Google’s decade-long bet on TPUs company’s secret weapon in AI race

Nvidia It has established itself as the undisputed leader in AI chips, selling large quantities of silicon to many of the world’s largest technology companies on its way to a market cap of $4.5 trillion.

One of Nvidia’s important customers GoogleIn the cloud, AI is offloaded to the chipmaker’s graphics processing units, or GPUs, to keep up with the growing demand for computing power.

While there’s no sign that Google will slow down its purchases of Nvidia GPUs, the internet giant is increasingly showing that it’s not just a buyer of high-power silicon. He is also a developer.

On Thursday, Google announced that its most powerful chip, called Ironwood, will be widely available in the coming weeks. This is the seventh generation of the Tensor Processing Unit (TPU), Google’s custom silicon that it has been working on for over a decade.

TPUs are application-specific integrated circuits, or ASICs, that play a key role in artificial intelligence by providing highly specialized and efficient hardware for specific tasks. Google says Ironwood is designed to handle the heaviest AI workloads, from training large models to powering real-time chatbots and AI agents, and is four times faster than its predecessor. AI startup Anthropic plans to use up to 1 million of these to power its Claude model.

At a time when all hyperscalers are racing to build giant data centers and AI processors can’t be produced fast enough to keep up with demand, TPUs offer a competitive advantage for Google. Other cloud companies are taking a similar approach, but their efforts lag well behind.

Amazon Web Services made its first cloud AI chip, Inferentia, available to customers in 2019, followed by Trainium three years later. Microsoft won’t announce its first dedicated AI chip, Maia, until the end of 2023.

“Among the ASIC players, Google is the only one using these things in really large volumes,” said Stacy Rasgon, an analyst who covers semiconductors at Bernstein. “For the other big players, this takes a long time, a lot of effort, and a lot of money. They are the furthest along among other hyperscalers.”

Originally trained for internal workloads, Google’s TPUs have been available to cloud customers since 2018. Recently Nvidia has shown some concern. When OpenAI signed its first cloud contract with Google earlier this year, the announcement spurred Nvidia CEO Jensen Huang to initiate further conversations with the AI ​​startup and its CEO Sam Altman. Wall StreetJournal.

Unlike Nvidia, Google doesn’t sell its chips as hardware; instead, it provides access to TPUs as a service via the cloud, which has emerged as one of the company’s biggest growth drivers. Google parent Alphabet said in its third-quarter earnings report last week that cloud revenue rose 34% from a year earlier to $15.15 billion, beating analyst estimates. The company ended the quarter with a backlog of $155 billion.

“We are seeing significant demand for our AI infrastructure products, including TPU-based and GPU-based solutions,” CEO Sundar Pichai said on the earnings call. “That’s been one of the key drivers of our growth over the past year and I think we continue to see very strong demand going forward and we’re investing to meet that.”

Google does not disclose the size of its TPU business in the cloud segment. Analysts at DA Davidson estimated in September that the value of a “standalone” business consisting of TPUs and Google’s DeepMind AI division could rise to about $900 billion, up from $717 billion in January. Alphabet’s current market cap is more than $3.4 trillion.

The company has seen increased demand for TPUs alongside Nvidia’s processors in its cloud business and is increasing GPU consumption “to meet significant customer demand,” a Google spokesperson said in a statement.

“Our approach is selection and synergy, not replacement,” the spokesman said.

Tuesday on Google announced The Suncatcher Project explores “how an interconnected network of solar-powered satellites equipped with our Tensor Processing Unit (TPU) AI chips can harness the full power of the Sun.”

As part of the project, Google said it plans to launch two prototype solar-powered satellites carrying TPU in early 2027.

“This approach will have tremendous scale potential while minimizing impact on terrestrial resources,” the company said in its announcement. “This will test our hardware in orbit and lay the foundation for the future era of large-scale computing in space.”

Google’s largest TPU deal on record occurred late last month, when the company announced a major expansion of its deal with OpenAI rival Anthropic, worth tens of billions of dollars. With the partnership, Google is expected to bring over one gigawatt of AI computing capacity online by 2026.

“Anthropic’s choice to significantly expand TPU usage reflects the strong price-performance and efficiency their teams have seen in TPUs for several years,” Google Cloud CEO Thomas Kurian said at the time of the announcement.

Google invested $3 billion in Anthropic. While Amazon remains Anthropic’s deepest embedded cloud partner, Google now provides the underlying infrastructure to support next-generation Claude models.

“There’s such demand for our models that I think the only way we can serve as much as possible this year is with this multi-chip strategy,” Anthropic Chief Product Officer Mike Krieger told CNBC.

This strategy covers TPUs, Amazon Trainium, and Nvidia GPUs, allowing the company to optimize for cost, performance, and redundancy. Krieger said Anthropic has done a lot of preliminary work to ensure its models can work equally well across silicon providers.

“I’ve seen the investment pay off because we can go online with these massive data centers and meet our customers where they are,” Krieger said.

Two months before Anthropic deal, Google signs six-year cloud deal Meta is worth more than $10 billion, but it’s unclear how much of the regulation involves the use of TPU. Although OpenAI said it would move away from Microsoft and start using Google’s cloud, the company said: Reuters It does not distribute GPUs.

Alphabet CFO Anat Ashkenazi attributed Google’s cloud momentum in the latest quarter to growing enterprise demand for Google’s entire AI stack. The company said it signed more billion-dollar cloud deals in the first nine months of 2025 than in the previous two years combined.

“On GCP, we are seeing strong demand for enterprise AI infrastructure, including TPUs and GPUs,” Ashkenazi said, adding that users are also flocking to services “like cybersecurity and data analytics” as well as the company’s latest Gemini offerings.

Similar sentiments are expressed by Amazon, which reported 20% growth in its market-leading cloud infrastructure business last quarter.

AWS CEO Matt Garman told CNBC in a recent interview that the company’s Trainium chip line is gaining momentum. “Every Trainium 2 chip we place in our data centers today is sold and used,” he said, promising further performance gains and efficiency gains with Trainium 3.

Shareholders have shown they are willing to take on large investments.

Google raised the upper end of its capital spending forecast for this year to $93 billion, from the previous estimate of $85 billion, expecting an even steeper rise in 2026. The stock price rose 38% in the third quarter, its best performance of any period in the last 20 years, and rose another 17% in the fourth quarter.

Mizuho recently noted Google’s clear cost and performance advantage with TPUs, noting that the chips were originally built for in-house use, but Google is now gaining external customers and larger workloads.

Nvidia’s GPUs will remain the dominant chip provider in AI, but developers’ increasing familiarity with TPUs could be a meaningful driver of Google Cloud’s growth, Morgan Stanley analysts wrote in a June report.

Analysts at DA Davidson said in September they were seeing so much demand for TPUs that Google should consider selling the systems to “external customers,” including leading AI labs.

“We continue to believe that Google’s TPUs remain the best alternative to Nvidia, with the gap between the two narrowing significantly over the last 9-12 months,” they wrote. “We have seen increased positive sentiment around TPUs during this time.”

WRISTWATCH: Amazon’s $11 billion data center goes live: Here’s an inside look