China Galaxy Securities: Google (GOOGL.US) will list TPUv7 to reshape the competitive landscape of AI chips
The Zhitong Finance App learned that China Galaxy Securities released a research report saying that the AI chip market competition will be more intense in the future, and Google (GOOGL.US) is expected to increase its AI chip market share with TPU v7 series products. The bank believes that with the launch of Google's TPU v7 next year, the domestic liquid cooling/power supply/PCB sector is expected to bring new development opportunities. At the same time, as the competitive pattern of AI chips continues to deepen, the trend of domestic hashrate chips is rising for a long time.
The main views of China Galaxy Securities are as follows:
Google is about to launch TPU v7, technical indicators are comparable to Nvidia B200
Google will soon officially launch the seventh-generation TPU chip “Ironwood”, marking a major breakthrough in AI computing power technology. The chip's single-chip peak computing power reached 4614 TFlops (FP8 accuracy), and is equipped with 192GB HBM3e memory. The memory bandwidth is as high as 7.4 Tb/s, and the power consumption is about 1000W. Compared with its predecessor, Ironwood's computing power has increased 4.7 times, and the energy efficiency ratio has reached 29.3 TFlops per watt, which is double that of the previous generation. In terms of server cooling, the server uses a 100% liquid cooling architecture, using a large cold plate design, covering 4 TPUs and VRMs; the maximum cluster size supports 144 racks interconnected, that is, a cluster of 9216 TPU chips. The overall technical indicators are comparable to the Nvidia B200 chip.
The product focuses on AI inference scenarios for its own Gemini model
TPU v7 focuses on AI inference scenarios and supports low latency requirements such as real-time chatbots, intelligent customer service, and intelligent driving; at the same time, it can also be expanded to large-scale model training, such as Anthropic plans to use one million TPUs to train Claude series models. On the customer side, Meta plans to deploy TPU in data centers starting in 2027 and rent computing power through Google Cloud in 2026. At the same time, Google itself uses training and services for models such as Gemini. Its hyperscale clustering capabilities and low cost advantages are becoming the preferred solution for AI companies to reduce inference costs. In addition to the chip, Google is also simultaneously launching a series of upgrades aimed at making its cloud services cheaper, faster, and more flexible to compete with Amazon AWS and Microsoft Azure
The AI competitive landscape is about to be reshaped. Focus on Google's AI chip industry chain
The bank believes that Google's launch of TPU v7 products will comprehensively drive the transformation of the AI industry chain from hardware to software ecology, drive upstream demand for ASIC chips, PCBs, packaging, HBM, optical modules, heat dissipation, manufacturing packaging, etc. from the top down, and inject momentum into AI model development and application popularization. Competition in the AI chip market will be more intense in the future, and Google is expected to increase its AI chip market share with TPU v7 series products.
Risk Alerts
The risk that downstream demand falls short of expectations, the risk of increased competition in the industry, the risk of new product development falling short of expectations, and the risk of increased uncertainty due to supply chain transfers.