Google is making an ambitious play for a bigger slice of the 21st century’s most important market: the chips that power artificial intelligence. Its financial might and years of technical development put it in prime position to succeed.
One obstacle it has had to contend with: tech companies’ fear of crossing Jensen Huang, Nvidia’s uber-territorial chief executive.
On the southern shore of Lake Ontario, a short drive from Niagara Falls, Google has been demonstrating how it can use Nvidia’s own playbook to win customers.
The site in western New York is host to an AI data-center cluster known as Lake Mariner. Alphabet-owned Google has provided a $3.2 billion financial guarantee for the project, whose developers will rent the computing power from thousands of its microprocessors to AI giant Anthropic, according to people familiar with the matter.
It is the same strategy Nvidia has used time and again to stoke already blazing demand for its own artificial-intelligence chips.
Until recently, Nvidia all but had that market to itself, its graphics processing units, or GPUs, coveted by tech companies for their power to train and run AI models. But as the AI race has morphed over the past year into a contest for computing resources, challengers have begun to edge in—none of them more formidable than Google.
“You have all these very well-capitalized companies who are big believers that this market around compute is going to have tremendous value,” said Nazar Khan, co-founder and chief technology officer of AI infrastructure company TeraWulf, which is developing Lake Mariner with FluidStack, a Google-backed cloud provider. “They want to be in the game, they don’t want to be left behind.”
In corralling customers for its chips—known as tensor processing units, or TPUs—Google has mimicked Nvidia’s practice of using financial guarantees to help data centers raise cheaper debt and providing so-called circular financing, where some of the money it invests flows back in the form of chip purchases.
A shake-up in the leadership of its Cloud unit has increased the level of urgency, people familiar with the matter say. Nvidia has a close partnership with and is a major investor in OpenAI; Google has a similar relationship with Anthropic, as well as its own frontier model, Gemini.
In private and in public, Huang has played down Google’s ability to meaningfully compete with his company.
In April, appearing on podcaster Dwarkesh Patel’s show, Huang said Nvidia enjoys a wide lead over Google and other makers of custom chips, known as ASICs, and argued that Anthropic is Google’s only significant external customer for the TPU.
“Our market reach is far greater than any TPU or ASIC can possibly have,” Huang said. “I would love to hear them demonstrate the cost advantage of TPUs. It makes no sense in my mind.”
In its most direct challenge to date, Google recently struck a $5 billion deal with Blackstone to establish a new cloud-services company that would compete with CoreWeave and Nebius, two Nvidia-backed cloud providers that exclusively use the chip giant’s hardware stack.
“They’re clearly being more opportunistic and more aggressive about monetizing what they have, relative to a few years ago,” said Stacy Rasgon, a tech analyst at Bernstein. “But a few years ago, the opportunity wasn’t there. Today, all we’re hearing is that nobody has enough compute.”
Placing its chips
Google saw the computing crunch coming a long way off, beginning with what one of its top scientists called a “thought experiment.”
In 2013, Jeff Dean was working alongside other artificial-intelligence researchers on speech recognition, using the neural-network technology that underpins today’s large language models.
“I said, ‘OK, if we want to have this speech model that we roll out to 100 million users, and they use it a few minutes a day, that would require doubling the number of computers Google had,” said Dean, who is now chief scientist at Google’s DeepMind AI lab, in an interview. His conclusion: “We need to build specialized hardware.”
At first, the company kept that hardware to itself. It used the chips to develop AI models and features for its search engine and other products.
As demand for chips exploded, the company began making them available to other companies through its Cloud platform. The move has driven rapid growth of that unit.
“Is this the end of Nvidia’s dominance?” asked SemiAnalysis, an influential tech research firm, in a November post tied to the release of Google’s seventh-generation TPU, which Anthropic has used to train its models.
Going direct
The showdown is intensifying. Google in May upped the ante by announcing plans to sell its chips directly to customers. The company also unveiled its first-ever TPU customized for inference, the type of AI computing involved in serving queries. Its product will likely go head-to-head with Nvidia’s new Groq 3 LPU.
