Alphabet's Tensor Processing Unit (TPU) business could be worth as much as $900 billion if spun off, according to a new report by investment bank D.A. Davidson. The estimate marks a sharp increase from the bank's earlier valuation of $717 billion, reflecting the rapid expansion of Google's in-house AI accelerator efforts and the growing importance of Google DeepMind's research.

Google has drawn strong interest from researchers and engineers with its custom-built TPUs for machine learning and AI workloads. Demand surged following the large-scale rollout of the sixth-generation Trillium TPU in December 2024, and adoption is expected to accelerate further with the seventh-generation Ironwood TPU, unveiled at the Google Cloud Next '25 conference earlier this year.

Ironwood delivers 4,614 TFLOPS of peak performance per chip, significantly boosting inference workloads. It also features 192 GB of high-bandwidth memory (HBM)—six times more than Trillium—and up to 7.2 Tbps bandwidth, a 4.5x increase. With twice the performance-per-watt of its predecessor, Ironwood is positioned to handle increasingly large AI models and datasets while improving energy efficiency.

Despite these advances, many smaller cloud providers remain reliant on Nvidia GPUs due to CUDA ecosystem maturity and overall cost efficiency. However, Google has made progress convincing partners: London-based Fluidstack, which primarily rents Nvidia GPUs to AI startups, is reportedly planning a new data center in New York. Sources said Google may provide up to $3.2 billion in financial support to secure deployment of TPUs at the site if Fluidstack faces rental constraints.

Alphabet currently works exclusively with Broadcom on TPU development, but reports suggest MediaTek may join as a backend design partner for Ironwood. Meanwhile, AI firms are signaling stronger adoption of TPUs—Anthropic has begun hiring TPU engineers to diversify from AWS Trainium, while Elon Musk's xAI is exploring TPU support following improvements to the JAX-TPU toolchain.

D.A. Davidson's analysts believe Alphabet's AI hardware business remains undervalued, though they see a spin-off of TPU operations as unlikely in the current environment. Instead, the company is expected to integrate TPU advancements more deeply with Google DeepMind and across Google's broader product ecosystem.