In 2025, US companies continue to compete to build the largest data centers for training AI algorithms. Meta’s CEO, Mark Zuckerberg, announced plans for 2025 even before revealing the company’s financial results. Meta plans to invest between 60 billion to 65 billion US dollars, mostly in data centers. By the end of the year, the company aims to operate more than 1.3 million GPU accelerators, including those already acquired, with several hundred thousand more to be added. Along with necessary processors and other infrastructure, the new 2025 hardware alone is expected to consume more than one gigawatt of electrical power. Meta typically purchases its accelerators from both Nvidia and AMD.
To accommodate new hardware, Meta is constructing a new data center in Richland Parish, Louisiana, designed for two gigawatts. The source of the energy was not disclosed by Zuckerberg. The layout is planned to be large enough to cover about half of Manhattan, which is 59 km² in size.
Amazon, Google, and Microsoft revealed their plans last fall. Amazon, during an analyst conference, projected capital expenditures of about 75 billion US dollars for the entire year of 2024, with expectations to increase in 2025. Google’s parent company, Alphabet, forecasted capital expenditures of approximately 52 billion US dollars in 2024 and also plans to spend more in 2025. Microsoft operates on fiscal years that are offset by half a year from the calendar. From July 2024 to the end of June 2025, Microsoft plans to allocate 80 billion US dollars, mainly for AI data centers. Together, these four industry giants could reach around 300 billion US dollars this year.
The companies are under pressure due to the Stargate project. With support from the US government, OpenAI, Softbank, and Oracle aim to invest 500 billion US dollars in AI development over the next few years. Additionally, the Chinese company Deepseek is causing concern in the US AI landscape by claiming to have trained its own model more cheaply and efficiently than its competitors.