Amazon Redesigns Indiana Data Center for In-House AGI Push

Amazon is transforming portions of a major AI data center campus in rural Indiana into what it calls an "AGI SuperCluster" designed to train the company's next generation of frontier AI models, according to internal planning documents.

The redesign consolidates multiple data centers and deploys thousands of servers powered by Amazon's custom Trainium chips, part of a broader initiative the company internally refers to as "AGI Pivot." The effort supports Amazon's AGI organization and its development of future in-house AI models, according to people familiar with the plans who spoke on condition of anonymity.

Why it matters

The infrastructure overhaul demonstrates Amazon's continued commitment to building proprietary frontier AI models despite recent layoffs in its AGI division and the discontinuation of its Nova model series. By repurposing existing facilities rather than building from scratch, Amazon is attempting to maximize returns on infrastructure investments while racing to meet accelerated training deadlines—a strategy that could influence how other hyperscalers approach capacity constraints in an overheated AI market.

Accelerated deployment schedules

The planning documents describe an "emergent request" to deploy more than 6,000 Trainium-powered AI servers on expedited timelines, moving launch schedules forward by several weeks. The accelerated pace was intended to prepare Amazon's next AI model for unveiling at this year's re:Invent conference, typically held in early December.

The documents also outline plans to expand data storage infrastructure to support multimodal AI training, which requires processing large volumes of high-resolution images while repeatedly saving model checkpoints during training runs.

Working alongside Anthropic

The Indiana campus houses facilities that are part of Project Rainier, the Trainium-based AI supercomputer Amazon built primarily for Anthropic. The new AGI initiative operates from the same data center complex but uses separate infrastructure that won't affect existing Project Rainier servers.

To create the AGI SuperCluster, Amazon is redesigning networking, storage, and fiber-optic systems so multiple data centers can function as a unified computing environment. Some existing buildings are being converted into "annexes" that share core networking equipment with neighboring facilities rather than operating independently. Certain locations are also replacing older Trainium 2 systems with newer Trainium 3 servers.

Trainium revenue acceleration

The infrastructure push comes as Amazon's custom chip business gains momentum. Last week, the company disclosed that its custom chip division—which includes Trainium AI accelerators and Graviton processors—is on track to generate more than $25 billion in annual revenue, up from a $20 billion projection the previous quarter.

Amazon recently increased its projected 2026 capital expenditures to $220 billion from $200 billion, citing continued demand for AI computing capacity that outstrips supply and rising component costs.

CEO Andy Jassy said last week that AI infrastructure investments should deliver attractive long-term returns because data centers remain productive for decades while servers and networking equipment can be upgraded over time.

One person familiar with the effort told Business Insider that the AGI push "has not slowed down" despite last month's workforce reductions in the AGI organization.

An Amazon spokesperson said the company is "always designing, upgrading, and improving our data center infrastructure to serve the diverse needs of our customers and teams," adding that the work improves "speed, cost, and sustainability."

These details were first reported by Business Insider.