Amazon opens $11 billion AI data center in rural Indiana as rivals race to break ground

NEW CARLISLE, Indiana — A year ago, it was farmland. Now, the 1,200-acre site near Lake Michigan is home to one of the largest operational AI data centers in the world. It’s called Project Rainier, and it’s the spot where Amazon is training frontier artificial intelligence models entirely on its own chips.

Amazon and its competitors have pledged more than $1 trillion towards AI data center projects that are so ambitious, skeptics wonder if there’s enough money, energy and community support to get them off the ground.

OpenAI has Stargate — its name for a slate of mammoth AI data centers that it plans to develop. Rainier is Amazon’s $11 billion answer. And it’s not a concept, but a cluster that’s already online. 

The complex was built exclusively to train and run models from Anthropic, the AI startup behind Claude, and one of Amazon’s largest cloud customers and AI partners.

“This is not some future project that we’ve talked about that maybe comes alive,” Matt Garman, CEO of Amazon Web Services, told CNBC in an interview at Amazon’s Seattle headquarters. “This is running and training their models today.”

Tech’s megacaps are all racing to build supercomputing sites to meet an expected explosion in demand. Meta is planning a 2-gigawatt Hyperion site in Louisiana, while Google parent Alphabet just broke ground in West Memphis, Arkansas, across the Mississippi River from Elon Musk’s Colossus data center for his startup xAI.

In the span of a month, OpenAI committed to 33 gigawatts of new compute, a buildout CEO Sam Altman says represents $1.4 trillion in upcoming obligations, with partners including Nvidia, Advanced Micro Devices, Broadcom and Oracle.

Amazon is already delivering, thanks to decades of experience in large-scale logistics. From massive fulfillment centers and logistics hubs to AWS data centers and its HQ2 project, Amazon has deep and close relationships with state and local officials and a playbook that’s now being used to get AI infrastructure set up in record time.

“These deals all sound great on paper,” said Mike Krieger, chief product officer at Anthropic, which has raised billions of dollars from Amazon. “But they only materialize when they’re actually racked and loaded and usable by the customer. And Amazon is incredible at that.”

The public unveiling of Rainier comes a day ahead of Amazon’s third-quarter earnings report. Investors will be listening closely for commentary on capital expenditures, but they also want to know how quickly capex projects will convert into revenue, and eventually, profit.  

On Tuesday, Amazon announced 14,000 layoffs as part of a broader push to flatten management and reallocate resources to priority areas like AI and the company’s Trainium chips.

The genesis of the Rainier complex dates back to the spring of 2023.

Roughly six months after ChatGPT launched, Amazon started scouting land in rural Indiana, working with American Electric Power through its Indiana Michigan Power subsidiary. A year later, it signed an $11 billion agreement with Indiana, the largest capital investment in the state’s history.

Construction began in September of last year and, as of this month, seven buildings are already online, with two more campuses underway. The full site will eventually span 30 buildings and draw more than 2.2 gigawatts of electricity, enough to power more than 1.6 million homes.

Josh Sallabedra, who’s spent 14 years building data centers for Amazon, is now the Indiana site lead. He relocated from the West Coast last year to oversee the project. Sallabedra brought on four general contractors to accelerate the timeline and says he’s never seen the company move this fast. 

“That’s the customer demand right now,” Sallabedra told CNBC. “As we saw AI and machine learning coming, we changed to a different building type.”

While some tech giants are throwing up temporary structures to move faster — Meta is building under giant tents in Ohio — Amazon took a more deliberate path. Midway through construction, it updated its facility design to speed up deployment.

“It’s not just fast,” said Garman. “It is secure and reliable AWS infrastructure … an industrial, enterprise-scale data center.”

Or, as Garman described it, “Cornfields to data centers, almost overnight.”

The site still feels raw. Workers in safety vests move between trailers as steel beams rise in the distance. Convoys of pickup trucks kick up dust past unfinished warehouse shells. From the security gate, a line of streetlamps stretches toward the data center core, where lifts haul crates packed with chips.

