An AI Data Centre in Georgia – How Much Electricity and Water Would It Need?
Debates about AI infrastructure often begin with chips, investment, and digital sovereignty. Yet a data centre is a physical facility before it is a digital one. It consumes electricity continuously, removes heat continuously, and depending on cooling design may consume substantial water. Georgia must therefore ask not only whether it can attract an AI facility, but whether the chosen grid node can support it in every season and what local water source would carry the cooling burden.
The model below assumes 10 MW of IT load: the servers and networking equipment rather than total facility demand. It assumes continuous full-load operation for 8,760 hours. Actual utilisation may vary, but grid infrastructure still has to accommodate credible peak demand.
Energy is becoming AI’s binding constraint
The International Energy Agency’s 2026 update projects global data-centre electricity use rising from about 485 TWh in 2025 to roughly 950 TWh in 2030. Electricity use by AI-focused facilities grows much faster, approximately tripling over the period. The challenge is not only annual energy. Data-centre demand is geographically concentrated, persistent, and increasingly subject to rapid power swings, creating requirements for substations, transformers, redundant feeds, and batteries.
Georgia consumed about 14.9 TWh of electricity in 2025. Hydropower gives the country a strong renewable profile, but summer hydropower availability is not the same product as dependable winter supply. An AI facility needs hourly deliverability, grid capacity, redundancy, and a predictable long-term commercial framework not merely an attractive annual energy balance.
| Indicator | Value | Status / meaning |
|---|---|---|
| Global data-centre use, 2025 | ≈485 TWh | IEA updated estimate |
| Global data-centre use, 2030 | ≈950 TWh | IEA central projection |
| AI-focused facility growth, 2025–2030 | ≈3× | IEA central projection |
| Georgia electricity consumption, 2025 | ≈14.9 TWh | Annual power-system data |
| Modeled IT capacity | 10 MW | BTUAI illustrative assumption |
Electricity: 87.6 GWh before cooling
According to BTU researchers – A continuous 10 MW IT load consumes 87.6 GWh a year: 10 MW multiplied by 8,760 hours. Cooling, uninterruptible power, distribution losses, lighting, and other infrastructure raise facility demand. Power Usage Effectiveness, or PUE, is total facility energy divided by IT energy. A lower number is better, with 1.0 representing the theoretical floor.
The model uses PUE 1.15 for an energy-efficient evaporative configuration, 1.20 for hybrid cooling, and 1.30 for a lower-water dry configuration. These are transparent design assumptions, not vendor guarantees. Actual performance depends on climate, load, engineering, and operations.
| Cooling scenario | PUE | IT energy | Total annual energy | Average load | Share of Georgia 2025 use |
|---|---|---|---|---|---|
| Evaporative / energy-efficient | 1.15 | 87.6 GWh | 100.74 GWh | 11.5 MW | 0.68% |
| Hybrid | 1.20 | 87.6 GWh | 105.12 GWh | 12.0 MW | 0.71% |
| Dry / lower-water | 1.30 | 87.6 GWh | 113.88 GWh | 13.0 MW | 0.76% |
Formula: annual facility energy = 10 MW × 8,760 × PUE. The result is nationally manageable around 0.7% of Georgia’s 2025 consumption but 11.5–13 MW of continuous new demand can still be material for a specific substation and region.
Water: the same computing capacity, a fourteen-fold range
Water Usage Effectiveness, or WUE, divides annual onsite water consumption in litres by IT energy in kWh. A U.S. Department of Energy example of an efficient hybrid facility reports WUE of 0.7 L/kWh, while another DOE practical example shows approximately 1.42 L/kWh for evaporative cooling. Our rounded scenarios use 0.7 and 1.4; the 0.1 dry-cooling case is a modeling assumption allowing residual water for humidification and servicing.
| Cooling scenario | WUE | Annual onsite water | Daily average | Energy trade-off |
|---|---|---|---|---|
| Dry / lower-water | 0.1 L/kWh | 8,760 m³ | 24 m³ | Lowest water demand; modeled PUE rises to 1.30 |
| Hybrid | 0.7 L/kWh | 61,320 m³ | 168 m³ | Middle-ground water-energy balance |
| Evaporative | 1.4 L/kWh | 122,640 m³ | 336 m³ | Highest water demand; modeled PUE falls to 1.15 |
Formula: annual water = 87.6 million kWh of IT energy × WUE. The result ranges from 8,760 to 122,640 m³ a year. These figures describe onsite consumption, not merely water circulated or withdrawn. They exclude the indirect water footprint of electricity generation, which should be disclosed separately to avoid methodological double counting.
Georgia’s opportunity and its constraint
According to BTU researchers – Georgia can offer a renewable-heavy grid, locations with useful cool-season conditions, regional fibre connectivity, and room for careful site selection. More free-cooling hours can reduce both electricity and water demand. But the statement that Georgia is “water-rich” is not a project assessment. Water is uneven across regions and seasons, while local basins may already serve households, agriculture, ecosystems, and industry.
The grid constraint is equally local. A 0.7% national share may look small, yet the facility needs power when hydropower is low and the system relies more heavily on imports or thermal generation. Site selection therefore requires an interconnection study, an hourly seasonal model, two independent feeds, storage or other backup, and a clear decision on who pays for network reinforcement.
The conditions Georgia should require
A data centre can be a valuable Georgian investment if it leaves more than an electricity bill: skilled operations, cybersecurity capability, local cloud services, university research links, and financing for additional generation or grid capacity. Otherwise, the country risks supplying scarce resources to an asset with a limited employment footprint.
Permits and investment agreements should disclose IT and maximum facility MW, firm and flexible load, annual and monthly PUE, WUE and water source, drought operating mode, backup fuel and emissions, network-upgrade cost allocation, and hourly electricity sourcing. An annual renewable certificate alone does not demonstrate clean power in every operating hour.
| Required disclosure | Core metric | Risk controlled |
|---|---|---|
| Electrical design | IT MW, facility MW, annual MWh, ramping | Local overload and weak interconnection |
| Energy efficiency | Annual and monthly PUE | Hidden cooling overhead |
| Water balance | WUE, m³/year, source, drought mode | Competition for local water |
| Power sourcing | Hourly profile and PPA | Seasonal shortage and misleading emissions claims |
| System contribution | Substation, line, battery, demand response | Shifting infrastructure cost to other users |
Conclusion
A 10 MW AI data centre is not a small digital office. In our model, it requires 100.7–113.9 GWh of electricity a year, equal to roughly 0.68–0.76% of Georgia’s 2025 use. Onsite water consumption ranges from 8,760 to 122,640 m³. The ranges are the point: neither question has a single answer without a technical design.
According to BTU researchers – Georgia’s strategy should be to set resource conditions before approving the investment. Dry cooling can save water but use more electricity; evaporative cooling can save energy while consuming more water. A strong project prices this trade-off against the actual local climate, grid, and basin, finances the infrastructure it requires, and leaves measurable digital capability in the country.
Sources and methodological note
Principal sources: International Energy Agency, Energy and AI and Key Questions on Energy and AI; U.S. Department of Energy guidance on PUE, WUE, and cooling efficiency; Georgian State Electrosystem annual balance and system data. The 10 MW IT load, full-load operation, PUE 1.15/1.20/1.30, and WUE 0.1/0.7/1.4 are scenario assumptions. Results are not a project engineering opinion, vendor guarantee, or national forecast.