Morgan Stanley's latest report reveals that Google has outlined a roadmap for selling TPUs: shipping 1 million units by 2027, of which 500,000 could generate approximately $13 billion in incremental revenue for the company's cloud business, accounting for 11% of Google's overall cloud revenue and boosting earnings per share (EPS) by about 3%. After the news was released, Alphabet's stock rose nearly 20% over a month, outperforming the Nasdaq during the same period.

Insiders confirmed that Google is adjusting its TPU business model, shifting from an "in-house accelerator" to a "revenue-generating platform." The first key customers include Meta, which is evaluating a multi-billion-dollar purchase order for building new AI data centers. Compared to NVIDIA GPUs, TPUs have advantages in power consumption and cost in high-throughput training and specific large model inference scenarios. Google hopes to break NVIDIA's monopoly and reshape the AI chip landscape with this.

However, analysts caution that TPU exports still face variables such as pricing, software stack compatibility, and workload adaptation. Whether the target of 1 million units can be fulfilled depends on whether Google can quickly address weaknesses in its ecosystem and distribution channels in the short term. If the plan proceeds smoothly, Google will become the second major cloud provider, after NVIDIA, to directly generate revenue from AI chips, and the semiconductor industry competition map may be reorganized by 2027.