Internal Documents Detail Amazon’s Titus Data Center Modernization Program

Business Insider reported on internal AWS planning documents describing a multi-year data center modernization program internally codenamed Titus, covering power architecture, liquid cooling, server and rack layouts, and accelerated deployment timelines for next-generation AI workloads. After the story published, AWS Global Communications told Let's Data Science by email that the underlying initiative was first announced publicly in December 2024, that Titus is an internal codename for a broader, multi-year effort rather than a new announcement, and that its scope extends beyond GPU cooling alone. An AWS spokesperson said: "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 initiative reported here, which was announced in December 2024, reflects how we're designing our data centers to support the next wave of AI workloads and the growing demands our customers are bringing to us."

• •December 2024: AWS publicly announced new data center components designed to support next-generation AI workloads at its re:Invent keynote, according to AWS's own account.
• •July 2025: AWS unveiled the In-Row Heat Exchanger (IRHX), its custom direct-to-chip liquid cooling system, co-engineered around NVIDIA Blackwell-class GPUs.
• •May 2026: Business Insider reported on internal Titus planning documents; AWS subsequently clarified to Let's Data Science that Titus is part of the previously announced, broader program.

Per AWS's own account of the IRHX build, the system uses a direct-to-chip "cold plate" design that carries heat away in a closed liquid loop, which AWS says does not increase a data center's water consumption. AWS's Dave Klusas, senior manager of data center cooling systems, said the design went from a whiteboard concept to a working prototype in four months and to a first production unit in 11 months, and that the system is built to be added only where liquid cooling is needed rather than deployed uniformly across all data centers. Business Insider's reporting on the internal documents describes Titus as reworking generational data center design to accommodate larger GPU racks, denser power and cooling requirements, and faster build cycles.

Multi-year data center modernization programs at this scale matter because they signal where the broader infrastructure stack is heading: facility power delivery and distribution, higher-capacity direct-to-chip cooling, and denser rack-level electrical and mechanical design. Practitioners planning AI training or inference capacity on AWS should watch for instance types or availability zones that require or benefit from liquid cooling, since those are the practical, deployable signals of this program rather than the internal codename itself.

• •Published AWS instance specifications that require liquid cooling or higher power-per-rack envelopes, and which availability zones receive them first.
• •Amazon's realized capital expenditure this year against the reported $200 billion plan, as an indicator of how much of that spend is infrastructure versus other capex.
• •Whether competing hyperscalers disclose comparable direct-to-chip cooling builds or timelines, which would suggest this is now an industry-wide requirement rather than an AWS-specific bet.

AWS's correction is itself a useful signal: the company pushed back not on the underlying facts Business Insider reported, but on the framing that a new, GPU-specific program had just been launched, clarifying instead that this is a previously disclosed, multi-year effort. That distinction matters for practitioners trying to gauge how far along AWS's liquid-cooling rollout actually is, since "already in production and ramping" reads differently than "newly announced."