Google’s Ironwood TPUs outpace Nvidia GPUs in AI efficiency race
Google’s latest Tensor Processing Units (TPUs) are gaining ground as a strong alternative to Nvidia’s GPUs in AI computing. The seventh-generation Ironwood TPUs, launched in April 2023, now deliver twice the performance per watt of their predecessors. Their success has even caught the attention of major tech firms like Meta Platforms.
Alphabet Inc. unveiled the Ironwood TPUv7 in April 2023, marking a significant leap in efficiency. These chips powered the training of Gemini 3, currently the world’s leading multi-modal and reasoning AI model. The release of Gemini 3 has since boosted industry interest in Google’s TPU technology.
The Ironwood TPUs have doubled efficiency compared to the previous generation, making them a compelling choice for AI training. With Meta considering adoption and Google’s manufacturing needs potentially expanding, the TPU ecosystem could see further growth. For now, Broadcom remains Google’s sole TPU manufacturer, but future changes may reshape the market.
