Google Cloud AI Chips are taking a major step forward as the company unveiled two new custom processors designed to strengthen its position in the fast-growing artificial intelligence infrastructure market.
At its latest event, Google Cloud announced the eighth generation of its tensor processing units, or TPUs, split into two specialized chips. The TPU 8t is built for model training, while the TPU 8i is focused on inference — the stage where AI models respond after users submit prompts.
Google Cloud AI Chips Split Training and Inference Workloads
The biggest change in the Google Cloud AI Chips strategy is specialization.
Instead of one chip handling every task, Google is separating training and inference into two dedicated processors. The TPU 8t is optimized for training large models, while the TPU 8i is designed for real-time usage and deployment workloads.
That approach reflects how AI infrastructure is evolving, with different stages of the model lifecycle requiring different hardware priorities.
Performance Claims Highlight Efficiency Gains
Google also shared several performance improvements for its new TPUs compared with earlier generations.
According to the company, the chips can deliver up to three times faster model training, 80% better performance per dollar, and the ability to connect more than one million TPUs in a single cluster.
The broader message is clear: more computing power, lower energy use, and reduced costs for cloud customers.
Nvidia Still Remains in the Picture
Despite the launch, Google Cloud AI Chips are not replacing Nvidia outright.
Like other hyperscalers such as Microsoft and Amazon, Google is using its custom chips alongside Nvidia hardware rather than removing it from the stack.
Google also confirmed that Nvidia’s Vera Rubin chip will be available in its cloud later this year, showing that both ecosystems will continue to coexist.
A Long-Term Competitive Shift
Over time, cloud giants building their own AI silicon could reduce reliance on Nvidia as more enterprise workloads move into cloud platforms.
Still, Nvidia remains dominant. Industry analyst Patrick Moore noted on X that predictions from 2016 suggesting Google’s first TPU launch would seriously hurt Nvidia did not materialize. Since then, Nvidia has grown into a company valued at nearly $5 trillion.
For now, demand for AI infrastructure appears large enough to support both custom cloud chips and Nvidia’s products.
Google and Nvidia Are Also Collaborating
Interestingly, competition is only part of the story.
Google said it is also working with Nvidia to improve networking performance for Nvidia-based systems running in Google Cloud. The companies are enhancing Falcon, Google’s software-defined networking technology that was open-sourced in 2023 through the Open Compute Project.
That means while Google Cloud AI Chips compete in one area, the two tech giants are still partnering in another.
Why This Matters
Google Cloud AI Chips show how the AI hardware race is becoming more nuanced than a simple winner-takes-all battle.
Custom chips, cloud infrastructure, and strategic partnerships are all shaping the future of enterprise AI computing — and Google wants to be at the center of it.
Don’t miss out on our latest news—follow us for the latest AI news, breakthroughs