Google Unveils TPU 8T and 8i, Challenging Nvidia’s 50% Data Center Revenue
Google introduced the TPU 8T for AI model training and TPU 8i for model inference, both boasting improved power efficiency and targeted performance. These ASIC chips are being marketed to hyperscalers and AI firms like Anthropic, OpenAI and Meta, intensifying competition with Nvidia and spotlighting Microsoft’s compute capacity constraints.
1. Google’s New TPU 8T and 8i Chips
Google has introduced the TPU 8T, optimized for AI model training, and the TPU 8i, designed for high-throughput inference. Both ASICs offer increased power efficiency and specialized processing, marking Google’s shift from unified designs to task-specific processors.
2. Hyperscaler Adoption and Nvidia Implications
These chips will be offered to major cloud and AI providers, including Anthropic, OpenAI and Meta, mirroring Amazon’s Trainium push. Hyperscalers account for roughly 50% of Nvidia’s data center revenue, so a move to in-house silicon raises competitive pressure.
3. Microsoft’s Capacity Constraints and Stock Impact
Microsoft has flagged compute capacity constraints on recent earnings calls, which have limited its AI model training throughput and weighed on Co-Pilot product adoption. Increased competition for AI hardware could further strain Microsoft’s access to high-performance accelerators.