Cover: AI-generated conceptual illustration.
OpenAI's Project Camellia power agreement and Google Cloud's utility deployments show different ways of approaching the electricity needs of AI infrastructure. One concerns how a planned compute campus will adjust its demand; the others concern how utilities forecast and manage their systems. The evidence spans contractual commitments, reported savings and targets still awaiting operational validation.
Key takeaways
- Georgia Power says OpenAI committed up to 1,000 MW of flexible demand response under a 25-year agreement for Project Camellia.
- NextEra reports customer savings from AI-assisted dispatch and outage scheduling. Santee Cooper's projected 75% reduction in financial scenario runtimes concerns a separate application from its weather model.
- Google selected 28 energy-focused startups for its 2026 accelerator programs. Their advertised benefits need to be assessed individually.
OpenAI: a power contract and an energy-technology role
Georgia Power's July announcement says the utility can reduce electricity delivered to the planned campus at certain times under the agreement. The flexible-demand commitment concerns reducing consumption during high demand; it does not describe a new 1 GW generating plant.
OpenAI's project announcement places Camellia in Effingham County, Georgia, with 3.2 GW of contracted power scheduled to arrive in phases between 2028 and 2032. These are future delivery plans, not operating capacity today.
Separately, OpenAI's Clean Energy and New Technology Lead listing describes responsibility for evaluating energy technologies and turning promising options into deployable projects. Its scope includes storage, flexible demand and low-carbon backup. The listing establishes the role's intended responsibilities; it does not establish that a hire has been completed or that any particular technology has been installed.
NextEra: reported savings from dispatch optimization
At a September 2 Google Cloud media roundtable, NextEra Analytics president Rich Argentieri attributed more than $20 million in customer savings during 2026 to optimized dispatch and outage scheduling. He said the first version took under 12 weeks to build and described processing roughly half a trillion data points daily. These are company-reported results, as recorded in POWER's coverage.
Google Cloud's account separately describes NextEra's Optos suite, developed with Gemini Enterprise Agent Platform. Its Optos Composer application brings decisions about generation, fuel, maintenance, trading, reserves and storage together. The reported utility savings should not be read as measured savings for an AI data center using that software.
Santee Cooper: weather forecasting and financial planning
Santee Cooper's September 3 announcement describes two distinct efforts:
| Application | Intended use | Evidence and limits |
|---|---|---|
| WeatherNext 2 | Improve weather-dependent load forecasts | The announcement describes hundreds of possible weather scenarios generated in under a minute, with forecasts reaching 15 days ahead. |
| Gemini-based financial model | Speed financial scenario analysis | The utility expects runtime reductions of up to 75%; this is a target for financial planning, not a weather-forecast speed measurement. |
The distinction matters. CFO Tami Wilson said extreme winter temperature variance can expose the utility to spot-market costs of up to $100,000 an hour, according to the announcement. That illustrates the risk being addressed, rather than savings already achieved. The financial tool is intended to support decisions including a $10 billion grid-expansion budget.
POWER's reporting says the financial application had not yet been tested in production at the time of the roundtable. Its expected improvement should therefore remain separate from NextEra's reported results. Santee Cooper supplies electricity ultimately serving more than two million people, according to its announcement; that is not a count of direct customer accounts.
Google's accelerator: candidates for grid and campus problems
Google's August 26 cohort announcement lists 12 North American and 16 European startups receiving mentorship and access to its tools.
Google describes Camus Energy as helping utilities connect data centers with local energy sources three to five years faster. It lists Zendo Energy for optimizing data-center energy use, Kyra for forecasting grid capacity and SMPNet for network optimization. Those are descriptions of the companies' offerings, not a common performance benchmark or evidence that every campus can realize the same benefit.
What compute operators can take from this
Our editorial reading is that these examples call for different verification questions: what demand reduction a campus is contractually required to deliver, what a utility has measured in service, and what a software vendor has only projected. Comparing those categories carefully is more useful than treating them as a single proven design for AI power.
For the physical equipment side, see our NVIDIA GB300 power and cooling explainer. For another announced power architecture, read our Prometheus Hyperscale Texas campus coverage.
Sources
This article was researched and fact-checked against the following sources:
- NextEra, Santee Cooper Point to Real Dollar Savings From AI Deployments (finance.yahoo.com)
- Google Selects 28 AI Energy Startups for Grid and Data-Center Power Accelerator | Superpower Daily (superpowerdaily.com)
- NextEra, Santee Cooper Point to Real Dollar Savings From AI Deployments (powermag.com)
- OpenAI Integrates Energy Planning Within Data Center Operations: A Strategic Shift - DediRock (dedirock.com)
- Santee Cooper Partners with Google Cloud to Help Improve Grid Reliability Through Better Planning (prnewswire.com)
- Georgia Power to serve OpenAI project in Effingham County (georgiapower.com)
- Building AI infrastructure with the Effingham County community (openai.com)
- Clean Energy and New Technology Lead — OpenAI (jobs.ashbyhq.com)
- How utilities are using AI to generate the future (cloud.google.com)
- 28 startups using AI to transform the energy sector (blog.google)
