Amazon to spend $100B in AWS AI infrastructure

Amazon plans to spend roughly $100 billion this year on AWS infrastructure as the company tries to stay ahead of rivals Google and Microsoft in the AI race.

On Thursday, Amazon's stock price fell more than 4% in after-hours trading after the company forecast operating income of $14 billion to $18 billion for the current quarter ending in March, lower than analysts expected. The company's projected revenue of between $151 billion and $155.5 billion also failed to meet expectations.

Capital expenditures on AI infrastructure contributed to the disappointing numbers. Amazon spent $26.3 billion in the fourth quarter that ended in December. Amazon CEO Andy Jassy said the amount would be roughly the same during each quarter in 2025, which amounts to about $100 billion.

"The vast majority of that Capex is on AI for AWS," Jassy said during an earnings call Thursday.

Jassy defended plans to spend more in 2025 than the $75 billion spent last year because the company's infrastructure had to keep up with customer demand for AI cloud services.

"AI represents, for sure, the biggest opportunity since cloud, and probably the biggest technology shift in business since the internet," Jassy said. "So, I think that both our business, our customers and shareholders will be happy, medium to long-term, that we're pursuing the capital opportunity and the business opportunity in AI."

Andy Jassy

Amazon's cloud rivals Alphabet and Microsoft are also spending tens of billions of dollars annually on data centers, chips, servers and other infrastructure supporting their AI services. This week, Alphabet, Google's parent company, reported it planned to spend $75 billion in 2025 to support AI in Google Cloud. Last month, Microsoft President Brad Smith said the company would spend $80 billion on AI infrastructure for its Azure cloud by the end of the current fiscal year in June.

Jassy said he expects enterprises to spend more on AWS as AI service prices drop due to innovations that will lower the cost of models and associated technologies.

Advancements like DeepSeek, a Chinese reasoning model fine-tuned or inferenced on relatively inexpensive Nvidia GPUs, will eventually lead to lower AI pricing, Jassy said. However, people who believe this will lead to enterprises spending less are mistaken.

"Sometimes people make the assumption that if you're able to decrease the cost of any type of technology component ... that somehow it's going to lead to less total spend in technology," Jassy said. "We've never seen that to be the case."

Organizations will eventually spend more because they will build AI into more applications, he said. "I believe the cost of inference will meaningfully come down … [and] it will make it much easier for companies to be able to infuse all their applications with inference and with generative AI."