Google launches TPU 8tx and 8i Max, targeting confidential computing and long-context workloads to challenge Nvidia
According to Bloomberg, Google parent Alphabet is rolling out two dedicated AI chips — TPU 8tx and TPU 8i Max — as an expansion of its eighth-generation TPU product lineup, targeting secure computing and advanced AI workloads in a bid to challenge Nvidia. Google officially announced on its blog on November 9 that the two systems are system-level configurations based on the existing TPU 8t and 8i, rather than entirely new chip designs, making 2026 the first year in which it will launch four TPU systems in a single year.
TPU 8tx is based on the TPU 8t silicon and paired with a fifth-generation AMD EPYC CPU host, aimed at regulated enterprises and government agencies, and provides hardware-level confidential computing so that data and model weights remain encrypted and isolated during processing, making it suitable for confidential inference and small-scale training. TPU 8i Max packages two TPU 8i chips into a single package, providing double the computing power, double the SRAM and double the HBM bandwidth while maintaining the existing rack footprint, making it suitable for agentic workloads that need to handle ultra-long context windows, such as complete code repositories or hours of video.
Bloomberg Intelligence estimates that Google TPU-related sales could surge from about $12 billion in 2026 to more than $135 billion in 2028, including related software and hardware but excluding the portion rented out through Google Cloud. Major customers include Apple, Meta, Oracle and Anthropic; OpenAI is also reported to be planning purchases. Bloomberg Intelligence analysts said TPU costs about 30% less than Nvidia GPUs for certain tasks and offers competitive throughput and higher energy efficiency.
Original source:
Title: Google Escalates Nvidia Rivalry With New Versions of Its TPU Systems
Publisher: Bloomberg
Publication date: 2026-11-09
URL: https://www.bloomberg.com/news/articles/2026-11-09/google-escalates-nvidia-rivalry-with-new-versions-of-its-tpu-systems
Source type: News report
Verification: Cross-checked with Google’s official blog and multiple republications, confirming that the chip names, positioning, sales forecast, customer list and the roughly 30% cost advantage statement are consistent with the Bloomberg report.
Supporting content: Chip launches, challenge to Nvidia, sales forecast, major customers and cost comparison.
ABAB AI Insight
Google is rapidly expanding its TPU product line through system-level configurations rather than new silicon, covering the two emerging needs of confidential computing and long-context workloads and reinforcing its positioning as an Nvidia alternative. The sales forecast rising from about $12 billion to more than $135 billion is a Bloomberg Intelligence estimate and excludes cloud rentals; actual execution depends on supply and customer contracting progress.
The roughly 30% cost advantage also comes from analyst views and is limited to comparisons on specific tasks and total cost of ownership, rather than being an across-the-board lead. In the customer list, Apple, Meta, Oracle and Anthropic already show public signs of adoption, while OpenAI purchases remain at the reporting level. Overall, this is another step in Google’s move from internal use toward external sales, but supply constraints and software ecosystem maturity remain key variables.
ABAB News · Cognitive Law
AI chip competition is shifting from general-purpose computing power to scenario-specific configurations, and cost and energy-efficiency advantages need to be verified against specific workloads.