This quiet stretch of rural Indiana, dotted with grain silos, transmission lines, and the occasional barn, has become a magnet for ambitious infrastructure projects. General Motors and Samsung are jointly building a $3.5 billion electric vehicle battery plant next door. At peak, more than 4,000 construction workers have been showing up each day in a town with a population of just 1,900.

Locals don’t necessarily love the trend.

“It’s just difficult to keep losing farmland,” said Marcy Kauffman, president of New Carlisle’s town council. “And this took a lot of farmland.” 

Dan Caruso, a longtime resident of the area, worries that this is just the beginning.

“My friends tried to tell me, ‘You can’t let them come in, because once they get their toe in there, they’ll want more,'” Caruso said. “And that’s exactly what happened.”

Indiana Michigan Power says peak power demand will more than double by the end of the decade, raising questions about household utility bills. One report found that monthly electricity bills in neighborhoods near these new types of sites are 267% higher than five years ago.

And expansion isn’t slowing anytime soon. 

“We’re rapidly adding new capacity all over the place,” Garman said. “I don’t know that we’ll be done ever. We’re going to continue to build as our customers need more capacity.”

Rainier’s seven data center buildings are packed wall-to-wall with Trainium 2, Amazon’s custom-built chips. Nvidia’s market-leading graphics processing units are nowhere to be found. Amazon claims this is the largest known deployment of non-Nvidia compute anywhere in the world.

“They’re already running about 500,000 chips in Indiana today,” Garman said. “And in fact, it’s going so well that they’ve actually doubled down on that order.” Amazon expects the number to reach a million by the end of the year.

Trainium 3, developed in collaboration with Anthropic, is set to launch in the next few months.

It’s the latest example of the tightening bond between the two companies. Anthropic’s primary infrastructure runs on AWS, and it’s one of the first major AI labs to train models on Amazon’s custom silicon. Amazon has invested $8 billion in the startup as part of its broader AI strategy.

While Trainium can’t match Nvidia’s GPUs in raw performance, AWS says its technology offers greater density and efficiency, packing more chips into each data center to deliver higher aggregate compute while reducing power and cooling costs.

Amazon and Anthopic have co-designed silicon based on real-world training demands. Garman and Krieger both told CNBC that Anthropic provided direct input to speed up training, cut latency and improve energy efficiency.

With Trainium 3, one major goal is to better support frontier models.

“It gives better performance, it gives better latency characteristics, it gets better power consumption per flop,” Garman said. “That will be deployed inside of Indiana. It’ll be deployed in many of our other data centers all around the world.”

Prasad Kalyanaraman, vice president of infrastructure services at AWS, said it’s critical to be “able to control the stack all the way from the lower layers of the infrastructure” in order to “build the right set of capabilities that these model providers want.”

Anthropic is moving at a breakneck pace, and burning mounds of cash in the process, as it races to keep up with OpenAI and others.

The company’s annual revenue run rate is nearing $7 billion. Its Claude chatbot powers more than 300,000 businesses, a 300-fold increase over the last two years. The number of large enterprise customers, each producing more than $100,000 in annual revenue, has jumped nearly sevenfold in just a year.

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

But Anthropic isn’t counting exclusively on Amazon as it carves its future path. Last week, the company announced a partnership with Alphabet that gives Anthropic access to up to 1 million of Google’s custom-designed Tensor Processing Units, or TPUs. The deal is worth tens of billions of dollars,

Anthropic had already received funding from Google, and Krieger said the company needs all the processing power it can get.

“There is such demand for our models,” said Krieger, “that I think the only way we would have been able to serve as much as we’ve been able to serve so far this year is this multi-chip strategy.”

Garman is well aware of the multi-cloud and multi-chip efforts, and said Amazon has no plans to do anything drastic, like bidding to buy Anthropic.

“We love the partnership as it is,” he said.

— CNBC’s Katie Tarasov and Erin Black contributed to this report.

WATCH: CNBC’s interview with AWS CEO Matt Garman