Energy implications of AI inference workload growth in renewable hydrogen-powered data centers: A data-driven hourly assessment
INTERNATIONAL JOURNAL OF HYDROGEN ENERGY, cilt.288, ss.1-14, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 288
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.ijhydene.2026.158332
- Dergi Adı: INTERNATIONAL JOURNAL OF HYDROGEN ENERGY
- Derginin Tarandığı İndeksler: Academic Search Ultimate (EBSCO), Engineering Source (EBSCO), Scopus, Science Citation Index Expanded (SCI-EXPANDED), Artic & Antarctic Regions, Chemical Abstracts Core, Chimica, Compendex, Environment Index, INSPEC
- Sayfa Sayıları: ss.1-14
- Bursa Uludağ Üniversitesi Adresli: Evet
Özet
The rapid expansion of artificial intelligence (AI) is increasing data center electricity demand and challenging the integration of renewable energy. This study assesses the effects of increasing AI inference workloads on a renewable hydrogen-based data center using a data-driven hourly framework. A UK Power Networks utilization profile and an AI inference profile derived from measured Llama 3 70B power data on NVIDIA H100 GPUs are combined with hourly meteorological data for Ankara, Türkiye. The system integrates photovoltaic (PV) and wind generation, proton exchange membrane (PEM) electrolysis, hydrogen storage, and PEM fuel cells. Four scenarios use normalized demand and fixed renewable generation capacities, with additional mean AI-related demand ranging from 0% to 50% of the baseline mean demand. From the baseline to the highest-demand scenario, direct renewable coverage decreases from 55.08% to 47.91%, while renewable electricity surplus and potential hydrogen production both decline by 37.35%. Per MW of mean baseline demand, annual potential hydrogen production decreases from 71.74 to 44.94 t H2 and fuel-cell generation from 1.43 to 0.90 GWh, whereas unmet electricity demand increases from 2.51 to 5.97 GWh. Consequently, the load supply ratio (LSR) decreases from 71.42% to 54.73%. A sensitivity analysis of the PV-wind mix, electrolyzer-specific electricity consumption, fuel-cell efficiency, and hydrogen storage capacity indicates that the LSR decline persists across the investigated ranges of these parameters. The normalized framework supports scalable assessment of these interactions within the adopted model assumptions.