Google Partners with Marvell on Inference AI Chips to Compete with Nvidia

In a bid to rival Nvidia, Google is collaborating with Marvell Technology to create AI inference chips. The unveiling of their next-generation tensor processing units is set for this week at the Google Cloud Next conference in Las Vegas, with subsequent chips tailored for inference. This initiative allows Alphabet to address the increasing demand for AI software. According to Gartner analyst Chirag Dekate, there is a notable trend toward inference. Google’s Chief Scientist, Jeff Dean, highlighted the importance of differentiating between training and inference chips. Demand for Google’s TPUs has skyrocketed, especially after Meta entered into a multibillion-dollar agreement through Google Cloud. Anthropic has also expanded TPU availability to 1 million chips and is collaborating with Broadcom for chips expected in 2027. However, supply limitations could impede Google's goals, as top teams are prioritized for the available chips.