To honor a contracted GPU delivery with Microsoft, OpenAI, and Meta, which power sources should CoreWeave secure, in what order, and at what cost?
To honor a contracted GPU delivery with Microsoft, OpenAI, and Meta, which power sources should CoreWeave secure, in what order, and at what cost?
CoreWeave is a provider of GPU compute for AI workloads. The company leases data-center shells from third-party landlords, but procures its own power, signing power-purchase agreements (PPAs) directly with utilities and developers. To meet customer demand, CoreWeave must decide on a power generation technology(s): variable renewables paired with storage, firm zero-carbon generation, and natural gas. Five factors impact this decision: cost, speed-to-power, firmness, CO2 emissions, and technical feasibility. Lowest-cost generation usually wins, since power is a commodity, but AI buyers will pay more for speed and have publicly pledged to cut emissions. In this case, students create an optimal portfolio of power sources, weighing the advantages and disadvantages of each.
About the Author(s)