Mark Lohmeyer, vice president of AI and computing infrastructure for Google Cloud, said the inference-specialized chip, combined with improvements the company has made in making its chips work across multiple systems, has generated new interest in using TPUs.
“We’re seeing a set of customers that might not have considered it in the past,” he said.
Among them is Citadel Securities, a longtime Google Cloud customer that recently began using TPUs for some of its research software workloads. Josh Woods, the firm’s chief technology officer, said the company can run key workloads at a 30% lower cost and up to four times as fast with TPUs.
Astronomical demand for AI computing has emboldened a host of challengers, including veteran rivals such as Advanced Micro Devices and Broadcom, as well as newer entrants like Cerebras Systems, to take on Nvidia.
Success requires breaking through large customers’ loyalties and Nvidia’s defensive moats. Its plug-and-play connectivity hardware and easy-to-use programming library, known as CUDA, are powerful enticements for AI labs and large enterprise computing partners. Huang is protective of his company’s market share in AI chips, which is estimated at north of 90%, and sensitive about incursions by rivals, according to people familiar with the matter.
Some neo-clouds worry that they can’t stray from buying Nvidia’s full stack of hardware for fear of being put in “Jensen jail,” meaning they might lose their allocations of Nvidia chips, said Adam Fisher, a partner at Bessemer Venture Partners.
“Not all the Nvidia neo-clouds would say it this way—some would say Nvidia gives them what they need—but there are others that are dying for something else, but they can’t get it from another supplier,” Fisher said.
Huang has underscored in public comments that Nvidia welcomes customers buying a la carte.
“Nothing gives me more joy than when you buy everything from Nvidia,” the CEO said at a 2025 conference. “But it gives me tremendous joy if you just buy something from Nvidia.”
Balance sheet heft
Among Nvidia challengers, Google stands alone in the amount of financial firepower it can deploy to pick off customers. The company this month said it plans to raise $85 billion in equity, largely to fund its AI infrastructure needs.
Industry insiders pointed to Google’s deal with Blackstone, which has close ties to both Nvidia and CoreWeave, as a sign of the shifting dynamics created by the computing shortage. As recently as a year ago, they said, such a deal would have been unthinkable, because companies were nervous about angering Nvidia’s Huang.
“Anyone not named Nvidia probably has to spend more from their balance sheet to break in,” said TeraWulf’s Khan.
Google is backstopping another Anthropic deal, a $7 billion project known as River Bend, near Baton Rouge, La. And in Colorado City, Texas, Google is providing an additional $1.4 billion in financial guarantees for an AI computing lease.
Much of the change in approach comes under the leadership of Amin Vahdat, who in December was promoted to the position of chief technologist in charge of Google’s AI infrastructure build-out. The promotion expanded Vahdat’s portfolio to include chip design, supply and deployment. He now reports to both Thomas Kurian, head of Google Cloud, and Alphabet Chief Executive Sundar Pichai.
People who have worked with Vahdat say that as a boss, he demands excellence and has a quiet competitive streak. In 2021, Google raided Intel’s top talent in Israel, hiring 25-year Intel veteran Uri Frank to lead Google’s silicon efforts.
Current and former employees say Google heightened its focus on the commercial potential for its TPUs about two years ago, in particular by investing in their inference capabilities.
Since Vahdat’s promotion, Google’s AI infrastructure team has been operating with more urgency, some of the people said. One current employee said Vahdat is hyperfocused on improving chip performance, often challenging engineers to continually improve various functions by 10%—a difficult margin to achieve.
Vahdat said in an interview that he is not focused on competing with Nvidia or any other rival in particular, and said the chip giant is a key partner as well as a competitor because Google uses Nvidia GPUs in its data centers. His focus, he said, is simply making better products for Google and its customers.
“For me and for us, it’s not zero-sum,” Vahdat said. “There’s so much demand out there.”
Write to Robbie Whelan at robbie.whelan@wsj.com and Katherine Blunt at katherine.blunt@wsj.com