Santee Cooper backs a $10 billion grid plan with Google AI
A state utility is buying forecast accuracy as a hedge against $100,000-an-hour winter spot-market misses.
Santee Cooper, South Carolina's state-owned electric and water utility, is moving load forecasts and financial scenario planning onto Google Cloud's AI models, Data Center Dynamics reports, deploying WeatherNext 2 — Google's ensemble weather modeling system — and a financial-forecasting application built on Gemini Enterprise. The agreement carries no disclosed price.
WeatherNext 2 generates hundreds of ensemble scenarios in under a minute and returns probabilistic forecasts up to 15 days out, sharpening the hourly purchase decision that CFO Tami Wilson describes in dollar terms: overestimate demand and Santee Cooper spends on fuel and operations it does not need; underestimate and it is forced into the spot market, often at high prices. She puts that exposure at up to $100,000 per hour for a single degree of temperature variance on the coldest winter day.
The Gemini application attacks the slow end of the cycle, cutting financial scenario-building time by as much as 75 percent and compressing what once took weeks or months into days or hours. Wilson ties that urgency to what she calls South Carolina's rapid growth and its shift toward innovative energy sources, and the scenario output feeds planning for the utility's $10 billion grid expansion budget.
Google Cloud's message is that existing utility infrastructure is the constraint: Raiford Smith, the vendor's global director for power and energy, said utilities are drowning in real-time data that legacy systems were not built to process, and he pointed to Santee Cooper as a demonstration of Gemini Enterprise augmenting that infrastructure. The utility plans to extend Gemini Enterprise beyond finance so staff in other departments can query internal systems, with a governance structure built around the rollout to protect both organizational and customer data.
The partnership itself carries no financing structure, no ownership change, and no disclosed price; what it carries is a materiality claim that moves forecast error from an operational nuisance to the risk line of a grid budget. At $100,000 an hour for one degree of temperature variance on a winter peak, that claim has arithmetic behind it, enough to justify handing a $10 billion program's planning runbook to an AI model. Read that way, the rollout looks less like a technology purchase and more like a risk-management contract on grid expansion.
The Wrap