Nvidia still dominates AI silicon with an estimated 81% market share, but a new long-term risk is emerging from Alphabet’s in-house hardware push. While AMD, Intel and Broadcom have won marquee hyperscaler deals, Google’s seventh-generation TPU, Ironwood, aims squarely at large-model training and high-volume inference with a claimed fourfold performance gain over its prior generation. Google is also rolling out Axion, an Arm-based CPU it says delivers double the price-performance of x86 rivals for AI-centric workloads. Nvidia projects $1 trillion in Blackwell and Vera Rubin chip sales across 2026–27—dwarfing Broadcom’s and AMD’s targets—yet Alphabet’s vertical integration and steady TPU cadence could chip away at Nvidia’s grip on AI compute over time.
Related articles:
Google Cloud TPU overview and product family
AWS Graviton Arm-based processors for cloud workloads
