Sustainable Grid through Distributed Data Centers: Spinning AI Demand for Grid Stabilization and Optimization

TL;DR

This work tackles grid reliability and renewable integration challenges posed by rising HPC/AI workloads by proposing a grid-centric paradigm that actively places and schedules batchable HPC/AI tasks across distributed, grid-aware data centers to create spinning demand totaling $TWhrs$ of parallel compute workloads. It leverages massively parallel HPC Monte-Carlo workloads and Learn While Mining as energy-aware, pause-resilient tasks coordinated by a pooling agent to balance supply and demand, maximize renewable utilization, and stabilize grid frequency. Using a three-node simulation with solar, wind, and gas, the authors show that distributing HPC load reduces renewable curtailment, lowers gas usage, and lowers electricity costs (e.g., approximately 12% when HPC distribution increases) while enabling ancillary services such as regulation based on AGC signals. The work also analyzes market designs (merit order and piecewise-linear supply functions) for pricing HPC capacity and outlines future steps toward voltage/frequency regulation and two-sided market clearing. Overall, the framework offers a path to decarbonize AI workloads and defer expensive grid expansion by aligning compute demand with renewable energy and grid needs.

Abstract

We propose a disruptive paradigm to actively place and schedule TWhrs of parallel AI jobs strategically on the grid, at distributed, grid-aware high performance compute data centers (HPC) capable of using their massive power and energy load to stabilize the grid while reducing grid build-out requirements, maximizing use of renewable energy, and reducing Green House Gas (GHG) emissions. Our approach will enable the creation of new, value adding markets for spinning compute demand, providing market based incentives that will drive the joint optimization of energy and learning.

Figures (9)

• Figure 1: Vision for Sustainable Grid through distributed data centers: spinning demand of batch-able compute jobs distributed to stabilize the grid.
• Figure 2: Energy benefit of joint Monte Carlo and proof of work mining
• Figure 3: Potential utilization of HPC in US electric power markets
• Figure 4: (Left) Example simulation HPC configurations. (Right) Energy generation, curtailment, and usage time series in simulation cases a and b.
• Figure 5: Cost of energy (COE), renewable curtailment, and gas power utilization statistics with increasingly distributed HPC.