Google expands TPU lineup to press Nvidia on AI chips

What Google launched and how it is positioning it

Google is expanding its custom AI chip roster with two new tensor processing unit variants as it seeks to challenge Nvidia in an industry valued in the hundreds of billions of dollars. The 8tx and 8i Max are reconfigured takes on Google's latest TPUs, built to match customer requests rather than brand new designs, said chip chief Amin Vahdat.

The 8tx targets secure, confidential jobs for government agencies and regulated industries. As part of a collaboration between the companies, it operates alongside processors from Advanced Micro Devices and keeps data encrypted at the chip level so it is not exposed in memory while being processed. The 8i Max is tuned for workloads that need long context windows, which matters for drug discovery, math, and coding, and it integrates a pair of TPU chips into one package, boosting compute capacity and memory. Google said these specialized variants will be available next year.

Vahdat said Google now has four TPU offerings aimed at different needs. In his words, "the future is going to be about specialization," and focused versions can deliver "much more value." The move also aligns with where the industry is heading, toward AI agents and tighter security.

Faster cadence, hotter competition

Google's chip team is moving quicker. It used to unveil a new AI processor roughly every two years. This year it released two at once, and the latest additions, due to be announced Monday, extend that family with more specialized options.

The timing lands in a complicated backdrop. Worries that infrastructure spending could decelerate have weighed on tech shares, and following US midterms that put Democrats in charge, new doubts around data center projects have emerged. Nvidia, meanwhile, is contending with upheaval following Chief Executive Officer Jensen Huang's acknowledgment of a relationship with a person accused of spying for China. Even so, demand for Nvidia's GPUs still exceeds supply, leaving room for rivals to win business.

Nvidia, long a proponent of general purpose GPUs, is also leaning into specialization. It now sells its own CPUs, a category pioneered by Intel, to help manage AI agents, and it is offering chips more focused on inference using technology acquired from Groq.

Where TPUs stand in the market

Google began developing TPUs more than a decade ago with Broadcom, a partner that assisted in the design and in coordinating production. Originally meant to support Google's own growth, the chips proved effective for AI and are now sold beyond Google Cloud. Bloomberg Intelligence estimates TPU sales, including related software and hardware, at about $12 billion this year, rising to more than $135 billion by 2028, excluding revenue from customers who access TPUs via Google Cloud.

Price for performance is a selling point, said Bloomberg Intelligence's Mandeep Singh, adding that TPUs are roughly 30% cheaper than Nvidia's GPUs while still providing competitive throughput - how much data can be processed over time. That pitch has attracted major names including Apple, Meta Platforms, and Oracle. Oracle said last month it will, for the first time, make TPUs available to its cloud customers starting next year. Karan Batta - a senior vice president who oversees Oracle's cloud infrastructure - said, "We partnered with Google because customers need more choice as AI workloads become more demanding." "This partnership expands our ability to help customers build and run AI on the infrastructure that best fits their needs."

New technologies quietly reshape how investors spot opportunities and manage long term risk. We're hosting our first ever investor livestream on 11/17 to show you how to use ChatGPT and Claude to invest like a pro, hosted by Briefs Finance CEO Jaspreet Singh. Save your seat.

Google also named Jump Trading as a TPU customer during its latest earnings report. Jump CTO Alex Davies said the firm tested TPUs for several years before signing a larger deal about a year ago. The company now has thousands of TPUs to train models used in trading research.

Nvidia remains its main training platform, but a "vocal minority" really likes TPUs for their tasks, he said. Where TPUs match or beat Nvidia, they can be more power efficient, helping to lower costs. "It's actually attractive to have some problems being done differently," he said.

"You learn stuff."

According to a filing, Anthropic PBC - the company that created the Claude chatbot - holds an option to purchase up to $150 billion of TPUs between 2028 and 2031, and Bloomberg News has also reported that OpenAI intends to buy TPUs. Google's own Gemini models gave TPU developers early access to improve power efficiency and catch bugs, which Vahdat said reduces the odds of costly chip design errors.

Constraints, risks, and the investor angle

Supply remains tight. Google said every TPU it has available is already contracted, reflecting shortages of memory and limited manufacturing capacity across AI infrastructure. Profit margins on TPUs are thinner than Nvidia's, according to Singh.

New rivals are forming too. Tenspresso, co founded by two former TPU engineers, is building a chip for long-context inference, the same area 8i Max targets, but Bloomberg reported it is at least a year from finishing its design. OpenAI and Anthropic are also building in-house chip programs staffed with TPU veterans.

Macro headwinds could bite. Singh estimates backlash against data centers could cut the roughly $1.4 trillion tech companies were expected to spend by about 5% to 7% in 2027, including outlays for chips, networking, land, and other data center needs. The bottom line: Google is leaning into specialization with 8tx for confidential computing and 8i Max for long-context work, while demand across the market remains larger than Nvidia alone can fulfill. Alphabet shares climbed as much as 4.1% to $296.33 on Monday.

Being curious about emerging tools can give investors a thoughtful advantage over time. On 11/17, our CEO Jaspreet Singh goes live for our first ever investor livestream on how to use ChatGPT and Claude to invest like a pro. Register to join him.