Marvell's early process-node gamble pays off amid AI demand

Marvell skipped 7nm in 2018–2019 and jumped straight to 5nm, then 3nm, 2nm and 14 angstrom to lead in data infrastructure. The company is advancing SerDes and optical interconnects for AI clusters, and has secured a $2 billion investment from Nvidia and a custom AI chip deal with Google.

• Marvell skipped 7nm, moving from 12–16nm straight to 5nm
• About 85% of data transmission in scale-up networks is processor-to-processor
• Nvidia invested $2 billion in Marvell
• Marvell will design custom AI chips for Google