Google pitches TPU chips to Nvidia-focused neocloud providers select customers
Google $GOOGL is pitching its TPU chips to neocloud providers whose business is built on Nvidia $NVDA GPUs, The Information reports. The sales push targets the specialized AI clouds at the core of Nvidia's demand, and some providers are reportedly resisting the switch.
Google $GOOGL is taking its AI chip fight directly into Nvidia $NVDA territory. The company is pitching its tensor processing units, or TPUs, to neocloud providers: the specialized AI cloud companies whose entire business is built around renting out Nvidia GPUs, The Information reports. According to Yahoo Finance, Google has also begun selling TPU hardware to select customers rather than offering the chips only through its own cloud. Google's pitch centers on more stable performance and simpler networking than GPU-based systems, but the reception is mixed: providers such as Nscale, which are deeply tied to Nvidia's software and hardware ecosystem, are reportedly reluctant to switch. The choice of target is what makes this notable. Neoclouds are among the most committed buyers in Nvidia's order book, so every TPU rack that lands in one of these data centers displaces GPUs at the heart of Nvidia's demand rather than at its edges. For enterprises renting AI compute, a second credible chip supplier inside the neocloud market changes the negotiating position on price and availability, whether or not they ever run a workload on a TPU.