Google Announces 8th Gen TPU, New AI Agents As It Looks To Take On Nvidia, OpenAI, Anthropic

• The new TPU 8t chip is designed for training large language models, with Google stating that it reduces the frontier development cycle from months to weeks.
• The TPU 8i chip is built for inference and reinforcement learning, with Google saying it’s designed to support large numbers of specialized agents working together to generate user responses.
• Google also announced a slew of partnerships with NASA, Bosch, The Home Depot, KPMG, PepsiCo, and Unilever, among others.

Alphabet Inc.’s (GOOG, GOOGL) Google on Wednesday announced a slew of new launches, including the eighth generation of its Tensor Processing Units (TPU) in its latest attempt to take on Nvidia Corp. (NVDA).

Google also announced a slew of partnerships with NASA, Bosch, The Home Depot, KPMG, PepsiCo, and Unilever, among others. Google stated that these companies have deployed Gemini Enterprise to build and deploy agents on their respective platforms.

Alphabet’s Class A shares were up nearly 2% in Wednesday’s opening trade. Retail sentiment on Stocktwits around the company trended in the ‘neutral’ territory, with message volumes at ‘neutral’ levels at the time of writing.

Two New Architectures For Training And Inference

Google stated that its new eighth-generation TPU is split into two distinct architectures, each purpose-built for training and inference.

The new TPU 8t chip is designed for training large language models, with Google stating that it reduces the frontier development cycle from months to weeks. The company added that TPU 8t delivers nearly three times the compute performance per pod over the previous generation.

The TPU 8i chip is designed for inference and reinforcement learning. Google stated that TPU 8i has been developed to handle the work of many specialized agents that often swarm together to deliver answers to users.

Google added that TPU 8i delivers 80% better performance per dollar than the previous generation.

The launch of TPU 8i comes amid a rise in agentic AI, with rivals OpenAI and Anthropic offering software development kits (SDKs) for building and deploying agents.

Nvidia CEO Jensen Huang also urged organizations to adopt the technology. “For the first time, AI is producing work. It is collaborating with me and our engineers and supporting us in different ways. That's another way of saying AI is finally valuable,” Huang said during a fireside chat with Adobe CEO Shantanu Narayen on Monday.

Taking On Nvidia

Google has continued to develop its own TPUs for machine learning and training large language models, even though it remains one of Nvidia's customers.

While Google is not directly comparing the performance and efficiency of its new eighth-generation TPU with Nvidia’s offerings, the company is offering this chip as an alternative for organizations using its cloud services.

Google’s launch of TPU 8i also comes months after Nvidia talked about upcoming chip technology that will enable large language models to respond faster to user questions. The company later unveiled Nvidia Groq 3 LPX, describing it as an inference accelerator for its Vera Rubin chip aimed at enabling real-time agentic inference.

GOOGL stock is up 8% year-to-date, while NVDA stock is up over 7%.

The S&P 500 ETF (SPY) is up 35% over the past 12 months, while the Invesco QQQ Trust ETF (QQQ) is up 47%.

Also See: Tesla Needs ‘Way More’ Spending But Cash Is Tight, Paul Meeks Warns Ahead Of Q1 Earnings Print

For updates and corrections, email newsroom[at]stocktwits[dot]com.