Skip to content
World clockEU--:--UK--:--USA--:--CN--:--PLDEFRIT中文EN

portal about AI and technologyevents · analysis · interviews · technical background

Search
LIVE
›

OpenAI's Grid Protectors puts 1,000 MW of flexible demand into a Georgia Power deal

OpenAI is testing a program called Grid Protectors at its planned Effingham County campus in Georgia, where a 3.2 GW agreement with Georgia Power includes 1,000 MW of flexible demand response, Tom's Hardware reported on 3 November.

TechnologyNewsRachel NwosuPublished: 4 November 20264 min readSources 28
OpenAI's Grid Protectors puts 1,000 MW of flexible demand into a Georgia Power deal

The company says the point is speed as much as load. A campus that agrees to cut consumption on request can connect sooner instead of waiting for new generation to come online.

Data centers are among the fastest-growing loads on several US grids. How much demand operators agree to switch off at peak decides whether utilities build more plants or spread fixed costs across wires that already exist. In PJM and in Georgia, that difference lands in capacity costs and, eventually, in bills. OpenAI says the arrangements will differ by location.

What OpenAI has committed to

The 3.2 GW agreement with Georgia Power includes 1,000 MW of flexible demand response.

The buffer is enough to power around 800,000 homes, nearly 20% of the state's. The company describes the aim as making campus flexibility a resource utilities can call on when communities need it most.

"Whether it's heating homes during a northern cold snap or running air conditioning through a southern heat wave, families need reliable electricity," OpenAI wrote in the blog post.

Demand response is not new. Grid operators once phoned steel mills and asked them to power down. Automation now lets a data center shed load in under a minute, according to Tom's Hardware. OpenAI says it will share further commitments as they are finalised.

Google's gigawatt, and Big Tech's hiring

Google's arrangement with Voltus shows the same idea at grid scale. The two announced a target of 1 GW of virtual power plant capacity in PJM, up from an initial 100 MW target set in June, Canary Media reported on 22 October.

Virtual power plants pay households and businesses to cut consumption or discharge batteries when the grid is tight. Voltus is securing commitments from its customers to reduce demand or supply battery power through 2040, far longer than PJM's three-year capacity auction horizon. The company manages more than 9 GW of demand response and virtual power plant capacity. It targets 20 GW by 2030, after raising $225 million earlier in October.

"The first 100 megawatts was always the first bite," Voltus chief executive Dana Guernsey told Canary Media. "We were always looking at how we bring this to scale, because scale is what we need".

Hiring points the same way. AWS named Robert Fourqurean, until recently a policy advisor for energy and climate at the British Embassy in Washington, as a manager of energy policy, DataCenterDynamics reported on 3 November. He had held the embassy post for more than three years. AWS has also agreed with American Electric Power to explore almost 960 MW of small modular reactor capacity in Indiana. Separately, it signed a 20-year power purchase agreement with Constellation Energy for 690 MW from Calvert Cliffs.

Why PJM is the pressure point

PJM Interconnection is the biggest US grid operator, serving 13 mid-Atlantic, Midwestern and Southern states.

The cost of securing enough power for peak demand in PJM has risen more than eightfold in the past few years. PJM forecasts data centers and other large loads will add 30 to 34 GW of demand by the early 2030s, on a system that recently peaked at 162.6 GW. Its interconnection queue is clogged, so new plants are slow to arrive.

Batteries behind existing connections

One fix needs no new transmission. FERC has told grid operators to allow batteries at existing power plants under Surplus Interconnection Service rules. A wind or solar farm can then inject stored power through a connection it already paid for.

Historical SIS requests total 44 GW as of mid-2026, about 80% of them proposals to add batteries. MISO has 14.8 GW proposed across 138 active requests and SPP has 14.3 GW across 89. PJM has been a laggard on rewriting its procedures, according to clean energy groups that have pushed for the change for years.

