Amazon, Google, Meta, Microsoft, xAI, Oracle, and OpenAI Are All Building Their Own Power Plants — Here Is Why

Something unprecedented is happening in the energy sector: the seven largest AI companies in the world — Amazon, Google, Meta, Microsoft, xAI, Oracle, and OpenAI — are all actively building or contracting their own dedicated electricity generation. Not buying power from utilities. Building power plants. This shift from energy consumer to energy producer is one of the least-covered but most consequential developments in tech in 2026.

Why AI Is Causing an Energy Crisis

Training and running large language models consumes extraordinary amounts of electricity. A single H100 GPU draws 700 watts. A cluster of 100,000 H100s — a scale that multiple companies now operate — draws 70 megawatts continuously. Data centers for inference at scale require hundreds of megawatts. Microsoft's partnership with OpenAI alone requires the equivalent of a medium-sized power station running 24/7.

The US power grid was not built for this. In many regions, utility companies have years-long backlogs for new large commercial connections. AI companies needed gigawatts of power on timelines that utilities could not accommodate. The solution: build your own.

What Each Company Is Doing

Amazon (AWS): Amazon has signed agreements to power data centers with nuclear energy, including a deal with Talen Energy for a Pennsylvania nuclear facility. AWS is also investing in small modular reactors (SMRs) through partnerships with companies including X-energy. Total committed capacity: multiple gigawatts across nuclear, solar, and wind.

Google: Google signed the largest corporate nuclear power purchase agreement in history in 2024 — a deal with Kairos Power for SMRs to come online in the early 2030s. In the near term, Google is also building large-scale solar and wind installations specifically attached to data center campuses. The company announced it would match 100% of its AI compute electricity with carbon-free energy 24/7 by 2030.

Meta: Meta has committed to building dedicated natural gas generation facilities adjacent to its largest AI training campuses to guarantee power availability independent of grid constraints. It is also investing in offshore wind.

Microsoft: The highest-profile bet — Microsoft reopened Three Mile Island Unit 1 in Pennsylvania to power its AI data centers. The 835-megawatt nuclear plant came back online in late 2024 under a 20-year power purchase agreement. This is the first US nuclear plant restart in decades. Microsoft is also funding SMR development.

xAI: Elon Musk's xAI built the Colossus cluster in Memphis, Tennessee — reportedly powered partly by natural gas turbines that xAI installed directly at the site because the Memphis grid could not supply power fast enough. Colossus is reported to draw over 150 megawatts.

Oracle: Oracle is investing heavily in data center buildout across the US, including co-locating with power generation. The company has signed large nuclear power agreements and is building AI infrastructure campuses designed around energy self-sufficiency.

OpenAI: OpenAI's Stargate initiative — a joint venture with SoftBank, Oracle, and others — is a $500 billion commitment to US AI infrastructure that explicitly includes energy generation. The plan includes dedicated natural gas plants and nuclear partnerships. OpenAI CEO Sam Altman has separately invested personally in nuclear energy startups including Oklo.

What This Means for Developers and the Cloud

Cloud pricing pressure: Building power generation infrastructure costs billions of dollars. These costs will eventually appear in cloud pricing — either directly (higher compute costs) or indirectly (reduced margin on AI API calls meaning less aggressive price competition). The low-cost era of AI APIs may be extended by efficiency gains but will eventually encounter the reality of energy capital expenditure.

Reliability improvements: Self-generated power means less dependency on grid outages, utility pricing spikes, and brownouts. For developers running latency-sensitive or always-on AI workloads, cloud regions with dedicated power generation are more resilient.

New regions for performance: Companies are choosing data center locations partly based on where they can build or access power. This is already shifting where cloud regions are being built — Pennsylvania (nuclear), Texas (solar + gas), the Southeast (natural gas), and eventually the Pacific Northwest (hydro + nuclear). Where your workload runs will increasingly be influenced by energy availability.

The SMR bet: Small modular reactors — compact, factory-built nuclear reactors in the 50-300 MW range — are the long-term bet most companies are making. If SMR deployment succeeds at scale in the early 2030s, it changes the entire energy cost picture for AI. If it is delayed (as nuclear projects historically are), AI companies face a gas-heavy transition period.

The Grid Stability Question

There is a legitimate policy concern: if the largest power consumers build private generation rather than drawing from the grid, they take their capital and expertise out of the shared infrastructure. Communities and smaller businesses still depend on a grid that may be under-invested. This tension is already appearing in state utility commission proceedings in Virginia, Texas, and North Carolina — where data center buildouts are straining local grids while the data centers themselves increasingly bypass local utilities.

For developers: energy is now a first-order infrastructure consideration alongside compute, network, and storage. The companies that own their electricity will have structural cost and reliability advantages that pure cloud users do not. Watch which cloud providers have the most energy-independent infrastructure — that will increasingly predict which ones maintain pricing and reliability advantages in the AI era.