In the era of AI reasoning, the computing power competition has entered the "power grab mode"! From hoarding GPUs to hoarding megawatts, Tesla has locked in Arizona's solar storage

According to the Zhitong Finance APP, renowned industrial project developer ContourGlobal has announced that electric vehicle, AI, autonomous driving, and robotics leader Tesla (TSLA.US) has reached a significant long-term agreement to invest heavily in purchasing electricity generated by a large solar and battery project in Arizona, USA, supported by private equity giant KKR.

This latest power supply agreement is rare for Tesla, which has a network of battery storage and solar assets, and highlights the increasing need for Tesla to sign more power purchase agreements as the demand for AI-driven data centers tightens the electricity market in the United States.

According to a statement, the company will sell 90% of the electricity generated by the Sterling project to Tesla. The facility is planned to be operational by 2028 and will include a peak solar generation capacity of 509 megawatts and a battery storage system capable of sustainable discharge for four hours at 360 megawatts. The financial terms were not disclosed by either party.

Tesla locks in Arizona solar and storage project, as the computing race escalates into an energy battle

This agreement is uncommon for Tesla and underscores the increasing demand for companies to sign so-called power purchase agreements as the scale of data centers driven by AI computing needs grows, tightening the U.S. electricity market. In addition to Tesla, other American tech giants such as Microsoft, Google, and Amazon have been signing such contracts for years and have been major drivers of the clean energy boom in solar and wind over the past decade. The combination of renewable energy projects like solar with large battery storage systems has become one of the fastest ways for tech giants to add scalable power capacity, especially in regions where utility companies struggle to meet the growing electricity demand.

The choice of four hours is primarily because the project is not aimed at providing "24-hour off-grid power" but rather transferring the abundant, low-cost solar energy from Arizona during midday to the peak electricity demand in the evening, approximately 4 to 5 hours later. When solar generation rapidly declines in the afternoon while residential and commercial loads remain high, the battery can discharge to alleviate the typical "duck curve" and evening power shortages. ContourGlobal has explicitly stated that the storage component is designed to supply clean electricity during evening peak periods; CAISO has also noted that most existing large-scale storage in California consists of four-hour lithium-ion batteries, primarily supporting the system during hot evenings after solar generation declines.

Four hours is also a common balance point between the economics of current lithium-ion storage and the value in the electricity market. The arbitrage value of electricity prices typically concentrates in the first few hours of the day when prices are highest: the first hour can capture the highest price difference, and subsequent hours usually see diminishing returns; meanwhile, extending storage duration requires nearly proportional increases in battery cells, cabinets, fire protection systems, and capital investment. The U.S. Department of Energy's ARPA-E has pointed out that short-duration lithium batteries are suitable for handling intra-day energy transfer from midday to evening, while prolonged low wind, low sunlight, or cross-day electricity shortages require longer-duration storage.

Tesla typically purchases electricity resources from local utility companies and grid systems. The company did not respond to requests for comments outside of normal business hours. Notably, Tesla also owns some hard assets related to power generation The large power station will connect to the grid system managed by the Western Area Power Administration in the United States and will be accessible to California. As the construction boom of AI data centers puts severe supply-demand pressure on grid areas that previously had a significant surplus of electricity, power grids across the United States are facing unprecedented demand growth.

According to long-term statistical data from the U.S. Energy Information Administration dating back to the late 1990s, the average electricity price for ordinary consumers in the U.S. is expected to rise by 4.3% this year, reaching an unprecedented record of 14.22 cents per kilowatt-hour.

This is the first agreement reached between Tesla and ContourGlobal. Such agreements typically last 10 to 15 years, providing buyers with a long-term, monitorable, and predictable cost of electricity resources, while also offering developers the revenue certainty needed for financing new projects.

The Sterling project will become the largest renewable energy asset in ContourGlobal's portfolio. The company will acquire this project by the end of 2024 and will trade the remaining 10% of the electricity generated in the market.

From buying GPUs to buying power plants: Tech giants bring their own power, electricity becomes the ultimate constraint on AI capital expenditure

The Trump administration's acceleration of advanced nuclear reactor approvals, pushing tech companies to pay for new power generation and grid upgrades for data centers, and Tesla's early lock-in of a large solar and storage project in Arizona essentially point to the same structural change: AI competition has shifted from "can we obtain enough chips" to "can we obtain sufficiently stable, predictable electricity without raising residential prices in the long term."

The Berkeley Lab under the U.S. Department of Energy predicts that by 2030, data centers could account for about 11.8% of total electricity consumption in the U.S., with different scenario ranges from 9.5% to 15.3%; the International Energy Agency (IEA) expects global electricity consumption by data centers to increase from about 485 terawatt-hours in 2025 to about 950 terawatt-hours in 2030, indicating that electricity demand from data centers is nearly doubling.

The so-called AI inference era means that electricity demand is shifting from periodic model training loads to high-frequency online computing continuously triggered by applications such as search, AI agents, video generation, enterprise Copilot, and autonomous driving. While the energy efficiency of a single inference may decline due to improvements in chips and algorithms, the expansion of model invocation volume, context length, inference chain depth, and concurrent user scale may more quickly consume efficiency dividends; the IEA points out that AI training and model usage will also cause significant and rapidly changing electricity loads, making energy storage and supply flexibility important conditions for reliable operation. Therefore, the constraints on AI expansion in the future will not only be the number of GPUs but also the "effective megawatts" constituted by available electricity, grid capacity, transformers, transmission lines, cooling systems, and backup power sources.

Policy efforts have already begun to redistribute costs around this bottleneck. The Trump administration's "Electricity User Protection Commitment" requires large tech companies to be responsible for their new loads by building or expanding power generation facilities, bearing the costs of transmission and distribution upgrades, and signing special electricity pricing agreements, rather than passing the costs of data center expansion onto residents; however, this commitment is currently still largely voluntary, and its actual binding force and cost isolation effects remain controversial At the same time, nuclear regulatory reforms require a reassessment and streamlining of the advanced reactor approval process, and the NRC has also launched a new licensing pathway for advanced reactors. The implications of this policy combination are very clear: in the short term, rapidly increasing capacity through natural gas, solar energy, and energy storage, while in the long term, relying on nuclear power and other stable sources of energy to support high-utilization AI infrastructure.

Tesla's purchase of 90% of the power generation from the Sterling project is a reflection of this trend on the corporate procurement side. The project is scheduled to be operational by 2028, featuring 509 megawatts of peak photovoltaic capacity and 360 megawatts, approximately 1.4 gigawatt-hours of four-hour energy storage, with an expected annual power generation of over 1 terawatt-hour. It can shift daytime photovoltaic energy to evening peak hours and lock in costs and supply through long-term power purchase agreements, but the four-hour energy storage still cannot independently bear the all-weather load of data centers, ultimately requiring collaboration with the grid, nuclear power, natural gas, or other stable power sources. This indicates that Tesla's move may not directly equate to powering a specific AI data center, but it clearly reflects that, against the backdrop of tightening electricity supply, large technology and manufacturing companies are elevating long-term energy procurement to a strategic level equal to that of chip procurement