Why Anthropic Uses Google Cloud TPUs for AI Infrastructure

The scramble for computing power has become one of the AI industry’s most pressing challenges – and Anthropic isn’t taking any chances. 

The company behind the AI Claude chatbot is making a bet on Google Cloud, committing to deploy up to one million tensor processing units (TPUs) in a deal worth tens of billions of dollars.

It’s a massive infrastructure play that reflects just how quickly things are moving. 

Anthropic now serves more than 300,000 business customers, with its biggest accounts – those paying over US$100,000 annually – growing nearly seven times in just the past year. 

The expansion, set to bring more than 1GW of capacity online in 2026, is essentially about keeping pace with demand that shows no signs of slowing.

How TPUs are fixing the global AI chip shortage

TPUs are Google’s answer to the chip shortage plaguing AI development. 

Unlike general-purpose processors, these chips are built specifically for the matrix multiplications that power neural networks, making them faster and more efficient for training and running large language models. 

Google is now on its seventh generation, codenamed Ironwood.

“Anthropic’s choice to significantly expand its usage of TPUs reflects the strong price-performance and efficiency its teams have seen with TPUs for several years,” says Thomas Kurian, CEO of Google Cloud. 

“We are continuing to innovate and drive further efficiencies and increased capacity of our TPUs, building on our already mature AI accelerator portfolio, including our seventh generation TPU, Ironwood.”

The partnership between Anthropic and Google Cloud stretches back several years, during which the AI company has been evaluating TPU performance for its workloads. 

This latest expansion is a vote of confidence in the technology after that extended testing period.

How Anthropic maintains a diversified approach across three chip platforms

Despite the sizeable Google commitment, Anthropic isn’t putting all its eggs in one basket.

The San Francisco-based firm runs what it calls a diversified compute strategy across three chip platforms: Google’s TPUs, Amazon’s Trainium processors and Nvidia’s graphics processing units (GPUs).

This means that it is a pragmatic approach that’s become increasingly common in an industry where computing capacity is both scarce and critical. 

By spreading workloads across multiple providers, Anthropic protects itself against supply constraints while maintaining leverage in negotiations.

The computing resources will do more than just serve customer requests. 

Anthropic says the expansion will support more rigorous testing, alignment research focused on keeping AI systems behaving as intended – and what it terms responsible deployment at scale. 

These safety considerations have been central to the company’s identity since former OpenAI executives founded it.

“Anthropic and Google have a longstanding partnership and this latest expansion will help us continue to grow the compute we need to define the frontier of AI,” says Krishna Rao, Anthropic’s Chief Financial Officer (CFO). 

“Our customers – from Fortune 500 companies to AI-native startups – depend on Claude for their most important work and this expanded capacity ensures we can meet our exponentially growing demand while keeping our models at the cutting edge of the industry.”

Why AWS remains primary training partner

Amazon Web Services (AWS) remains Anthropic’s primary training partner, the company is keen to stress. 

Work continues on Project Rainier, a sprawling compute cluster with hundreds of thousands of AI chips spread across multiple US data centres. 

The setup means Anthropic is simultaneously deepening ties with Google – which also holds an equity stake from 2023 – while maintaining its core partnership with Amazon.

The 1GW of planned capacity is worth putting in perspective – because that’s enough electricity to power a small city, all dedicated to training and running AI models. 

The energy intensity of these operations is becoming harder to ignore as the industry scales up, with data centres requiring substantial power not just for processors but for the cooling systems needed to prevent hardware failure.

“Anthropic will continue to invest in additional compute capacity to ensure our models and capabilities remain at the frontier,” Krishna says.