkey’s Enka. It has also been working on hydropower studies for Albania’s Skavica project and has experience in programme management for complex infrastructure.
Bechtel’s cooperation with Nvidia on modular designs for gigawatt-scale artificial-intelligence data centres adds an unusual strategic dimension. A contractor familiar with both large energy infrastructure and the physical requirements of AI campuses could help Serbia integrate power planning with technology-sector investment. That relationship, however, does not determine the project’s final economics or remove the need for competitive procurement, independent review and strong protection of the state’s commercial position.
An engineering, procurement and construction structure could place responsibility for design, equipment procurement and construction under a single contractor or consortium. Civil works would probably include reservoirs, tunnels, waterways and an underground or partially underground powerhouse, while specialised manufacturers such as Voith, Andritz, GE Vernova or Toshiba could compete to supply reversible pump-turbines, generators and control systems.
The attraction of single-point EPC responsibility is clear, but so is the risk. Underground construction carries major exposure to geological uncertainty, design changes and claims. A poorly developed scope transferred too early into an EPC contract can produce expensive variations even when the headline arrangement is described as fixed-price or turnkey. The growth in the cost of the Morava Corridor is a warning that contractor reputation cannot substitute for complete technical preparation and transparent cost governance.
Before committing to the main construction package, Serbia will need a defensible reference design, updated geological investigations, hydraulic modelling, an environmental and social impact assessment, transmission studies, a procurement strategy, a financial model and clearly allocated cross-border risks. Independent Owner’s Engineer oversight would be particularly important because the state must retain enough technical capacity to challenge designs, certify progress, assess claims and verify commissioning performance.
The Owner’s Engineer function should extend beyond construction supervision. It would need to integrate interface management between civil works and electromechanical suppliers, programme controls, cost forecasting, risk registers, environmental compliance, grid-code testing, reservoir commissioning and verification of the plant’s actual efficiency and response characteristics. For a €2.63bn project, small weaknesses in early-stage design or contract allocation can translate into hundreds of millions of euros in additional public exposure.
The revenue model also needs to recognise the separation between national-system value and commercial market income. Đerdap 3 could lower balancing costs, reduce curtailment, free existing hydropower plants from part of their reserve burden and improve Serbia’s ability to manage winter import exposure. Some of these benefits accrue to EPS, some to EMS, some to renewable developers and some to the wider economy. Unless those benefits are translated into contractual revenue, they do not automatically service project debt.
A plausible structure would therefore combine public strategic ownership with disciplined commercial contracting. Large data centres and industrial consumers could provide long-term demand commitments. Renewable projects could supply low-cost charging energy. EMS could procure defined ancillary services. EPS could operate the facility or manage its market exposure, while financing could combine sovereign support, export-credit participation, international financial institutions and long-tenor infrastructure debt.
The principal threat is that the data-centre narrative becomes a convenient justification for accelerating construction before Serbia has resolved the project’s engineering, market and environmental fundamentals. Artificial-intelligence demand is expanding rapidly, but individual technology projects can change location, design and energy strategy far faster than a pumped-storage plant can be permitted and built. Đerdap 3 must remain viable under scenarios in which no hyperscale campus is secured.
Its investment case is strongest when viewed as national and regional flexibility infrastructure capable of serving multiple markets over several decades. Data centres could improve the demand outlook, anchor renewable-energy procurement and create an additional source of contracted revenue. They cannot compensate for uncontrolled construction costs, weak project preparation or an incomplete framework for monetising system services.
Properly structured, Đerdap 3 would allow Serbia to sell something more valuable than electricity alone: the ability to shift energy across time, stabilise a renewable-heavy grid and support consumers that cannot tolerate interruptions. That combination could reinforce Serbia’s role in the Southeast European power market and make the country more credible as a location for capital-intensive digital infrastructure. The decisive value will lie in the quality of project preparation, contractual discipline and the integration of storage, generation and transmission—not in the scale of the headline ambition.
