A Japanese data center provider forecasts that AI compute will require 1 million GPUs going forward.

Japanese cloud computing provider Sakura Internet has launched an ambitious three-year project to scale out a data center focused on artificial intelligence, expecting to be fully booked by the time it's complete in 2027. To meet existing and future demand, the company is acquiring significant quantities of Nvidia GPUs. According to the company's CEO, while Sakura is in talks with Nvidia to buy 10,000 GPUs a year, demand is exploding, making a more appropriate target between 50,000 and 100,000 GPUs each year to meet current needs alone.

The scale of future demand is projected to be substantially larger. The CEO stated that if demand remains as high as it is now, the data center would need 1 million GPUs to keep up 10 years from now. When the facility becomes fully operational in 2027, it will allow local companies, universities, and research institutes to tap into the AI infrastructure Sakura is building. The CEO noted that its compute capacity is currently fully booked, with a waiting list that extends for years, driving the urgency to acquire more Nvidia hardware.

The company anticipates that demand for compute may currently be somewhat sporadic but will likely explode when AI becomes commonplace in cars and household appliances. The CEO noted that he doesn't believe AI will fundamentally change the experience of using a smartphone or a PC, a perspective that Google, Apple, and Microsoft would likely contest.

Sakura Internet was the first company in Japan to be chosen by the government to build what it calls a "public cloud," and it relies on government subsidies to help foot the bill for its upcoming Nvidia hardware purchases. According to Bloomberg, this hardware purchase will total almost $60 million in the next two months.