One thing from yesterday’s $MRVL Investor Day I also think about.
Marvell says it does not need new customer wins to reach its long-term custom targets. The programs are already in place.
And the opportunity is becoming much bigger than the XPU itself. Management sees 1–2 additional XPU attach sockets at roughly $1,000 ASP each across networking, memory, storage, infrastructure management and inference acceleration.
They think individual attach categories can become billion-dollar businesses.
That helps explain how custom goes from ~$1.5B in FY26 to >$12B in FY29 and ~$30B by FY31.
显示更多
From yesterday’s $MRVL Investor Day:
Marvell just gave us a pretty wild look at what they think this business can become.
FY2031 revenue target: $70–90B.
For comparison, Marvell did just $8.2B in FY2026.
And the revenue is supposed to come from:
Custom silicon: ~$30B
Interconnect: ~$37.5B
So Marvell is basically saying that moving data around AI clusters could become an even larger business for them than building the custom accelerators themselves.
And custom is hardly small either. They now see custom chip revenue reaching $12B by FY2029, up from the previous $10B target.
I’ve written a lot about Broadcom because I think the custom silicon opportunity is still underestimated.
But yesterday’s Investor Day was another reminder that this market is getting large enough for more than one winner.
If Marvell gets anywhere close to $30B in custom silicon and $37.5B in interconnect, the company we are looking at in 2031 will barely resemble the Marvell of a few years ago.
But 5 years out. Alot will and can happen in that time.
显示更多