Amazon Launches $11 Billion AI Data Center Project ‘Rainier’ in Indiana

NEW CARLISLE, Indiana – What was once sprawling farmland near Lake Michigan is now the epicenter of Amazon’s AI ambitions. Project Rainier, a 1,200-acre complex, stands as one of the world’s largest operational AI data centers. This isn’t just another brick-and-mortar structure; it’s where Amazon is actively training its cutting-edge artificial intelligence models using its proprietary chip technology.

Amazon, along with its competitors, has committed over $1 trillion to AI data center projects. While these endeavors signal a bullish outlook on AI’s future, their sheer scale has raised questions regarding the feasibility, financial viability, energy demands, and community impact of such massive undertakings.

While OpenAI envisions its “Stargate” project as a network of expansive AI data centers, Amazon’s $11 billion Project Rainier is already a tangible reality. This operational cluster is explicitly designed to train and run AI models developed by Anthropic, a leading AI startup and a key Amazon cloud customer and AI partner.

“This isn’t a conceptual project slated for future development,” Matt Garman, CEO of Amazon Web Services, emphasized in a recent interview with CNBC. “This facility is currently operational, actively training Anthropic’s AI models as we speak.”

The race is on among tech’s megacaps to establish supercomputing facilities, driven by the anticipated surge in AI demand. Meta is reportedly planning a 2-gigawatt Hyperion site in Louisiana, and Alphabet recently broke ground in West Memphis, Arkansas, positioning itself strategically across the Mississippi River from Elon Musk’s xAI data center, known as Colossus.

OpenAI’s aggressive expansion, involving 33 gigawatts of new compute commitments, reflects a projected $1.4 trillion in future obligations, with partnerships spanning Nvidia, Advanced Micro Devices, Broadcom, and Oracle.

Amazon’s execution in this domain benefits from years of experience in large-scale logistics. Bolstered by strong relationships with state and local authorities, Amazon leverages its expertise in constructing massive fulfillment centers, logistics hubs, AWS data centers, and its HQ2 project to accelerate the deployment of AI infrastructure.

“Deals can look promising on paper,” notes Mike Krieger, Chief Product Officer at Anthropic, “but their true value is realized when the infrastructure is fully operational and accessible to the customer. Amazon’s prowess in this area is unparalleled.”

The unveiling of Project Rainier precedes Amazon’s third-quarter earnings report. While investors will pay close attention to capital expenditures, they’ll also be scrutinizing the speed at which these investments translate into tangible revenue and, ultimately, profit.

Amazon recently announced 14,000 layoffs, indicating a strategic realignment and reallocation of resources toward high-growth areas like AI and the development of its Trainium chips.

The genesis of Project Rainier can be traced back to the spring of 2023. Shortly after the launch of ChatGPT, Amazon began scouting locations in rural Indiana, partnering with American Electric Power through its Indiana Michigan Power subsidiary. The following year, an $11 billion agreement was inked with the state, marking the largest capital investment in Indiana’s history.

Construction commenced in September of the previous year, and as of this month, seven buildings are operational, with two additional campuses in progress. Upon completion, the facility will encompass 30 buildings and consume over 2.2 gigawatts of electricity, enough to power over 1.6 million homes.

Josh Sallabedra, with 14 years of experience building data centers for Amazon, now leads the Indiana site. Relocating from the West Coast, Sallabedra enlisted four general contractors to accelerate the project timeline, noting the unprecedented speed of this undertaking.

“This pace reflects the current customer demand,” Sallabedra stated. “Anticipating the rise of AI and machine learning, we adjusted our building design accordingly.”

While some tech companies opt for temporary structures to expedite construction, Amazon continues to pursue a more robust approach. Despite the inherent time savings, Amazon maintains a commitment to quality and security.

“It’s not just about speed,” Garman stressed. “It’s about providing secure and reliable AWS infrastructure – an industrial-grade, enterprise-scale data center.”

Summarizing the rapid transformation, Garman remarked, “From cornfields to data centers, practically overnight.”

