Google aims to challenge NVIDIA with TPU sales push, targets 1M units by 2027

According to a recent Morgan Stanley report, Google has unveiled plans to commercialize its Tensor Processing Units (TPUs), aiming to ship 1 million units by 2027. Of these, 500,000 could add roughly $13 billion in revenue to Google Cloud—about 11% of its total cloud income—and lift Alphabet’s earnings per share by around 3%. Following the announcement, Alphabet’s stock surged nearly 20% in a month, outpacing the Nasdaq.

Insiders say Google is shifting its TPU strategy from an internal tool to a revenue-generating product. Early adopters include Meta, which is considering a multi-billion-dollar order for new AI data centers. TPUs offer advantages over NVIDIA GPUs in power efficiency and cost for high-throughput training and certain large-model inference tasks.

Still, analysts warn that success hinges on overcoming challenges like pricing, software compatibility, and ecosystem readiness. If Google executes its plan, it could become the second major player—after NVIDIA—to monetize AI chips directly, potentially reshaping the semiconductor landscape by 2027.