With Google's parent company Alphabet owns an increasingly powerful TPU (Tensor Processing Unit) business, and with Google DeepMind's focus on AI research, investing in bank D.A. Davidson analysts believe that if the TPU business is independent, the overall value may be as high as $90 billion, a significant increase from the estimated $717 billion earlier this year.
The accelerator TPU, specially designed for machine learning and AI workload, has attracted the attention of AI researchers and engineers. The sixth-generation Trillium TPU has been in a strong demand since its massive launch in December 2024. The seventh-generation Ironwood TPU, designed specifically for the recommendation, was released at this year's Google Cloud Next 25 conference and is expected to receive a large number of applications from customers.
Learn further, each chip of Ironwood TPU can provide 4,614 TFLOPS computing power, drive thinking and recommended models at peak. The Ironwood TPU significantly increases the high-frequency wide memory (HBM) capacity, providing 192GB per chip, which is 6 times that of Trillium TPU, to handle larger models and data collection calculations; at the same time, the Ironwood TPU can reach 7.2Tbps high-speed access per chip, which is 4.5 times that of Trillium TPU. Not only that, the Ironwood TPU's performance and power consumption ratio is twice that of Trillium TPU, providing customers with more computing power per watt for their AI load.
In terms of partners, Alphabet currently only cooperates with Broadcom to produce TPUs, but has issued an opportunity to explore cooperation with Joint Development Corporation, which is produced by Joint Development Corporation for the upcoming Ironwood TPU. At the same time, many AI companies have accelerated their adoption of TPUs. For example, Anthropic started recruiting TPU core engineers, which may balance the dependence on AWS Trainium chips. Elon Musk's xAI has been improved by JAX-TPU tools and has shown interest in TPUs.
D.A. Davidson analysts believe that Alphabet's value in the AI hardware field is undervalued, but to split the TPU business, it is unlikely that it will happen in today's environment. TPU will combine Google DeepMind research to continue to integrate more Google product combinations.
Google may be sitting on a $900 billion gem that could disrupt Nvidia’s dominance Extended reading: Zhongli customers bite NVIDIA? Google promotes its own TPU to advance to the small cloud market Google: TPU provides computing options, not to reduce the dependence on NVIDIA Not affected by the cooperation between Google and Technological Technology, we will continue to win the order of Google's seventh generation TPU