Advertise
AI

AWS CFO says AI server costs recouped within three years

John Felton pointed to a nearly $500 billion order backlog and $25 billion in annualized AI revenue as evidence the spending is sustainable.

John Felton, chief financial officer of Amazon Web Services, said the company recovers the cost of its AI server investments within three years, according to remarks reported by Axios. Felton said current AI spending levels are sustainable because the investment is driving meaningful innovation.

He cited an order backlog of close to $500 billion as of the second quarter as evidence against the argument that AI spending has formed a bubble. Orders are coming evenly from customers of different sizes, he said, including large enterprises, large AI research and development companies, and small and medium-sized businesses.

AWS posted second-quarter revenue of $42.2 billion, up 36.7% year on year, while its AI annualized revenue exceeded $25 billion, more than double the previous year.

Productivity rather than headcount cuts

Felton said AI should be seen as a tool for raising productivity rather than only cutting costs, by reducing tedious chores so employees can spend time on more valuable work. An analysis of the company's internal token usage showed token consumption and productivity were nearly directly proportional, he said.

The remarks come as research and advisory firm Gartner has warned that more than $3.5 trillion in AI infrastructure investment could be written down by the end of 2030. Gartner also predicted that more than 80% of AI-specialized cloud companies will not survive independently by 2030.

Quick answers

How long does AWS take to recover its AI server investment?

CFO John Felton said AWS recovers the cost within three years.

What is AWS's AI annualized revenue?

It exceeded $25 billion, more than double the previous year, according to Felton's remarks reported by Axios.

What was AWS's order backlog in the second quarter?

Close to $500 billion, which Felton cited as evidence against the AI bubble argument.

Source