Google has expanded its lineup of TPU AI chips to compete with Nvidia

Google, a subsidiary of Alphabet, unveiled two new versions of its Tensor Processing Units (TPUs)—the 8tx and 8i Max. The company has expanded its lineup of specialized AI chips, which is challenging Nvidia—the leader in a market worth hundreds of billions of dollars— according to Bloomberg. Google’s new products have arrived at an extremely difficult time for Nvidia, which has been embroiled in a scandal surrounding CEO Jensen Huang’s affair with an alleged Chinese spy.

New TPU-based solutions are optimized for various tasks, Amin Vahdat, head of chip development at Google, said in an interview with Bloomberg. “This is fully in line with our philosophy here at Google: the future lies in specialization, and the benefits of it continue to be significant.”

The new TPU 8tx and 8i Max variants are not fundamentally new designs—rather, they have been reconfigured to meet the needs requested by customers, Vahdat explained.

The 8tx is designed for secure data processing for clients such as government agencies and companies in regulated industries, and works in conjunction with Advanced Micro Devices (AMD) processors as part of a partnership between the two companies. The 8i Max, in turn, combines two TPUs in a single package, providing more computing power and memory for tasks with long context—such as drug discovery, mathematical computations, and programming.

Google’s new status as a chip vendor is proving to be quite lucrative, Bloomberg notes. According to Bloomberg Intelligence, TPU sales will exceed $135 billion by 2028 (including related software and hardware), up from approximately $12 billion this year. This figure does not include customers using the chips through Google Cloud. TPUs cost about 30% less than Nvidia graphics processing units (GPUs) while offering comparable performance in terms of data processing volume, noted Bloomberg Intelligence analyst Mandip Singh.

TPU chips have already attracted a number of major customers, including Apple, Meta Platforms, and Oracle. Anthropic, the developer of the Claude chatbot, has an option to purchase up to $150 billion worth of TPUs between 2028 and 2031. OpenAI also plans to purchase TPUs, according to Bloomberg News.

In October, Oracle signed an agreement that will allow its cloud customers to use TPUs for the first time starting next year—previously, they had only been offered Nvidia and AMD chips. “We began this partnership with Google because customers need more choice as AI workloads become more complex,” said Karen Batta, senior vice president at Oracle.

Alphabet is aiming to capture a share of the chip market, which is currently led by Nvidia. Its management will likely be distracted in the coming weeks by the scandal surrounding CEO Jensen Huang, suggested Seaport Research analyst Jay Goldberg—the only “bear” on Wall Street regarding Nvidia stock, who advises selling them. He believes the situation surrounding Huang’s affair will become a “long-lasting” problem for the company.

Google is facing a shortage of key components—memory and manufacturing capacity: all available TPUs have already been contracted by customers, the company said. To ensure supply, Google is diversifying its suppliers beyond its main partner, Broadcom. In recent months, it has signed an agreement with Broadcom’s competitor, Marvell Technology, which includes some work on TPUs, although details have not been disclosed. It has also been reported that Taiwanese semiconductor developer MediaTek is fulfilling orders for Google’s chip division. Vahdat declined to comment on the roles of both companies.

Google is likely securing long-term agreements to guarantee supply and avoid bottlenecks, according to Bloomberg Intelligence analyst William Tong.

Competition is also intensifying from new players. The startup Tenspresso, founded by two former TPU engineers, has focused its first chip on inference with an expanded context window—exactly the same area that the 8i Max targets. OpenAI and Anthropic are developing their own chips, also drawing on veterans from the TPU team.

Among other risks is a potential reduction in spending on AI infrastructure: A negative reaction to data center construction threatens to cut the $1.4 trillion that tech companies planned to spend in 2027 by approximately 5–7%, according to estimates by Singh of Bloomberg. This applies to chips, networking equipment, land, and construction materials. The analyst notes that Google’s TPU business has lower profit margins than Nvidia’s.

Google’s announcement marks an acceleration in the pace of development within its chip division: previously, the company released a new processor every two years, but this year it unveiled two at once, Bloomberg notes. The release comes amid turbulence in the AI chip market—investors are concerned about a possible slowdown in infrastructure spending, and the U.S. midterm elections, which resulted in a victory for the Democrats, have heightened doubts about the prospects for data centers. At the same time, demand for Nvidia processors continues to outstrip supply, leaving room for Google and other competitors to grow their businesses, the agency notes.

This article was AI-translated and verified by a human editor