GlobalFoundries and Marvell expand capacity for AI data-centre links
The agreement focuses on the connectivity chips that move information through increasingly large and power-hungry computing clusters.
The story
GlobalFoundries and Marvell are expanding a manufacturing agreement for chips used in AI data-centre connectivity, Reuters reported.
Training clusters need high-speed links to keep thousands of processors working efficiently. Optical and networking components are therefore becoming as strategically important as the accelerators receiving most public attention.
The capacity deal suggests customers expect sustained demand. Execution will depend on yield, packaging and the ability to match rapid changes in data-centre architecture.
INNOVOX analysis
The AI hardware market is becoming a systems contest. Faster accelerators increase the need for efficient links, so value is migrating toward networking, optics and advanced packaging.
What to watch
Watch manufacturing yields, customer commitments and whether new capacity arrives before cluster architectures shift to another connectivity standard.