GOOGLE ORDERS 2M AI CHIPS FROM MARVELL

The partnership marks Google's continued push to reduce reliance on third-party chip manufacturers and build specialized silicon for its AI infrastructure. Google has been developing its own processors for years, including its Tensor chips for mobile devices and TPUs (Tensor Processing Units) for data centers. The new collaboration with Marvell suggests the company is expanding this strategy with additional custom designs tailored to specific workloads. Marvell Technology, a semiconductor design firm, specializes in creating chips for data centers and cloud infrastructure. The company does not manufacture chips itself but works with foundries to produce designs. This approach allows Google to leverage Marvell's expertise in data center architecture while maintaining control over specifications. The nearly two million chips Google is planning represent a significant commitment to custom silicon. The scale indicates these processors are intended for widespread deployment across Google's global data center network, likely supporting AI model training, inference, and other computational tasks. This move aligns with a broader industry trend where major cloud providers—including Amazon, Microsoft, and Meta—are developing proprietary chips to optimize performance and reduce costs. Custom silicon typically offers better efficiency than off-the-shelf options for specialized tasks. Google has historically worked with established chip manufacturers like TSMC and Samsung for chip production, and the Marvell deal likely follows this pattern, with manufacturing outsourced to a foundry partner. The initiative also reflects competitive pressure in the AI infrastructure space. As demand for AI chips accelerates, securing reliable supply chains and optimized designs has become critical for companies building large language models and other AI systems. Details on the specific purposes of the two chips and their expected capabilities remain unclear. Google declined to comment on the reports.