Understand this story
On November 9, 2026, Google announced that Pathways is now generally available (General Availability) on Google Cloud Platform. According to the Google Developers Blog, Pathways is a distributed system Google uses to train its own large models such as Gemini; a single Python client can describe the entire computation as a graph, and the distributed runtime handles placement, tensor movement, and scheduling, thereby supporting execution modes beyond SPMD such as pipeline parallelism and RL post-training; virtual devices can remap tasks when a host fails, and the Shared Pathways Service allows multiple clients to share the TPU capacity of the same cluster. On the same day, in an article on the Google Blog about expanding the eighth-generation TPU product line, Google also announced that Cloud Pathways and GKE dynamic slicing are now generally available. There are currently no further developments.
AI-generated from public reports
Latest development · Report
Google launches two new systems: TPU 8tx and TPU 8i MaxDevelopments
1 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 10
Analytics India Magazine
Google launches two new systems: TPU 8tx and TPU 8i Max
Google has added TPU 8tx and TPU 8i Max to its eighth-generation TPU product line, targeting confidential inference and multi-agent large-context workloads, respectively. TPU 8tx and TPU 8t use the same chip and are paired with fifth-generation AMD EPYC CPU hosts; the company says latency is up to 2 times lower than TPU 8t with near-zero performance loss. TPU 8i Max packages two TPU 8i chips together, with twice the compute, SRAM, and HBM bandwidth of TPU 8i, and can support million-token-level contexts. Google also announced software updates including open-source TorchTPU, XLA Serving Engine, TPU Sync, and the preview-stage TPU Agent Suite, and said its TPU business already has nearly 100 customers.
Jiwei.com
Google Launches Two In-House Chips: TPU 8TX and TPU 8I Max
Google announced the launch of two new in-house tensor processors, TPU 8TX and TPU 8I Max, expanding the eighth-generation TPU family to four systems within a year. TPU 8TX focuses on full-stack confidential inference; it is the first TPU system to integrate hardware confidential computing features, and it introduces the Boardfly network topology, reducing the inter-chip communication diameter by more than 50% and halving inference latency compared with the previous generation. TPU 8I Max uses a dual-chip package, integrating two 8I chips within a single module, providing 17.2 TB/s of memory bandwidth, 768 MB of on-chip SRAM and 384GB of HBM; within the same rack space and power consumption, its compute density and data throughput are twice those of TPU 8I. Both chips and the acceleration software will be available on Google Cloud.
Jiwei.comRelated discussion
Jiwei.com joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
TMTPost · AGI
Edge AI Daily Morning Brief: Nvidia Under Investigation, Anthropic's $895.4 Billion Compute Commitment and…
On the same day Nvidia's board launched an independent investigation, Google unveiled two eighth-generation TPUs, the TPU 8tx and TPU 8i Max, targeting confidential computing and high memory bandwidth. Anthropic plans to list on Nasdaq before Thanksgiving, with its valuation range lowered to $1.2 trillion to $1.3 trillion, and cumulative purchase and lease commitments of approximately $895.4 billion. OpenAI completed an initial funding round of about $11 billion, corresponding to a valuation of $1.25 trillion, with the total for this round reaching up to $20 billion.
Phoenix Technology
Google releases two self-developed AI chips, 8tx and 8i Max, with up to twice the speed
Google has released two self-developed AI chips redesigned based on TPU, 8tx and 8i Max, respectively targeting high data security requirements and high-intensity AI computing scenarios. 8tx combines Google's self-developed technology with AMD processors, supports hardware-level data encryption, and boosts processing speed to up to twice that of the previous generation with almost no performance loss due to encryption; 8i Max integrates two chips in the same package to provide larger memory and stronger computing power, suitable for AI coding assistants, complex mathematical reasoning, and drug development. After the announcement, Alphabet's stock rose more than 3% in early trading.
