Google may be Nvidia's biggest rival in chips — and now it's upping its game

Google is rolling out its seventh-generation Ironwood Tensor Processing Unit (TPU), a custom AI chip reportedly four times more powerful for training and inference than its predecessor, designed for demanding AI workloads. This development significantly strengthens Google's position as a formidable alternative to Nvidia in the AI silicon market, with analysts like D.A. Davidson's Gil Luria valuing a potential spun-off TPU/DeepMind unit at $900 billion. The adoption of these TPUs by Anthropic for its Claude LLMs, anticipating substantial price-performance gains, further validates Google's advanced chip strategy and its potential to accelerate Google Cloud growth and broader AI innovation.

Analysis

Google has launched its seventh-generation Tensor Processing Unit (TPU), Ironwood, which is touted as four times more powerful for AI training and inferencing than its predecessor. This custom-built chip is designed for demanding AI workloads, including large-scale model training and high-volume, low-latency inferencing, with the capability to connect up to 9,216 units in a POD for massive shared memory access.

This development significantly strengthens Google's competitive position against Nvidia in the AI chip market, offering a credible alternative to GPUs. The strategic importance is highlighted by Google's internal use for Search, YouTube, and Gemini AI, alongside external validation from Anthropic, which plans to deploy up to 1 million Ironwood units for its Claude LLMs, expecting "impressive price-performance gains." This adoption by Anthropic, previously reliant on AWS, signals a notable shift in the AI infrastructure landscape.

Analysts are highly optimistic, with D.A. Davidson's Gil Luria suggesting Google's TPUs have closed the gap with Nvidia and valuing a potential spun-off TPU/DeepMind unit at $900 billion. Melius Research's Ben Reitzes notes the TPUs as the "most proven ASIC," contributing positively to Google Cloud (GCP) growth and even Broadcom's AI revenues. This indicates a strong inflection point for Google's custom silicon strategy, driving both internal innovation and external revenue streams.

Key Decisions for Investors

• Monitor Google's (GOOGL/GOOG) continued traction in custom AI silicon, as the Ironwood TPU's capabilities and external adoption by Anthropic present a growing competitive challenge to Nvidia's (NVDA) dominance in AI accelerators.
• Evaluate the potential for accelerated growth in Google Cloud Platform (GCP) revenues, driven by both internal and external demand for Google's advanced AI infrastructure.
• Assess the evolving competitive landscape in AI hardware, considering the implications for market share and long-term strategic positioning for key players including Nvidia and Amazon (AMZN), which offers its own custom AI chips.
• Consider the positive implications for indirect beneficiaries like Broadcom (AVGO), given analyst commentary suggesting their involvement in Google's custom silicon efforts could boost their AI-related revenues.