Amazon Web Services (AWS) CEO Matt Garman has said that the company is still running six-year old Nvidia A100 servers, in part due to customer demand for GPU capacity continuing to outstrip supply.

Speaking to Jeetu Patel, Cisco’s president and chief product officer, during the Cisco AI Summit earlier this week, Garman said: "Because there is so much more demand than supply, there typically still is demand for the older chips... we actually are completely sold out and have never retired an A100 server.”

Nvidia first unveiled its A100 GPUs in 2020.

His comments echoed those made by Google’s VP and GM of AI and infrastructure, Amin Vahdat, last year, when he told attendees of the Andreessen Horowitz Runtime event that Google has seven generations of its TPU hardware in production and that its “seven and eight-year-old TPUs have 100 percent utilization.”

He went on to note that demand for the company’s TPUs is so oversubscribed that Google is having to turn customers away.

While AWS’ Garman said demand was undoubtedly a key reason why the hyperscaler was still offering A100 instances, he went on to note that there are also structural reasons why customers want to keep using old hardware.

“The industry today is getting so much gain on the AI side [by] reducing the floating point accuracy,” he explained. “I was talking to some customers recently, and they were saying ‘I really can’t move to Blackwells, I have to use [Intel] Haswells because I’m doing HPC-style calculations and the precision is not enough for me.’“

In June 2025, AWS announced that it was reducing costs for accessing Nvidia H100, H200, and A100 GPUs on its platform, with the on-demand price for A100 instances decreasing by 33 percent across both P4d and P4e instance types.