Prediction: Alphabet's New TPUs Are Another Reason to Buy the Stock
Alphabet has launched its eighth generation of Tensor Processing Units (TPUs), enhancing its position in the AI market. These new TPUs, designed for both training and inference, offer significant performance improvements and cost advantages over competitors. The chips support popular frameworks like PyTorch and JAX, expanding their market reach. With a focus on inference and agentic AI, Alphabet's TPUs are expected to drive new revenue streams through Google Cloud and partnerships. Analysts suggest that these developments make Alphabet's stock a strong buy.
Alphabet (GOOGL +1.71%) (GOOG +1.35%) just unveiled the eighth generation of its highly successful Tensor Processing Units (TPUs), which could be another catalyst for the stock's growth. The company has really separated itself in the artificial intelligence (AI) race with its custom AI chips, and the newest iterations look set to extend its lead in this area.
Alphabet developed its TPUs more than a decade ago as a cheaper way to run its internal workflows, and wrapped its entire hardware and software ecosystem around these chips. However, it was a smart pivot away from its Tensorflow framework, for which the chips are named, that has positioned its chip business to be where it is today.
Image source: The Motley Fool.
Alphabet uses its TPUs not only to run its internal workflows, but also to train its Gemini foundational large language model (LLM) and to run inference. This has given the company a significant structural cost advantage over competitors who are still largely reliant on Nvidia's graphics processing units (GPUs) to run their AI workloads.
Meanwhile, the move to have its TPUs support PyTorch and JAX (its new internal framework) has opened the market for its customers to deploy the chips at scale. This is also leading to entirely new revenue streams, with TPUs being used more widely in Google Cloud and even being sold directly through its co-developer partner, Broadcom.
NASDAQ: GOOGL
Key Data Points
With its eighth generation of TPUs, meanwhile, Alphabet is splitting them into two, with one version for training and another for inference. This makes a lot of sense given that inference is becoming a larger share of the AI market.
While Alphabet's seventh-generation Ironwood chip was built with inference in mind, its TPU 8i is designed specifically for inference and agentic AI. The chip will have both more static random-access memory (SRAM) and high-bandwidth memory (HBM) to reduce the lag caused by data transfers between chips. It can also be paired with Alphabet's Axiom central processing unit (CPU) to better handle agentic AI. Overall, Alphabet said its chips will offer "80% better performance-per-dollar" versus its prior-generation TPU.
Time to buy the stock
Alphabet's new TPUs are just another reason to buy the stock, in my view. The specs on the chips look great, and it shows that the company is moving toward where the ball is headed, with a focus on inference and agentic AI.
The company has a significant advantage over other hyperscalers, thanks to its early lead in TPUs. Customers looking to use them, either through Google Cloud or buying them through Broadcom, just adds another revenue stream. Don't underestimate the company's burgeoning CPU business, which could be an additional growth driver.
With the most complete AI stack, this is an AI stock to own.
