Hardware quality problems and server supply chain kinks slow Amazon’s $100 billion AI build

Reverses life extensions for some servers it now feels aren’t useful in the inferencing age

Amazon Web Services is struggling to get the high-quality servers it needs to build AI infrastructure and has retired other hardware early to make room to accelerated machines.

Those peeks into the state of AWS infrastructure came from CEO Andy Jassy and CFO Brian Olsavsky, who revealed the info during the Thursday earnings call that saw Amazon detail its 2024 financial year and outline future plans.

Olavsky used his prepared remarks to reveal that Amazon recently conducted a “useful life study” for its servers and networking equipment, and found “an increased pace of technology development, particularly in the area of artificial intelligence and machine learning.”

Those findings mean the company decided to reduce the useful life for a subset of its servers and networking equipment from six years to five years, as of January 2025. Amazon also retired some other servers and networking boxes early, for unexplained reasons.

Just a year ago, Amazon increased the working life of its servers from five to six years, up from the five years it decided on in 2022.

Also on the call, a Wall Street analyst asked if AWS’s AI infrastructure build is being hampered by supply chain issues.

Jassy said that is the case.

“We could be growing faster if not for some of the constraints on capacity,” he admitted, adding that “Chips from our third-party partners come in a little bit slower than before.”

He also said it can “take a little bit of time to get the hardware actually yielding the percentage healthy and high-quality servers we expect.”

Shortages of components like server motherboards are also a constraint, as is energy supply.

Supply problems matter because Amazon spent $26.4 billion on capital expenditure in the final quarter of 2204, and Olavsky said most of that went towards AI infrastructure for AWS. The CFO said $26.4 billion will be typical of Amazon’s quarterly capex in FY 2025, suggesting hardware spend approaching $100 billion for the year.

Spending on AI infrastructure is a hot topic after Chinese model-maker DeepSeek claimed it was able to train its wares on modest hardware in a short time, suggesting huge builds might not pay for themselves.

Jassy said Amazon is unworried by such arguments, because “we think virtually every application that we know of today is going to be reinvented with AI inside of it and with inference being a core building block just like compute and storage and database.”

The CEO said the faster AWS grows, the more its capex grows “because we have to procure data center and hardware and chips and networking gear ahead of when we're able to monetize it.”

“We don't procure it unless we see significant signals of demand,” he said. “And so, when AWS is expanding its Capex, particularly in what we think is one of these once-in-a-lifetime type of business opportunities like AI represents, I think it's actually quite a good sign medium-to-long term for the AWS business.”

The signs are already good for AWS, which posted annual revenue of $107.6 billion, 19 percent up from last year’s tally. Q4 revenue for AWS was $29 billion, meaning its annual revenue run rate is $115 billion. By way of comparison, Dell’s most recent full year revenue was $88.5 billion, IBM’s was $62.8 billion, Cisco racked up $53.8 billion of revenue and HPE won $30.1 billion. Microsoft’s annual revenue of $245.1 billion lies ahead – but it’s hard to disentangle the consumer, gaming, and advertising revenue from Redmond’s haul to make a direct comparison of business tech revenue. We think AWS is a few tens of billions behind for now.

Amazon’s revenue for the year ended December 31, 2024, was $638.0 billion, up eleven percent from FY 2023’s $574.8 billion. Net income of $59.2 billion almost doubled FY 2023’s $30.4 billion.

The company thinks those numbers will improve as it invests more in robotics, which will also account for chunk of capex spend. Jassy said the 1,000-plus generative AI applications already running at Amazon may also help. Among the uses of generative AI at Amazon are:

• An app for third-party sellers that lets them “take a picture of an image or point to a URL, and the Generative AI app will fill in most of the rest of the information they have to fill out, which speeds are getting selection on the website and easier for sellers.”
• A tool that compares catalogs from different clothing manufacturers and means Amazon knows “which brands tend to run big or small relative to each other. So when you come to buy a pair of shoes, for instance, it can recommend what size you need.”
• On Amazon Prime’s broadcasts of American Football, AI helps to issue “defensive alerts, where we predict which players is going to blitz the quarterback or defensive vulnerabilities where we were able to show viewers what area of the field is vulnerable.”

As 2025 unfurls, Jassy said shoppers on Amazon will see its “Rufus” agent appear more often and offer advice on what to buy.

Investors were advised that Q1 2025 will see revenue of $151.0 to $155.5 billion, growth of between five and nine percent. In case the guidance felt low, Amazon reminded shareholders that 2024 was a leap year in which the extra day’s trading on February 29th meant an additional $1.5 billion of sales,

Investors weren’t thrilled, with after hours trading seeing Amazon shares drop from over $239 apiece to just over $229.

The Q1 forecast appears to have been one reason for shareholders’ lack of enthusiasm. AWS’s growth rate may be another, as it’s accelerating more slowly than rivals Google and Microsoft. ®