NVIDIA and OpenAI announced a major strategic partnership on September 22, 2025, to deploy at least 10 gigawatts (GW) of NVIDIA-powered computing systems. This massive scale involves millions of GPUs and is aimed at training and running OpenAI’s next-generation AI models, including pursuits toward superintelligence. NVIDIA will invest up to $100 billion in OpenAI, tied to deployment milestones, with the first 1 GW expected online in the second half of 2026 using NVIDIA’s Vera Rubin platform. The collaboration also includes joint optimization of hardware, software, and infrastructure roadmaps, positioning NVIDIA as OpenAI’s preferred partner for expanding AI factories.
The partnership’s 10 GW compute capacity represents the power draw. Assuming continuous full utilization (24 hours a day), this equates to approximately 240 gigawatt-hours (GWh) of energy consumption per day. For context, that’s roughly equivalent to the daily electricity use of over 18 million average U.S. households. Note that this is for the planned full deployment starting in 2026; current OpenAI operations (like training GPT-4) used far less, around 1,300 megawatt-hours (MWh) total for that model. No specific environmental mitigation details were mentioned in the announcement.
Microsoft and OpenAI’s partnership, primarily through Azure cloud infrastructure, has driven massive energy demands for AI training and inference. To meet these, Microsoft has secured several high-profile power purchase agreements (PPAs) and investments in nuclear and renewable sources. While not all deals explicitly name OpenAI, they directly support Azure data centers hosting OpenAI’s models (e.g., GPT series). These efforts align with Microsoft’s goal of carbon-negative operations by 2030. Below are the key deals as of October 2025.
Microsoft is betting heavily on nuclear power for reliable, low-carbon baseload energy to fuel AI growth, with several projects tied to data center expansion for partners like OpenAI.
Helion Energy Fusion PPA (2023, Construction Started July 2025): Microsoft signed a landmark PPA with Helion Energy for electricity from the world’s first commercial fusion power plant, Orion, located in Malaga, Washington. The deal commits to up to 50 MW of fusion-generated power starting in 2028, after a one-year ramp-up, delivered via Washington’s grid to nearby Microsoft data centers. Helion, valued at $5.4 billion and backed by OpenAI CEO Sam Altman, raised $425 million in a Series F round to fund the project. This fusion tech aims to produce net-positive energy, complementing Microsoft’s fission-based nuclear commitments.
Three Mile Island Unit 1 Restart (Announced September 2024, Online 2028): Microsoft inked a 20-year PPA with Constellation Energy to revive the shuttered Three Mile Island nuclear reactor in Pennsylvania, renaming it the Crane Clean Energy Center. The ~835 MW facility will supply dedicated power to Microsoft’s AI data centers, addressing surging demand from workloads like OpenAI’s. Restart costs are estimated at $1.6 billion, with the plant expected to operate for decades, creating jobs and supporting grid stability.
These nuclear initiatives are part of a broader trend where Microsoft and peers (e.g., Amazon, Google) are investing billions in nuclear to power AI, with projections of $250 billion in AI infrastructure spend in 2025 alone, including Microsoft’s $80 billion allocation.
Microsoft has expanded its portfolio of renewable PPAs to ensure 100% clean energy matching for data centers by 2025, indirectly powering OpenAI via Azure.
| Deal | Partner(s) | Value & Details | Timeline & Capacity |
|---|---|---|---|
| Shizen Energy Solar PPAs | Shizen Energy (Japan) | Three 20-year agreements for solar power; supports Microsoft’s 2025 renewable targets and 2030 carbon-negative goal. | Ongoing; data through August 2025, offsite PPAs only. |
| Nscale & Aker Renewable AI Infrastructure | Nscale (AI hyperscaler), Aker (industrial investor) | $6 billion+ for 100% renewable-powered AI data centers, including wind and solar projects. | Announced September 18, 2025; multi-year rollout. |
| Broader Clean Energy Commitment | Various (wind, solar projects) | $10 billion total for renewables to offset AI growth; part of Azure OpenAI Service sustainability push. | 2025 onward; aligns with global energy transition efforts. |
These deals address the enormous power needs of OpenAI’s operations on Azure—e.g., training large models can consume gigawatt-hours equivalent to small cities. However, challenges include permitting delays, fusion tech risks, and grid bottlenecks, with AI data centers projected to strain U.S. power supplies through 2030. Microsoft views these as long-term solutions, blending nuclear reliability with renewable scalability. For more on OpenAI’s separate Stargate project (with Oracle/SoftBank), see related announcements, though it doesn’t overlap directly with Microsoft energy pacts.
