Google Unveils TPU 8tx and TPU 8i Max, Its Newest AI Chips to Challenge Nvidia

Published November 10, 2026 | Tech & AI | TrendWire Daily

Google is escalating its challenge to Nvidia. On Monday, November 9, 2026, the company introduced two new AI systems — TPU 8tx and TPU 8i Max — expanding its eighth-generation Tensor Processing Unit lineup to four systems in a single year, the first time Google has ever moved that fast with its custom chips.

What Google Announced

Direct answer: Google unveiled the TPU 8tx and TPU 8i Max on November 9, 2026. The 8tx is built for secure, confidential AI inference, keeping data encrypted even while it is processed, while the 8i Max packs two TPU 8i chips into one package for AI workloads that need massive context windows. Both are specialized variants of existing eighth-generation silicon, not brand-new chips, and Bloomberg reported they are due to become available in 2027.

Google introduced the first two systems of this generation — TPU 8t for large-scale training and TPU 8i for real-time reasoning — at Google Cloud Next in April 2026. Google says the 8t offers up to 2.7 times better training performance than the previous generation, and the 8i offers 80 percent better inference price performance. The two new systems build on that foundation, targeting workloads the first pair did not cover.

Amin Vahdat, Google's senior vice president and chief technologist for AI infrastructure, wrote in Google's official blog post that the company has "always co-designed our TPUs in partnership with Google DeepMind" and is now "increasingly getting feedback from external organizations on their specific needs." That shift — from building chips for Google's own use to building them for outside customers — is the heart of the story.

TPU 8tx: Confidential Computing for Regulated Industries

The TPU 8tx is Google's third eighth-generation system, optimized for secure and confidential inference and also suited to small-scale training. It uses the same silicon as the TPU 8t, paired with 5th Generation AMD EPYC host processors, and builds protection into the hardware so data and model weights stay isolated while in use.

The target customers are banks, governments and other regulated organizations that need to run AI on sensitive data — health records, financial transactions, classified material — without exposing it. Google says the 8tx borrows the Boardfly network from the TPU 8i and delivers up to 2 times lower latency than the TPU 8t, with a near-zero performance penalty from the security layer. Google has not published independent security audits of those claims, and no third party has yet tested the system.

TPU 8i Max: Double the Memory for Huge Context Windows

The TPU 8i Max attacks a different problem: the memory wall. As AI models handle longer conversations, bigger codebases and more complex reasoning, they need to keep more information close to the processor. Google says the 8i Max scales compute and memory in lockstep, delivering double the compute, double the SRAM and double the HBM bandwidth of the TPU 8i.

In concrete terms, that means 768 MB of on-chip SRAM, 17.2 TB/s of memory bandwidth and 384 GB of HBM memory — about a third more memory than the TPU 8i — according to trade-press analysis of Google's figures. The target workloads are AI agents, coding, drug discovery and complex mathematics: jobs that need huge context windows and the highest compute density per rack. All performance figures are Google's own claims, not independent benchmarks.

The Nvidia Challenge — and the OpenAI Question

Bloomberg first reported the launch and framed it as Google stepping up its rivalry with Nvidia, which still dominates the market for AI accelerators. The economics help explain why customers are listening: Bloomberg Intelligence analyst Mandeep Singh said Google's TPUs cost about 30 percent less than Nvidia's GPUs while delivering competitive throughput. Bloomberg Intelligence estimates Google's TPU-related sales, including hardware and software, could exceed $135 billion by 2028, up from an estimated $12 billion in 2026.

Google says it now has nearly 100 TPU customers, from frontier AI labs to trading firms, retailers and universities, and that all of its available TPU supply is already contracted to customers, with shortages of memory and manufacturing capacity. Anthropic is a major customer and holds an option to buy as much as $150 billion worth of Google chips between 2028 and 2031, according to a regulatory filing — a deal that also underpins Anthropic's massive data-center buildout.

One report remains unconfirmed: Bloomberg has reported that OpenAI is planning to buy Google TPUs, a move that would be remarkable given OpenAI's deep ties to Microsoft and Nvidia. Google has not confirmed an OpenAI deal, and no OpenAI commitment to the 8tx or 8i Max has been announced. Watch for named customers before treating that report as settled.

Gloved hands holding an advanced AI processor chip in a clean laboratory

What to Watch Next

Google has not published full specifications or prices for either new system, and both are due in 2027, not now. Oracle announced last month that its cloud customers will be able to access Google's TPUs from next year, widening distribution further. On the software side, Google pointed to TorchTPU, which it open-sourced last month so PyTorch code and Hugging Face models run natively on TPUs — a direct pitch to developers who have built their workflows around Nvidia's CUDA ecosystem.

The bigger picture is a chip market that is no longer a one-company story. Nvidia still leads, and it faces its own pressures — including a Justice Department probe into alleged chip smuggling and a board investigation into its CEO. But Google, with four TPU systems in one year and a growing external customer list, is now the most credible alternative. The test will be whether customers actually commit to the 8tx and 8i Max when they arrive in 2027 — and whether independent benchmarks back up Google's claims.

Frequently Asked Questions

What are Google TPU 8tx and TPU 8i Max?

They are two new AI accelerator systems Google announced on November 9, 2026, expanding its eighth-generation TPU portfolio to four systems. The 8tx is designed for secure, confidential inference; the 8i Max combines two TPU 8i chips in one package for workloads needing huge context windows and high memory bandwidth.

Are the TPU 8tx and 8i Max new chips?

Not entirely. Google describes them as new systems built on existing eighth-generation silicon: the 8tx uses the same silicon as the TPU 8t, and the 8i Max packages two existing TPU 8i chips together. The new part is how they are configured and what workloads they target.

When will TPU 8tx and TPU 8i Max be available?

Bloomberg reported both systems are due in 2027. Google has not published a precise availability date or pricing for either system.

How do Google's TPUs compare with Nvidia GPUs?

Bloomberg Intelligence estimates TPUs cost about 30 percent less than Nvidia GPUs while delivering competitive throughput for many workloads. Nvidia still dominates the overall market, and Google's performance figures are its own claims — no independent benchmarks of the new systems exist yet.

Is OpenAI buying Google TPUs?

Bloomberg has reported that OpenAI plans to buy Google TPUs, but neither Google nor OpenAI has confirmed a purchase. Treat it as an unconfirmed report until a named commitment is announced.

Who else buys Google TPUs?

Google says it has nearly 100 TPU customers. Anthropic is a major one, with an option to buy up to $150 billion of Google chips between 2028 and 2031. Oracle said last month its cloud customers will be able to access Google TPUs from next year.

Search This Blog

TrendWire Daily