Google is shifting its tensor processing units (TPUs) into rival data centres, aiming to disrupt Nvidia's GPU dominance, according to a report. The tech giant has initiated partnerships with firms like CoreWeave and secured a $3.2 billion deal with Fluidstack for hosting its chips in New York. Google's strategy seeks to expand TPU access beyond its cloud, yet it must persuade developers to shift from Nvidia's established tools to attain widespread adoption amidst intensifying competition in AI hardware.