Google partners with Marvell Technology to design custom AI chips, stepping up challenge to Nvidia's market dominance

Google is working with chipmaker Marvell Technology to build two new AI chips: a memory processing unit (a chip that handles data storage and retrieval) to pair with Google's existing Tensor Processing Unit (TPU), and a new TPU optimized for running large language models (AI systems that understand and generate text). The partnership aims to reduce Google's reliance on Nvidia chips, which currently dominate AI computing.

The new memory chip addresses a bottleneck in current AI systems — right now, AI models spend a lot of time waiting for data instead of processing it. By pairing faster memory access with dedicated processing power, Google's setup could let AI systems produce answers more quickly and handle more requests simultaneously, especially for business applications like AI chatbots and search.

For business professionals and students using Google Cloud services or relying on Google's AI tools, this means cheaper, faster AI features down the line — Google can reduce the cost of powering these services and pass savings along to customers. For the broader tech industry, this signals that Nvidia no longer has automatic dominance in AI chips; companies with deep pockets and in-house engineering (like Google, Amazon, and Meta) are now building their own, which could eventually lower AI computing costs across the board.