What happens next

That may be about to move. PJM's Market Implementation Committee is set to take up the next steps at its 4 November meeting, according to Jon Gordon, a senior director at the trade group Advanced Energy United. If a plan wins approval, PJM could ask FERC for permission to enact the changes as early as April 2027.

For OpenAI, the open question is the shape of the commitment. The terms of that flexibility, how fast the load can drop and for how long, are not public.

Comments 0

Sources

28
  1. 01OpenAI introduces program to reduce AI data center energy demand during peak hoursEN
  2. 02Google plans big virtual power plant in energy-strapped PJM gridEN
  3. 03AWS appoints British Embassy advisor as manager of energy policyEN
  4. 04PJM desperately needs grid batteries. Here's a way it could get some.EN
  5. 05China’s Mengxi grid tops 20 GW of storage as renewables approach 60% of capacityEN
  6. 06Grid Modernization and the Smart Grid - Department of Energy (.gov)EN
  7. 07Microgrids, Large Electric Loads & Grid Support: How to Leverage Microgrids to Support Utilities and Large Load Customers - Department of Energy (.gov)EN
  8. 08More of the Same Expensive Dependence on Fossil Fuels in TEP and APS 2026 Energy PlansEN
  9. 09DOE Announces $8.5M for Fusion Energy Public-Private PartnershipsEN
  10. 10Bringing Design Data Management into the Developer WorkflowEN
  11. 11On the pitch and on the grid, Ukraine is not waiting for the war to end and is asking the world to invest in itEN
  12. 12OpenAI deepens Abu Dhabi foothold as Sam Altman inks deals spanning energy, government and investmentEN
  13. 13Percepto launches Axion AI platform to unify inspection data analysisEN
  14. 14Electric Hydrofoils Save Energy, But Do They Earn Their Price?EN
  15. 155 ballot proposals that could drive or derail clean energyEN
  16. 16Eugene Becomes 5th Oregon City to Adopt Home Energy Score RequirementEN
  17. 17The Ford Escape may return as an EV based on the UEV platform [Update]EN
  18. 18Einstein's equations may explain the universe without dark energyEN
  19. 19How digital membership platforms help sports NGBs leave the spreadsheets behindEN
  20. 20Fintech platform Sageworks taken offline after suspected cyberattackEN
  21. 21Cloudflare’s Post-Quantum GeoKey Manager Enters Closed Beta without Attribute-Based EncryptionEN
  22. 22Can battery swapping and energy storage power CATL’s next phase of growth?EN
  23. 23Puerto Rico Energy Resilience Fund - Department of Energy (.gov)EN
  24. 24Solar Photovoltaic System Cost Benchmarks - Department of Energy (.gov)EN
  25. 25Solar Topics in Small Business Innovation Research and Small Business Technology Transfer (SBIR/STTR) - Department of Energy (.gov)EN
  26. 26SoftBank teams up with five Japanese firms to build quantum-based platformEN
  27. 27More heat pumps would actually be good for the grid, researchers findEN
  28. 28As US grid batteries surge, risk of fires plummetsEN

All figures and quotations in this text come from the sources listed below.

Content prepared by the editorial team with AI assistance.

Rachel Nwosu

Rachel Nwosu

AI, models and technology

Rachel Nwosu covers AI, models and technology for FLASH24, working from public model documentation, benchmark releases and repository histories rather than press summaries, and she skips announcements that arrive without reproducible numbers. She checks training-data claims against dataset cards and reruns reported metrics where code is available. She spends much of her week interviewing researchers and engineers, tracking model launch calendars, and comparing vendor benchmarks with independent evaluations. Outside the desk she runs 3D printers, restores old computers, and tests how models learn from internet junk. She does not publish benchmark figures she cannot trace to a source.

Newsroom →

Comments

0
  1. No comments yet — be the first.

Write a comment

Comments are public. We do not publish abuse, spam or advertising.