Google has ramped up investments in clean energy to meet the surging power demands of its AI operations, including training models like Gemini and expanding data centers. In 2025, the company plans to spend $75 billion on AI infrastructure alone, with a focus on nuclear for reliable baseload power and renewables for scalability. Key goals include achieving 24/7 carbon-free energy matching by 2030 and securing over 8 GW of new clean energy contracts in 2024. These deals support Google’s data centers, which consumed enough electricity in 2024 to power 1.5 million U.S. homes, with AI workloads driving a 48% year-over-year increase in emissions.
Google is leading Big Tech’s nuclear revival, signing multiple agreements for advanced small modular reactors (SMRs) and next-gen fission to provide dedicated, low-carbon power for AI facilities.
Kairos Power and TVA Agreement (August 2025): Google, Kairos Power, and the Tennessee Valley Authority (TVA) inked a deal for up to 500 MW of advanced nuclear power by 2035, starting with the Hermes 2 demonstration plant in Oak Ridge, Tennessee. Initial output ramps from 28 MW to 50 MW starting in 2030, feeding into TVA’s grid to supply Google’s data centers in Montgomery County, Tennessee, and Jackson County, Alabama. This marks the first utility purchase of power from a Generation IV reactor, emphasizing clean attributes for AI-driven loads.
Elementl Power Development Agreement (May 2025): Google agreed to fund the development of three new U.S. sites for advanced nuclear reactors with Elementl Power, a nuclear developer. Details on capacity and timelines are pending site approvals, but it’s aimed at long-term AI data center powering.
Google prioritizes hydropower and other renewables for rapid deployment, blending them with nuclear for a diversified portfolio.
| Deal | Partner(s) | Value & Capacity | Timeline & Details | Regions |
|---|---|---|---|---|
| U.S. Hydropower PPA | Brookfield Asset Management (incl. Brookfield Renewable Partners) | $3 billion for 3 GW | 20-year agreements; initial deliveries in 2026, with Google investing $25 billion in data centers over two years | Starts in Pennsylvania; expands to mid-Atlantic and Midwest (PJM region) for AI/cloud growth |
| UK Renewables Management | Shell | Undisclosed; supports broader clean energy procurement | Ongoing; Shell manages Google’s UK renewable sourcing as part of the US-UK Tech Prosperity Deal | United Kingdom; tied to £31 billion AI/quantum/nuclear investments |
Cipher Mining AI Hosting Deal (September 2025): Google is backing $1.4 billion of a $3 billion, 10-year colocation agreement between U.K.-based AI startup Fluidstack and Bitcoin miner Cipher Mining. This repurposes mining sites for 168 MW of AI high-performance computing starting October 2026, with Google gaining an option for a 5.4% stake in Fluidstack. It leverages crypto’s energy infrastructure for AI, potentially unlocking $7 billion in value.
$4 Billion Arkansas Data Center (October 2025): Google broke ground on a major AI-focused facility in West Memphis, Arkansas, to boost cloud and agentic AI capabilities. While not a direct energy deal, it’s powered by regional clean sources amid concerns over rising power bills from AI expansion.
Google for Startups AI for Energy Accelerator (September 2025): Program selected 29 startups from North America and Europe to innovate in AI-driven energy solutions, fostering tech for efficient AI power use.
These deals position Google ahead in the AI energy race, with nuclear providing stability for always-on inference and renewables scaling for training bursts. However, experts warn of grid strains, permitting hurdles, and potential rate hikes for consumers. Google’s 2025 sustainability report highlights efficiency gains—Gemini AI’s power needs dropped significantly—offsetting some growth. The US-UK Tech Prosperity Deal amplifies this, with Google joining Microsoft and Nvidia in £31 billion commitments for AI infrastructure. Overall, AI could double data center power use by 2026, making these pacts critical for Google’s carbon-negative ambitions by 2030.