In the era of AI inference, the computing power competition enters the "power grabbing mode"! From hoarding GPUs to hoarding megawatts, Tesla (TSLA.US) secures Arizona’s solar storage

According to news from 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 major long-term agreement to spend heavily on purchasing electricity generated by a large-scale solar and battery project in Arizona, USA, supported by private equity giant KKR.

This latest power supply agreement is quite rare for Tesla, which already owns a network of battery storage and solar assets. It highlights the tightening of the U.S. electricity market as data center growth driven by massive AI inference demand accelerates, forcing Tesla to sign more power purchase agreements.

According to a statement, the company will sell 90% of the Sterling project’s power output to Tesla. The facility, scheduled to start operations in 2028, will include 509 MW of peak solar power generation capacity and a 360 MW battery storage system capable of sustained four-hour discharges. Financial terms were not disclosed by either party.

Tesla Secures Arizona Solar-Storage Project, Computing Power Race Escalates Into Energy Battle

This agreement is uncommon for Tesla and highlights a growing need for corporations to sign so-called power purchase agreements as U.S. electricity markets tighten, driven by the rapid expansion of data centers for AI inference computing. In addition to Tesla, other American tech giants such as Microsoft, Google, and Amazon have been securing such contracts for years and are the main driving force behind the boom in clean energy like solar and wind over the past decade. Combining renewable projects like solar with large battery storage systems has become one of the fastest ways for tech giants to add large-scale power capacity, especially in regions where U.S. utilities struggle to meet growing demand.

The four-hour duration was chosen primarily because the project does not aim for “24/7 complete grid independence,” but rather to shift a large amount of low-cost midday solar energy in Arizona to cover about four to five hours of peak evening demand. When solar output rapidly decreases in the afternoon but residential and commercial loads remain high, the battery discharges in a concentrated manner, helping to alleviate the typical “duck curve” and evening power shortages. ContourGlobal explicitly stated that the storage component is designed to supply clean power during evening peaks; CAISO also pointed out that most of California’s large-scale storage consists of four-hour lithium-ion batteries, which mainly support the grid during hot evening hours after solar output drops.

Four hours is also currently the common balance point between the economics of lithium-ion storage and electricity market value. The value of price arbitrage is usually concentrated in the hours of the highest daily electricity price: the first hour captures the highest price differential, and the incremental benefit typically decreases with each added hour. At the same time, extending storage duration nearly proportionally increases costs for cells, cabinets, fire protection, and capital. According to the U.S. Department of Energy’s ARPA-E, short-term lithium batteries are suitable for handling intraday energy shifting from midday to evening, while low wind, low sun, or multi-day shortages require longer-term storage solutions.

Tesla typically purchases electricity resources from local utilities and grid systems. The company did not respond to requests for comment outside normal business hours. It’s worth noting that Tesla also owns some power generation assets.

This large power station will connect to the grid system managed by the U.S. Western Area Power Administration and will have access to California. As the AI data center construction boom puts severe supply pressure on grid areas that until recently had surplus power, the U.S. is facing unprecedented growth in electricity demand across its grid.

According to long-term statistics from the U.S. Energy Information Administration dating back to the late 1990s, the average electricity price for U.S. residential customers is expected to rise 4.3% this year, reaching a record-high 14.22 cents per kilowatt-hour.

This is the first agreement reached between Tesla and ContourGlobal. Such agreements typically last for 10 to 15 years, providing buyers with long-term, reliable, predictable power costs and giving developers the revenue certainty needed to finance new projects.

The Sterling project will become ContourGlobal’s largest renewable energy asset. The company acquired the project at the end of 2024 and will independently trade the remaining 10% of its power output on the market.

From Buying GPUs to Buying Power Plants: Tech Giants Secure Their Own Power, Electricity Becomes the Ultimate AI Capital Expenditure Constraint

The Trump administration’s acceleration of advanced nuclear reactor approvals, its push for tech companies to pay for new power generation and grid upgrades to support data centers, and Tesla’s preemptive securing of a large-scale solar-storage project in Arizona, all point to the same structural shift: the AI competition has escalated from “can you get enough chips” to “can you secure sufficient, stable, predictable, and non-residential-rate-hiking electricity over the long run.”

According to Berkeley Lab under the U.S. Department of Energy, by 2030, data centers could account for about 11.8% of the total U.S. power consumption, with scenarios ranging from 9.5% to 15.3%. The International Energy Agency estimates that global data center power usage will rise from about 485 TWh in 2025 to around 950 TWh in 2030, nearly doubling data center electricity demand.

The so-called AI inference era means that electricity demand shifts from occasional model training loads to high-frequency online computing driven by applications like search, AI agents, video generation, enterprise copilots, and autonomous driving. While per-inference energy consumption may decrease with chip and algorithm improvements, the expansion in model call volume, context length, depth of inference chains, and number of concurrent users could quickly outpace these efficiency gains. The IEA points out that AI training and model use can lead to large, rapidly changing electricity loads, making storage and flexible power supply essential for reliable operations. As a result, the key constraint on AI expansion in the future may not just be the number of GPUs, but the combined “effective megawatts” available—including power, grid access, transformers, transmission lines, cooling systems, and backup power.

Policy makers are already reallocating costs around this bottleneck. The Trump administration’s “Electricity Customer Protection Pledge” requires large tech companies to be responsible for their additional load by building or expanding new power facilities, paying for transmission and distribution upgrades, and signing special rate agreements, instead of shifting data center expansion costs onto residential customers. However, this pledge is currently mostly voluntary and its binding effect and cost segregation remain open to debate.

Meanwhile, nuclear regulatory reform demands a reevaluation and reduction of advanced reactor approval timelines, and the NRC has introduced a new licensing pathway for advanced reactors. The policy mix is clear: short-term capacity increases will come from natural gas, solar, and storage, while long-term, all-weather stability depends on nuclear and other resilient power sources to support high-utilization AI infrastructure.

Tesla’s purchase of 90% of the Sterling project’s output is a microcosm of this trend in corporate procurement. The project is scheduled to launch in 2028, equipped with 509 MW peak PV and 360 MW (about 1.4 GWh) four-hour storage, expected to generate over 1 TWh annually. It can shift day-time solar power to night-time peaks and lock in cost/supply through long-term contracts, but four-hour storage still cannot independently support 24/7 data center loads, ultimately requiring coordination with the grid, nuclear, natural gas, or other stable sources. This shows that Tesla’s move is not necessarily directly powering a specific AI data center, but clearly reflects how, amid tightening power supply, large tech and manufacturing companies are elevating long-term energy procurement to the same strategic level of importance as chip acquisition.