Amazon building 30 data centers in Indiana running its custom chips to power Anthropic AI
First there was xAI’s Colossus. Then there was OpenAI’s Stargate. Now Amazon has “Project Rainier.”
Tech companies are racing to build ever larger, more powerful AI data centers and betting hundreds of billions of dollars on the construction.
These massive data centers are filled with powerful GPUs to both train and run AI models, on speculation that there will be enough demand to justify the expense.
More details about Amazon’s plans are emerging, and they’re as large as the mountain they are named after. Amazon is well underway in building 30 huge data centers on one site in Indiana that will use up to 2.2 gigawatts of power, enough to power a million homes, according to a report from The New York Times.
The absurd computing capacity will be used to power AI services from its partner Anthropic, which Amazon has invested $8 billion in.
Unlike most mega-super-jumbo data centers, it won’t be filled with Nvidia’s GPUs, like OpenAI’s Stargate and Meta’s Manhattan-sized data center in Richland Parish, Louisiana.
Instead, Amazon plans on running its own custom Trainium2 chips. While less powerful than Nvidia’s market-leading Blackwell chips, Amazon plans to cram more of the purpose-built chips in its data centers and thinks it can outperform on power and efficiency.
Amazon has said it’s on track to spend $100 billion in capex this year to build out AI infrastructure.
These massive data centers are filled with powerful GPUs to both train and run AI models, on speculation that there will be enough demand to justify the expense.
More details about Amazon’s plans are emerging, and they’re as large as the mountain they are named after. Amazon is well underway in building 30 huge data centers on one site in Indiana that will use up to 2.2 gigawatts of power, enough to power a million homes, according to a report from The New York Times.
The absurd computing capacity will be used to power AI services from its partner Anthropic, which Amazon has invested $8 billion in.
Unlike most mega-super-jumbo data centers, it won’t be filled with Nvidia’s GPUs, like OpenAI’s Stargate and Meta’s Manhattan-sized data center in Richland Parish, Louisiana.
Instead, Amazon plans on running its own custom Trainium2 chips. While less powerful than Nvidia’s market-leading Blackwell chips, Amazon plans to cram more of the purpose-built chips in its data centers and thinks it can outperform on power and efficiency.
Amazon has said it’s on track to spend $100 billion in capex this year to build out AI infrastructure.