East Money
Google Releases Two New Variants: TPU 8tx and 8i Max
Google has released two derivative variants of its eighth-generation TPU, the TPU 8tx and 8i Max, aimed respectively at confidential computing and long-context AI applications, and both will launch next year. The 8tx works in conjunction with AMD CPUs and can encrypt data at the chip level; the 8i Max integrates two TPUs in the same package, providing stronger compute power and larger memory. Amin Vahdat, who heads Google's chip business, said the new products are reconfigurations based on customer needs rather than entirely new designs, expanding the eighth-generation TPU lineup to four models.
East MoneyRelated discussion
East Money joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
ServeTheHome
Google launches TPU 8tx and TPU 8i Max, with the former paired with AMD EPYC Turin hosts
Google has added two new system configurations, TPU 8tx and TPU 8i Max, to its eighth-generation TPUs. TPU 8tx uses the TPU 8t chip, paired with fifth-generation AMD EPYC Turin host CPUs, protects data in use through confidential computing, and adopts the Boardfly topology; Google says its latency is up to 2 times lower than that of TPU 8t. TPU 8i Max packages two TPU 8i chips together, doubling HBM bandwidth to 17.2TB/s, doubling on-chip SRAM to 768MB, and providing 384GB of HBM capacity.
Unite.AI
Google Cloud makes Pathways Runtime generally available on GKE
Google Cloud announced that Cloud Pathways is now generally available on GKE, providing a single-controller runtime for large-scale training and inference. Over the past 90 days, users have consumed more than 130 million Cloud TPU chip hours using Pathways on GKE. The runtime was originally developed by Google DeepMind to train and deploy frontier models such as Gemini. By setting JAX_PLATFORMS=proxy, a single JAX client can orchestrate multiple TPU slices spanning thousands of chips as local devices, and it supports elastic training based on host memory snapshots. Google says a large consumer technology company is already running production batch processing and reinforcement learning inference on tens of thousands of TPU v5p, Trillium v6e, and Ironwood TPU7x chips, consuming more than 100 million chip hours per quarter; integration with Slurm clusters remains in Preview.
Data Center KnowledgeRelated discussion
Data Center Knowledge joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Sundar Pichai
Google's eighth-generation TPU adds TPU 8i Max and TPU 8tx
Google is expanding its eighth-generation TPU product line, adding the TPU 8i Max, aimed at massive context windows and the highest compute density per rack, as well as the TPU 8tx, aimed at secure confidential inference. Amin Vahdat said this is the first time Google has launched four different TPU systems within one year and within the same generation. Developers can run native PyTorch and Hugging Face models directly through TorchTPU without rewriting code; the open-source TPU Sync reduces KV transfer latency by more than 90%, and Agent Compute Engine is used to reduce wasted idle compute.
MarkTechPostRelated discussion
MarkTechPost joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
cnBetaRelated discussion
cnBeta joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Nov 9
ZhidxSelected
Google Releases TPU 8i Max and TPU 8tx, Doubling HBM Bandwidth
Google has released two eighth-generation TPUs: the TPU 8i Max, aimed at ultra-large context windows, doubles the compute, SRAM, and HBM bandwidth of the TPU 8i; the TPU 8tx, which supports secure and confidential inference, can reduce latency by up to half compared with the TPU 8t. This is the first time Google has launched 4 TPUs within a single year, with the eighth-generation products covering the full lifecycle of agents. Google has also upgraded its open-source software stack, opening TPU development capabilities to PyTorch developers and open-sourcing the TPU agent toolkit, XLA Serving Engine, and TPU Sync.
Cailian Press
Google releases two TPU variants, the 8tx and 8i Max, targeting confidential computing and long-context applications respectively
Google has released two new variants of its eighth-generation TPU, the TPU 8tx and 8i Max, targeting confidential computing for government and regulated industries and AI applications requiring long context windows, respectively, with launch set for next year. The 8tx works in tandem with AMD CPUs and can encrypt data at the chip level; the 8i Max integrates two TPUs in a single package, delivering greater compute power and larger memory. Amin Vahdat, who heads Google's chip business, said the new products are reconfigured offerings based on customer needs rather than entirely new designs.
