Net sustainability assessment of AI-driven data centers in the GCC: an MCDA approach framework

Overview

To fill the literature gap that examined the infrastructure footprint and distributed benefits of AI separately, the framework compares three 2024-2035 scenarios: BAU, moderate transition and aggressive decarbonisation. A three-tier boundary protocol separates the direct boundary (Scope 1+2), an extended boundary (partial Scope 3 embodied carbon) and an offset boundary (AI benefits that pass four attribution tests). MCDA dimension weights are carbon 0.40, water 0.25, social 0.20 and governance 0.15, derived from content analysis of GCC NDCs, checked against the literature (0.35-0.45 typical for carbon) and sense-checked by three experts.

The Net Impact Score is NIS = Σ(w_i·B_i) − Σ(w_i·F_i) on a −1 to +1 scale. Results: BAU −0.478 (ΣwF = 0.620, ΣwB = 0.142), moderate −0.032, aggressive +0.190. With all AI benefits excluded, every scenario is negative (BAU −0.619, moderate −0.198, aggressive −0.082). Net-positive outcomes require at least about 40% renewable procurement; renewable share is the most influential parameter, followed by the AI-benefit scaling factor. The ranking OPT > MOD > BAU holds under ±20% weight variation.

Scenario assumptions: BAU uses standard air cooling, 65% utilisation and a 4-year refresh; moderate uses hybrid air/liquid cooling, a 15% PUE reduction, 75% utilisation and a 3.5-year refresh; aggressive uses full liquid immersion, a 30% PUE reduction and 85% utilisation. GCC operational grid intensities (IEA 2023): Saudi Arabia 620, UAE 430, Qatar 520, Kuwait 650, Bahrain 580, Oman 490 gCO2/kWh, weighted average 548. Projected GCC data-centre emissions in 2035 are 47.3 Mt CO2e (BAU), 31.8 (moderate, −32.8%) and 19.4 (aggressive, −59.0%), up from 20.7 Mt in 2023. Even under the aggressive scenario, scaled AI benefits (14.7 Mt CO2e/yr) offset only 75.8% of data-centre emissions. Under BAU, 2035 water use is 2.50 million m³/yr: 1.49 direct cooling plus 1.01 indirect through desalination energy. Embodied carbon adds 15-25% to the operational footprint.

Limits: the outputs are scenario-conditional estimates built on secondary data, not forecasts; weights rest on expert judgement and NDC content analysis. The result is highly sensitive to renewable procurement share and AI-benefit scaling.

Key points

1. MCDA dimension weights: carbon 0.40, water 0.25, social 0.20, governance 0.15, derived from content analysis of GCC NDCs and checked against the literature (0.35-0.45 typical for carbon).
2. NIS = Σ(w_i·B_i) − Σ(w_i·F_i) on a −1 to +1 scale: BAU −0.478, moderate −0.032, aggressive +0.190; with all AI benefits excluded every scenario is negative (BAU −0.619, moderate −0.198, aggressive −0.082).
3. Net-positive outcomes require at least about 40% renewable procurement; renewable share is the most influential parameter; the ranking OPT > MOD > BAU holds under ±20% weight variation.
4. Scenarios: BAU standard air cooling, 65% utilisation, 4-year refresh; moderate hybrid air/liquid, 15% PUE reduction, 75% utilisation, 3.5-year refresh; aggressive full liquid immersion, 30% PUE reduction, 85% utilisation.
5. GCC operational grid intensities (IEA 2023): Saudi Arabia 620, UAE 430, Qatar 520, Kuwait 650, Bahrain 580, Oman 490 gCO2/kWh; weighted average 548.
6. Projected GCC data-centre emissions in 2035 are 47.3 Mt CO2e (BAU), 31.8 (moderate, −32.8%) and 19.4 (aggressive, −59.0%), from 20.7 Mt in 2023; even under aggressive, AI benefits (14.7 Mt/yr) offset only 75.8% of emissions.
7. Under BAU, 2035 water use is 2.50 million m³/yr (1.49 direct cooling + 1.01 indirect through desalination energy); moderate and aggressive cut the total by 28% and 52%.
8. In the extended boundary, embodied carbon adds 15-25% to the operational footprint.

Using it in design

• In AI data-centre investment assessment in hot, arid regions, weigh carbon, water, social and governance criteria together with explicit weights (0.40/0.25/0.20/0.15); indirect water (desalination energy) was about 0.68 times direct cooling water.
• Define the renewable procurement share (at least about 40%) as a project condition for a net-positive outcome, and put AI-benefit claims through the four attribution tests.
• Include embodied carbon (15-25% of the operational footprint) in the boundary definition, and report scenario results as conditional estimates, not forecasts.

Who should read it

Planners and policy specialists assessing the net sustainability of AI data-centre investments in the Gulf region.

Document contents

1. 1 Introduction
2. 2 Theoretical framework (2.1 Socio-technical systems theory, 2.2 Ecological modernization, 2.3 Conceptual tension: infrastructure)
3. 3 Literature review (3.1 Data center energy consumption, 3.2 Water-energy nexus in arid climates, 3.3 Applications of AI for environmental benefit, 3.4 Existing sustainability assessment frameworks, 3.5 Literature synthesis)
4. 4 Methodology (4.1 Research design, 4.2 Data sources and quality, 4.3 Case study selection protocol, 4.4 Sensitivity analysis protocol, 4.5 System boundary definition)
5. 5 Results (5.1 Environmental footprint: carbon, water, energy, embodied carbon; 5.2 AI-enabled sustainability benefits, 5.3 Net sustainability impact)
6. 6 Discussion (6.1 Interpretation of net Impact scores, 6.2 Theoretical implications, 6.3 Methodological contributions, 6.4 Policy and managerial implications, 6.5 Limitations)
7. 7 Conclusion

Related standards

• GHG Protocol Corporate Standard
• GHG Protocol Scope 2 Guidance
• EU Code of Conduct for Data Centres

Read the document

Publication details

Authors

Abdelrehim Awad; Bshair Alharthi; Hiyam Abdulrahim; Sara A. Ghorashi; Waleed M. Abdelfattah; Suleiman Ibrahim Mohammad; Asokan Vasudevan; Zahid Hussain

Publisher

Frontiers in Environmental Science (Frontiers Media SA)

Year

2026

Volume

14

Pages

1852462

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.3389/fenvs.2026.1852462

Licence

CC BY 4.0 — share with attribution

Licence evidence

doi.org

Citation