Amazon (AMZN) is giving its massive Amazon Web Services (AWS) data centers a facelift. The company announced it’s improving the efficiency, availability, and design of its enormous computing facilities to keep them up and running for its customers and provide them with the kind of processing power they need to run new generative AI applications.
“We’ve done a ton of improvements in terms of simplification of our electrical and mechanical design that will improve the availability for our customers even more,” Prasad Kalyanaraman, VP of infrastructure services at AWS, told Yahoo Finance.
“Generative AI requires a meaningful amount of power for these racks and for these chips. And we’ve been innovating on our power delivery design to these servers as well,” he added.
AWS is Amazon’s most profitable business unit. The segment brought in $90.7 billion in 2023, lower than the retail division’s $484 billion in revenue. Operating income, however, topped out at $24.6 billion, eclipsing Amazon’s retail operating income of $12.2 billion.
Any changes to AWS’s infrastructure are sure to impact the company’s bottom line over time. AWS is the world’s largest cloud services provider, ahead of second-place Microsoft (MSFT) and third-place Google (GOOG, GOOGL). And anything that improves the cloud business’s availability and performance has the potential to draw more customers while ensuring existing users don’t flee to Amazon’s rivals.
Amazon says the latest updates include simplified electrical and mechanical designs for its data centers that ensure better than 99% infrastructure availability. In other words, the company streamlined its hardware to ensure its data centers stay up and running without any hiccups.
The tech giant says it’s also preparing for Nvidia’s (NVDA) upcoming Blackwell chips by outfitting its data centers with liquid cooling capabilities. Data centers generally use air-cooled servers that dissipate heat their chips and servers generate using fans. But according to Kalyanaraman, servers running Nvidia’s Blackwell chips or Amazon’s own Trainium 2 chips run so hot that the only way to cool them is using a liquid cooling system.
You can think of a liquid cooling system as akin to the radiator system in a car. Generally, coolant circulates through a series of pipes to a block attached to a server chip. The heat generated by the chip transfers to the block and then the coolant, which then passes through a radiator where it’s cooled and recirculated to the chip.
Heat is a massive problem for computer components. If a system gets too hot, it’ll either slow down or shut off completely to prevent serious damage. So keeping new AI servers cool is paramount for cloud computing companies.
