Optimal Workload Scheduling and Energy Management of AI Data Centers with Demand Response

Lehigh Professors Shalinee Kishore, Alberto J. Lamadrid, Javad Khazaei, and Ph.D. Candidate Morteza Ghorashi have devised a mixed-integer linear programming (MILP) framework designed to reduce operation costs in data centers by 20% through effective demand response strategies.

Data centers are among the largest energy consumers, heavily relying on electricity and water-intensive cooling techniques, putting pressure on local infrastructure. Their operations allow for flexibility; thus, they can engage with the power grid through Demand Response (DR). This interaction enables data centers to shift energy use to off-peak hours, yielding cost savings and enhancing grid reliability.

The framework developed by the research team optimally schedules workloads while managing energy sources—grid power, local renewables like solar, and battery storage. Key components of their model include dynamic pricing from the main grid, local solar microgrids for sustainable self-generation, energy storage systems, and the various demands of servers and cooling units.

A practical application of this framework was illustrated through a case study involving a 24-hour operational schedule across two grid-connected data centers. The study demonstrated that utilizing DR can lead to strategic energy consumption, as tasks were scheduled during cheaper off-peak hours and incorporated local solar and battery resources. This approach not only minimized reliance on expensive peak-hour grid power but also maintained efficiency and stability in operations.

Future efforts will expand on this work, incorporating detailed cooling models and analyzing economic factors alongside external grid conditions. This research outlines a pathway for data centers to evolve into proactive participants in energy management, ultimately reducing costs and contributing to grid stability. Support for this research was provided by a grant from GTI Energy and was presented at the Innovating Energy and Water Solutions for Tomorrow's AI Data Centers Symposium in October 2025.

Doubling energy investment in Africa necessitates urgent measures to reduce financing costs and improve capital accessibility. The International Energy Agency (IEA) emphasizes that approximately 600 million people across Africa lack electricity, with recent advancements in electrification slowing due to factors like rising debt and the impacts of the COVID-19 pandemic. While off-grid solutions, particularly solar and battery systems, rapidly expand—accounting for over half of new electricity connections in 2022—the efficiency of these solutions increasingly hinges on intelligent planning and reliable data.

To enhance electricity access, the IEA, with collaboration from institutions like MIT, is developing an open-source, AI-driven model aimed at mapping electricity demand across Africa. This innovative model integrates satellite imagery, building footprint data, and utility meters to accurately assess which structures have electric access and their energy needs. Using machine learning, it successfully recognizes patterns in energy consumption, achieving over 80% accuracy in identifying electrified buildings and reducing demand estimation errors by 40% compared to traditional methods.

The implications of this AI mapping tool are substantial. It empowers utilities and off-grid service providers to efficiently locate high-potential customers and underserved areas without extensive on-ground surveys, significantly cutting customer acquisition costs. This precision aids in optimal planning for grid expansions, mini-grids, and standalone systems. Moreover, the model helps target informal settlements and urban areas on the brink of electrification, where connection expenses may be lower, and the ability to pay is anticipated to rise.

Ultimately, this technological advancement transforms the landscape of energy accessibility in Africa, making it easier and cheaper to deliver power solutions to communities in need, thus driving sustainable development in the region.

John Ketchum, CEO of NextEra Energy, emphasizes that the partnership with Google represents a pivotal moment where energy and technology are becoming deeply connected. Google Cloud and NextEra Energy are set to establish gigawatt-scale data centers with dedicated power, specifically aimed at enhancing enterprise AI adoption, energy efficiency, and creating resilient infrastructure.

As the demand for energy-efficient data centers escalates, this collaboration highlights several concrete applications of AI to transform the energy sector. The companies plan to develop new data centers powered directly by dedicated power plants, leveraging Google Cloud’s AI capabilities to accelerate NextEra Energy’s deployment of innovative, AI-driven solutions. The first product of this collaboration is slated for release by mid-2026 on the Google Cloud Marketplace.

By integrating Google’s generative AI with NextEra's asset data, the partnership aims to reduce operational costs and improve worker safety. The AI system will proactively address potential disruptions in supply chains and weather-related challenges, enhancing overall operational resilience. Furthermore, incorporating Google’s open-source forecasting models, such as TimesFM 2.5 and WeatherNext 2, will fortify grid resilience against extreme weather.

This strategic alliance not only promises to significantly lower NextEra Energy’s operational costs but also ensures a stable energy supply. As large-scale AI deployment drives the need for data center capacity, this collaboration positions both companies to reshape energy infrastructure, creating a forward-looking model that integrates technological innovation with energy expertise. The revitalization of the Duane Arnold Energy Centre in Iowa for nuclear energy further exemplifies their commitment to this transformative vision.

Recent advancements in government and industry suggest that fusion energy could significantly impact national power grids by the 2030s, evolving from experimental research to real-world application. A notable collaboration between Google DeepMind and Commonwealth Fusion Systems (CFS) exemplifies how artificial intelligence (AI) can address complex challenges in fusion physics. As fusion emerges as a strategic technology with the potential to generate trillions in clean energy, the International Atomic Energy Agency (IAEA) emphasizes its growing importance.

The U.S. Department of Energy has outlined a roadmap to integrate fusion technology into the energy landscape by the early 2030s. AI is expected to facilitate breakthroughs in areas such as materials science and digital modeling, enabling the efficient control of plasma in magnetic confinement systems. This is crucial for achieving net-energy gain from fusion, a goal that has been pursued since the 1950s.

The partnership between DeepMind and CFS focuses on leveraging AI capabilities like deep reinforcement learning to maintain plasma stability, an essential factor for power generation. Tools like the TORAX plasma simulator will help explore operational efficiencies and optimize fusion processes. Supported by significant investments from major tech firms like Google and Eni, CFS aims to deploy its first grid-scale fusion plant in the 2030s.

Moreover, with 160 fusion facilities either operational or in development globally, the IAEA projects an increase in fusion-generated energy from 2 TWh in 2035 to 375 TWh by 2050. This growth could massively boost global GDP as demand for clean electricity escalates. As a result, fusion energy is transitioning from scientific research to a pivotal economic sector, demonstrating AI's significant role in shaping the future of energy.