Amazon is changing the way it builds data centers to make these giant facilities “future-proof” for the age of AI.
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 purpose of Titus’ diverse program portfolio is to deliver the next generation of AWS data center designs,” says one internal document.
This effort highlights how the AI boom is reshaping the physical infrastructure behind cloud computing as AI systems become increasingly power-hungry and data centers move beyond traditional designs.
The document shows AWS is pursuing faster deployment schedules, advanced liquid cooling systems, and more flexible power architectures while meeting emissions targets and more stringent noise requirements.
AI infrastructure is increasingly a priority for Amazon, with the company planning a record $200 billion in capital spending this year, much of it related to AI data centers. The company also launched another modular data center initiative, internally dubbed “Houdini,” Business Insider previously reported. And last month, the company promoted Prasad Kalyanaraman, vice president of AWS Infrastructure Services, to the elite S-Team leadership group.
Faster construction and more liquid cooling
According to one of the documents, the Titus project initially focused on reducing data center construction time to meet rapidly growing demand. However, this effort has since expanded to broader upgrades of AWS infrastructure for future AI chips.
According to a document created earlier this year, AWS aims to reduce the timeline from “starting a shell” to the first working server room to less than 35 weeks, well below industry standards. (Shell start is the point at which construction of the basic exterior of the data center building begins.)
According to one of the documents, Titus is increasing its total computing power per site from about 58 megawatts to about 68 megawatts.
Titus can support broader water cooling deployments, reflecting the industry’s growing recognition that traditional air cooling may be inadequate for increasingly power-hungry AI hardware. According to internal documents, the system aims to achieve a 15% reduction in cooling power consumption compared to current AWS designs.
At the heart of that effort is AWS’s “in-row heat exchanger” system (IRHX), according to one of the documents. This is an in-house developed liquid cooling technology that does not require a major redesign of existing data centers.
Amazon CEO Andy Jassy shared some details about IRHX last year, saying it will enable AWS to “support traditional workloads and demanding AI applications in the same facility.”
IRHX Liquid Cooling System from Amazon Noah Berger/Noah Berger
Lake Knudsen, an analyst at Semianalysis, told Business Insider that Amazon’s push to speed up data center deployments and make widespread use of liquid cooling represents a “significant strategic move.”
“We’re seeing Amazon really jump into the race with new designs that are optimized for faster deployment,” Knuhtsen said.
An AWS spokesperson told Business Insider that the Titus initiative is part of the company’s efforts to “support the next wave of AI workloads,” but noted that the Titus initiative is separate from the Houdini initiative.
“We are constantly innovating to provide our customers with the fastest, most resilient, safest and most sustainable infrastructure, and our scale allows us to do so while keeping costs down,” the spokesperson said.
More demanding Nvidia systems
Titus’ roadmap suggests that AWS is preparing a new generation of hardware that is even more power-intensive.
The document repeatedly mentions higher rack power density and support for new GPUs and servers, including Nvidia GB200 systems and newer. These are one of the most demanding AI server platforms in the industry.
One of Amazon’s internal documents mentions plans for wider aisles designed to accommodate “GB200 and future generations of racks with increased space requirements.” This is partially due to the new cables creeping out the front of the server rack and taking up more space.
The new version of Titus Data Center is expected to be rolled out in the first half of 2027. This should come in time for Nvidia’s latest Vera Rubin GPU server system, which promises higher performance, greater memory capacity, faster GPU-to-GPU networking, and dramatically higher power consumption.
“Stuck Power”
Titus also strives to reduce costs and environmental impact. Internal goals call for a 10% reduction in cost per kilowatt of IT capacity compared to previous AWS designs, while also meeting the company’s 2028 carbon intensity goals and more stringent noise standards.
The company is also looking to make its facilities more adaptable over time. Titans is designed to quickly scale data center PODs, or blocks of servers and other equipment, without requiring custom redesign, potentially allowing AWS to scale capacity more quickly.
Engineers are also looking to reduce “stranded power,” or unused electrical capacity, by lowering the minimum power requirements for racks, one of the documents said.
In general, if your data center is only equipped with the latest power-hungry GPU systems, you may have less flexibility when handling less demanding AI workloads. Having a wide variety of server racks allows you to start up low-power systems on demand, ensuring you use as much of the available power as possible.
According to the document, Titus involves a wide range of component upgrades across AWS data centers, rather than routine infrastructure upgrades.
As one internal planning document states, “How can we future-proof our data center design to better respond to capacity demand signals from existing and future AWS services?”
Any tips? To contact this reporter via email, please specify the following address: ekim@businessinsider.com or on Signal, Telegram or WhatsApp 650-942-3061. Use a personal email address, non-work WiFi network, and non-work device. Here’s a guide to sharing your information securely.