Google Unveils Two New TPUs, Boosting TSMC

According to Google, the company announced on November 9, 2026, the launch of two new in-house Tensor Processing Units (TPUs), the TPU 8TX and TPU 8I Max, further expanding its AI chip portfolio and strengthening its position in the rapidly growing artificial intelligence market.

Industry analysts point out that Google's eighth-generation TPU family relies on TSMC's ecosystem, from advanced semiconductor manufacturing processes to system-level packaging. As Google continues to expand its AI chip lineup and attract adoption from major technology companies, TSMC is also expected to benefit from growing demand for advanced chip manufacturing and packaging services.

Mark Lohmeyer, Vice President and General Manager of AI and Computing Infrastructure at Google Cloud, said the company's AI chip development cycle is accelerating. Previously, Google typically introduced one AI chip every one or two years. It has now reached a milestone of launching four specialized systems within a single year, reflecting its faster hardware development capabilities and commitment to meeting growing AI computing demands.

Earlier this year, Google introduced the TPU 8T, designed for large-scale AI model training, and the TPU 8I, optimized for low-latency inference. With the addition of the TPU 8TX and TPU 8I Max, Google has completed the expansion of its eighth-generation TPU lineup, marking the introduction of four TPU systems within one year.

Industry analysts believe that AI is moving beyond traditional systems of record toward systems of action, which can interact with enterprise resource planning (ERP) platforms and databases to independently execute tasks. This transition is significantly increasing demand for computing resources, memory bandwidth, and high-speed data communication.

The two new chips serve different strategic purposes. The TPU 8TX focuses on end-to-end confidential inference and is Google's first TPU system to integrate hardware-based confidential computing capabilities. It is deeply integrated with AMD's x86 processor architecture, which manages system-level operations. Through hardware encryption and authentication across the entire PCIe communication link, the system protects AI model weights and private data throughout the inference process. This is particularly important for industries such as finance and healthcare, where data security and regulatory compliance are critical.

The TPU 8TX also introduces a new network topology called Boardfly, which reduces the communication diameter between chips by more than 50%. This architectural improvement cuts inference latency by approximately half compared with the previous generation, enhancing performance for demanding AI workloads.

Meanwhile, the TPU 8I Max is designed to handle large document collections, extensive code repositories, and highly parallel AI agent workflows. It adopts a dual-chip packaging architecture that integrates two TPU 8I chips into a single module. The configuration delivers up to 17.2 TB/s of memory bandwidth, doubling the previous capacity, along with 768 MB of on-chip SRAM, also twice that of its predecessor, and 384 GB of HBM, representing a 33% increase.

Under the same rack space and power consumption conditions, the TPU 8I Max delivers more than twice the computing density and data throughput of the standard TPU 8I, making it suitable for increasingly complex AI inference and agent-based applications.

Google emphasized that next-generation AI infrastructure is no longer defined solely by the number of accelerators in a system. Instead, performance depends on the integrated design of silicon chips, high-speed interconnects, memory bandwidth, and workload scheduling software.

The new TPU systems and accompanying acceleration software will be made available through Google Cloud, providing enterprises worldwide with advanced AI infrastructure that combines high performance, flexible resource allocation, and stringent security and compliance capabilities. The latest launches also highlight Google's efforts to strengthen its in-house AI hardware ecosystem as competition in the global AI computing market intensifies.