Cailian PressRelated discussion
Cailian Press joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Lisa Su
AMD's Lisa Su confirms EPYC serves as host processor for Google TPU 8tx
AMD CEO Lisa Su congratulated Google Cloud and Amin Vahdat, and confirmed AMD EPYC as the host processor for TPU 8tx, with the two sides collaborating on this. The cited original Google post states that the eighth-generation TPUs add TPU 8i Max and TPU 8tx for secure confidential inference, marking the first time Google has launched four TPU systems within the same year.
Yicai
AI Evolution Express: Google to Launch Two New TPU Chips, ByteDance Releases Seedream5.5
Google will launch two new tensor processing chips, TPU 8tx and TPU 8i Max, expected to hit the market next year. ByteDance released its Doubao image creation model Seedream5.5; its API continues the two-part billing of input image plus output image, with a single 2K square image edit costing about 0.28 yuan, less than half that of Seedream 5.0 Pro. Alibaba Cloud raised the subscription quota for the personal edition of Token Plan; the monthly fees for the Lite, Essential, and Standard tiers remain unchanged, while monthly quotas increased by about 74% to 78%.
Testing Catalog
Google releases two new chips, TPU 8tx and TPU 8i Max
Google announced the launch of two new TPUs, TPU 8tx and TPU 8i Max. TPU 8tx targets large-scale training, with target customers being regulated enterprises and government agencies that run models and autonomous agents on sensitive information; TPU 8i Max targets low-latency inference, addressing the memory needs of agents that handle large codebases, long videos, and million-token contexts.
SiliconANGLERelated discussion
SiliconANGLE joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
The Decoder
Google launches four TPU systems within a year, adding TPU 8tx and TPU 8i Max
Google has added two systems, TPU 8tx and TPU 8i Max, to its eighth-generation AI accelerator lineup, and says this is the first time it has delivered four TPU systems within a year. TPU 8tx is aimed at confidential computing, with data and model weights kept encrypted and isolated during use; TPU 8i Max places two TPU 8i chips on a single package, achieving twice the compute, twice the memory and twice the memory bandwidth within the same rack specification. Google has also open-sourced TorchTPU, enabling PyTorch code to run on TPUs with only minor changes, and supports vLLM and SGLang; external TPU customers now number close to 100.
Google BlogFirst-partySelected
Google Releases TPU 8tx and TPU 8i Max, Expanding the Eighth-Generation TPU Product Line
Google is expanding its eighth-generation TPU product line with the addition of the TPU 8tx for confidential inference and the TPU 8i Max for ultra-long context and the highest per-rack compute density, forming four systems together with the TPU 8t and TPU 8i; this is also the first time Google has launched four TPUs within a year. The TPU 8tx is based on the same chip as the TPU 8t and is paired with a fifth-generation AMD EPYC CPU host, protecting data and model weights in use at the hardware layer, with latency reduced by up to 2 times compared with the TPU 8t; the TPU 8i Max packages two TPU 8i chips together, with compute, SRAM, and HBM bandwidth all twice those of the TPU 8i. Google has also open-sourced TorchTPU, TPU Sync, and the XLA Serving Engine, and announced the general availability of Cloud Pathways and GKE dynamic slicing.
Google Developers BlogFirst-partySelected
Google officially opens the Pathways distributed training system on Google Cloud
Google announced that Pathways is now generally available (General Availability) on the Google Cloud platform; this is the distributed system Google uses to train its own large models such as Gemini. Pathways lets a single Python client describe the entire computation as a graph, with the distributed runtime responsible for placement, tensor transfer, and scheduling, thereby supporting execution modes beyond SPMD such as pipeline parallelism and RL post-training. The article also explains that virtual devices can remap tasks when a host fails, and that the Shared Pathways Service allows multiple clients to share the TPU capacity of the same cluster.
Heat trend for this story
- Current
- 101
- Period peak
- 124 Nov 10 10:00
- 24-hour change
- —
Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.