Google Cloud launches two new AI chips to compete with Nvidia

2026-04-22 · Source: TechCrunch · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Emerging Technologies & Innovation · Depth: Intermediate, quick

Summary

Google Cloud has unveiled its eighth generation of custom-built AI chips, Tensor Processing Units (TPUs), which are now specialized into two distinct types: the TPU 8t for model training and the TPU 8i for inference. These new TPUs offer significant performance improvements over previous generations, including up to 3x faster AI model training and 80% better performance per dollar. They also support massive scalability, allowing over 1 million TPUs to operate together in a single cluster, aiming to deliver more compute power with reduced energy consumption and cost. Despite these advancements, Google Cloud continues to integrate Nvidia's latest chips, such as the Vera Rubin, into its infrastructure and is collaborating with Nvidia to enhance networking efficiency for Nvidia-based systems using Google's Falcon technology.

Key takeaway

For CTOs and VP of Engineering evaluating cloud AI infrastructure, Google Cloud's new TPU 8t and 8i offer compelling performance and cost efficiencies for both training and inference. Your teams should consider these specialized TPUs for demanding AI workloads, while also noting Google's continued integration of Nvidia's latest chips and collaborative networking enhancements, which could further optimize hybrid AI deployments.

Key insights

Google Cloud's new specialized TPUs enhance AI training and inference while complementing, not replacing, Nvidia's hardware.

Principles

• Specialization improves performance
• Scalability drives efficiency

Method

Google is enhancing cloud AI infrastructure by developing specialized TPUs for training and inference, and by collaborating with Nvidia on networking improvements like Falcon.

In practice

• Utilize TPU 8t for large-scale model training
• Deploy TPU 8i for efficient AI inference workloads

Topics

• Google Cloud
• AI Chips
• Tensor Processing Units
• NVIDIA
• AI Model Training

Best for: CTO, VP of Engineering/Data, MLOps Engineer, AI Architect, Director of AI/ML, Tech Journalist

Related on AIssential

• Google Releases Two New AI Chips — Google's new TPU 8t and 8i chips specialize in AI training and inference, aiming for efficiency and cost reduction.
• Google Cloud Debuts New AI Chips — AI compute capacity and proprietary silicon are critical differentiators in the competitive cloud and AI model landscape.
• Google to Release New AI Chips, Challenging Nvidia — Google is specializing its TPU program for AI inference to meet surging demand and maintain its competitive edge.
• Google doesn't pay the Nvidia tax. Its new TPUs explain why. — Google's new specialized TPUs and vertical integration aim to reduce AI compute costs and enhance performance.
• Here’s how our TPUs power increasingly demanding AI workloads. — TPUs are Google's custom chips optimized for high-speed AI model computation.
• Google's TPU AI Compute Arbitrage — Google Cloud uses custom TPUs to create a structural cost advantage, funding competitive GPU offerings.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by TechCrunch.