The landscape still bears the marks of recent construction, with workers moving between trailers and steel beams ascending in the distance. A steady stream of pickup trucks generates clouds of dust as it passes unfinished warehouse shells. From the security gate, a row of streetlamps stretches towards the data center’s core as lifts carry crates of chips.

This relatively undeveloped area of rural Indiana, known for its grain silos, transmission lines, and scattered barns, has become a focal point for ambitious infrastructural development. General Motors and Samsung are collaborating on a $3.5 billion electric vehicle battery plant nearby. During peak construction periods, the area has witnessed over 4,000 construction workers in a community with a population of only 1,900.

Local residents express mixed feelings about these developments.

“It’s increasingly difficult to see farmland disappear,” stated Marcy Kauffman, president of New Carlisle’s town council. “This project has consumed a significant portion of our farmland.”

Dan Caruso, a long-time resident, expressed concerns about the potential for further expansion, referencing cautionary advice from his friends.

Indiana Michigan Power anticipates a more than doubling of peak power demand by the decade’s end, raising concerns about rising household utility bills. A recent report indicated that monthly electricity bills in neighborhoods surrounding these new facilities are 267% higher compared to five years prior.

And expansion shows no sign of slowing down. “We are rapidly adding new capacity across our global infrastructure,” Garman stated. “We will continue to build as our customers’ needs for capacity increase.”

Project Rainier’s seven data center buildings are densely packed with Trainium 2 chips, Amazon’s custom-designed silicon. Notably absent are Nvidia’s market-leading graphics processing units (GPUs). Amazon asserts that this represents the largest deployment of non-Nvidia compute in the world.

“We already have approximately 500,000 Trainium chips actively running in Indiana today,” Garman said. “And the performance has been so positive that the initial order has been doubled.” Amazon anticipates deploying one million Trainium chips by the year’s end, signaling a significant bet on its in-house silicon.

Trainium 3, co-developed with Anthropic, is slated to launch in the coming months, further solidifying the partnership between the two companies.

Anthropic’s key infrastructure runs on AWS, making them one of the earliest AI labs to train models on Amazon’s custom silicon. Amazon’s $8 billion investment in the startup reinforces its comprehensive AI strategy.

While Trainium may not match the raw performance of Nvidia’s GPUs, AWS touts its technology’s superior density and efficiency. This allows for more chips per data center, facilitating higher aggregate compute while lowering power and cooling costs.

Amazon and Anthropic have collaborated on customized silicon designs tailored to real-world training requirements. Garman and Krieger confirmed that Anthropic provided direct input to optimize training speed, reduce latency, and enhance energy efficiency.

The goals behind Trainium 3 are to better support frontier models.

“It offers better performance, lower latency, and improved power consumption per flop,” Garman explained. “It will be deployed within Indiana and in numerous data centers worldwide.”

Prasad Kalyanaraman, Vice President of Infrastructure Services at AWS, emphasized the critical “ability to control the stack from the lower layers of the infrastructure” to “build the right set of capabilities that these model providers want.” This level of vertical integration is a strategic advantage in the rapidly evolving AI hardware landscape.

Anthropic is rapidly expanding, fueled by a substantial cash burn, as it strives to stay competitive with OpenAI and other AI innovators. Anthropic’s current annual revenue is near $7 billion, Claude’s chatbot supports >300,000 businesses which is a 300x increase in the last two years. Large enterprise customers are up to ~7x in one year.

Claude Code, Anthropic’s new agentic coding assistant, generated $500 Million in annualized revenue within two months.

Anthropic is diversifying its infrastructure partnerships. Last week, they announced a partnership with Alphabet, giving Anthropic access to up to 1 million of Google’s Tensor Processing Units (TPUs). The deal is worth billions of dollars.

Acknowledging existing funding from Google, Krieger emphasized Anthropic’s need for diverse processing power, highlighting the “multi-chip strategy.”

“The demand for our models is so high,” said Krieger, “that the only way we have been able to serve to this point is this multi-chip strategy.”

Garman is cognizant of the multi-cloud and multi-chip trends, and affirmed that Amazon has no intention of pursuing an acquisition of Anthropic.

“We value our relationship as it stands,” he stated.