Amazon is changing the way it builds data centers, to “future-proof” these giant facilities for the AI era.
Loading audio narration…
An internal AWS initiative known as “Titus” aims to dramatically accelerate data center construction while redesigning facilities to meet the demands of next-generation AI hardware, according to internal planning documents reviewed by Business Insider.
“The objective of the Titus portfolio of program variants is to deliver the next AWS generational Data Center design,” one internal document states.
The effort underscores how the AI boom is reshaping the physical infrastructure behind cloud computing, as increasingly power-hungry AI systems push data centers beyond traditional designs.
The documents show AWS pursuing faster deployment timelines, advanced liquid-cooling systems, and more flexible power architectures, while also meeting emissions targets and stricter noise requirements.
AI infrastructure has become a growing priority for Amazon, which plans a record $200 billion in capital expenditures this year, much of it tied to AI data centers. The company has also launched a separate modular data-center initiative internally dubbed “Houdini,” Business Insider previously reported. And last month, it promoted AWS infrastructure services VP Prasad Kalyanaraman to its elite “S-team” leadership group.
Faster construction and more liquid cooling
The Titus project initially focused on shortening data center construction time to meet rapidly growing demand, according to one of the documents. But the effort has since expanded into a broader upgrade of AWS infrastructure for future AI chips.
AWS aims to reduce the timeline from “shell start” to the first operational server room to under 35 weeks, well below industry standards, according to the documents, which are from earlier this year. (Shell start is the point when construction begins on the basic outer structure of a data center building).
Titus also raises total compute capacity per site to about 68 megawatts from about 58 megawatts, according to one of the documents.
Titus can support much broader liquid-cooled deployments, reflecting the industry’s growing recognition that traditional air cooling may not be sufficient for increasingly power-hungry AI hardware. Internal documents say the system aims to achieve a 15% reduction in cooling power consumption compared with current AWS designs.
Central to that effort is AWS’s “In-Row Heat Exchanger” system, or IRHX, an internally developed liquid-cooling technology that doesn’t require major redesigns to existing data centers, according to one of the documents.
Amazon CEO Andy Jassy publicly shared some details of IRHX last year, saying it enables AWS to “support traditional workloads and demanding AI applications in the same facilities.”
Noah Berger/Noah Berger
Reyk Knuhtsen, an analyst at SemiAnalysis, told Business Insider that Amazon’s push toward faster data center deployment and broader use of liquid cooling marks an “important strategy push.”
“We’re seeing Amazon really come out to the races with new designs optimized for faster deployment,” Knuhtsen said.
An AWS spokesperson told Business Insider the Titus initiative is part of the company’s effort to “support the next wave of AI workloads,” while noting Titus is separate from the Houdini initiative.
“We’re always innovating to provide customers the fastest, most resilient, most secure, and most sustainable infrastructure, and our scale lets us do that while keeping costs down,” the spokesperson said.
More demanding Nvidia systems
The Titus roadmap suggests AWS is preparing for a new generation of even more power-intensive hardware.
The documents repeatedly reference higher rack power density and support for new GPUs and servers, including Nvidia GB200 systems and beyond. These are among the industry’s most demanding AI server platforms.
One internal Amazon document mentions plans for wider aisles designed to accommodate “GB200 and future generations of racks with increased space requirements.” That’s partly due to new cables that squirm out of the front of the server racks, taking up more space.
A newer version of Titus data centers is expected to roll out in the first half of 2027. That should be in time for Nvidia’s latest Vera Rubin GPU server systems which promise higher performance, much larger memory capacity, faster networking between GPUs — and dramatically higher power use.
“Stranded power”
Titus also seeks to lower costs and reduce environmental impact. Internal targets call for a 10% reduction in cost per kilowatt of IT capacity compared with previous AWS designs, while also meeting the company’s 2028 carbon-emissions intensity goals and stricter noise standards.
The company is also attempting to make facilities more adaptable over time. Titus is designed to quickly scale data center PODs, or blocks of servers and other equipment, without requiring custom redesigns, potentially allowing AWS to expand capacity more quickly.
Engineers are also trying to reduce “stranded power,” or unused electrical capacity, by lowering minimum rack power requirements, one of the documents stated.
Generally, if a data center is packed with only the latest, power-hungry, GPU systems, that can reduce flexibility when handling less-intense AI workloads. Having a wider range of different server racks means some less power-hungry systems can kick in when needed — ensuring as much available electricity is used as possible.
The documents suggest Titus is more than a routine infrastructure upgrade, with a broad range of component upgrades across AWS data centers.
As one internal planning document puts it: “How do we future-proof our Data Center design and better respond to capacity demand signals of existing and future AWS services?”
Have a tip? Contact this reporter via email at ekim@businessinsider.com or Signal, Telegram, or WhatsApp at 650-942-3061. Use a personal email address, a nonwork WiFi network, and a nonwork device; here’s our guide to sharing information securely.
