Small AI Chip Maker Marvell Now More Valuable Than Intel
wsj.com
wsj.com
The real kicker is that they picked up their ARM license from Intel, when they bought the XScale line (now Armada) in 2006.
Actually, if you look at their quarterly report, the breakdown of their revenue by market has changed dramatically in the last year. Revenue from the data center market went from 36% to 71% of their revenue, a change which they attribute to AI creating demand. Meanwhile networking dropped from 24% to 10% of their revenue, and other categories have gone down as well.
From the report:
> Net revenue in the third quarter of fiscal 2025 was $1.5 billion and was 7% higher than net revenue in the third quarter of fiscal 2024. This was due to a 98% increase in sales from the data center end market compared to the three months ended October 28, 2023. The increase was partially offset by decreases in sales from the carrier infrastructure end market by 73%, from the enterprise networking end market by 44%, from the consumer end market by 43% and from the automotive/industrial end market by 22%.
> We have seen strong revenue growth from our data center end market, driven by robust demand for our electro-optics and custom compute products from AI applications.
https://app.quotemedia.com/data/downloadFiling?webmasterId=1...
They've sprinkled "AI" everywhere above the fold, but all the actual products they're talking about are still networking products.
> Meanwhile networking dropped from 24% to 10% of their revenue, and other categories have gone down as well.
Enterprise networking is a category they report as dropping from 19% to 10% year-over-year for a three month period. I think you got the 24% from an incomparable column reporting on a nine month period. But either way, enterprise networking is obviously just a small subset of all their networking business.
Marvell is clearly still primarily a networking business. AI is driving demand for some of their networking products (or at least, that's what they're attributing the demand to), but that's not the same thing as Marvell actually making products that are doing AI, or selling such products in enough volume to double their datacenter revenue.
Oops, you're right, the comparison I meant to make was 24% to 12% over a 9 month period.
> But either way, enterprise networking is obviously just a small subset of all their networking business.
Is it, though? That table [0] puts everything "consumer" into a chunk that's now only 6% of their revenue. Carrier infrastructure is another 6%. Together that adds up to roughly equal to the enterprise networking row, but that's still a tiny amount compared to the data center revenue.
Are you saying that "data center" includes a lot of networking equipment that isn't filed under "enterprise networking"?
[0] Looking at page 9
Yes, exactly. "Enterprise" and "data center" are treated as disjoint segments by many tech companies, even though both tend to refer to rackmount hardware. That's why I linked to Marvell's own page describing their data center products, so you could see for yourself that they are indeed networking products: switches, PHYs, NICs, etc.
All of the categories under the "Net revenue by end market" table are comprised primarily of networking products: "Data center" is mostly networking stuff, "Enterprise networking" is obviously networking stuff, "Carrier infrastructure" is where you find networking stuff with the now-passé "5G" buzzword, "Consumer" is probably mostly networking but might actually be the segment with the lowest proportion of networking products (though Marvell's SSD controllers aren't too common these days in consumer SSDs. Also, Realtek has recently started eating away at Marvell's Aquantia revenue for 2.5Gb and 5Gb Ethernet in consumer products.) "Automotive/industrial" is primarily about automotive Ethernet.
Most of their revenue increase is from data center expansion due to AI. Though their enterprise has grown in recent years quite a bit.
Well, no of course you can't.
They started with disk drive controllers, recently they’ve made a decent NVME controller. They bought global foundries custom asic devision in 2019 and that’s when they really got into the custom chip business (according to wikipedia).
As an aside, it's funny to read "Intel's corner office". Andy Grove eschewed a private office and other executive perks, opting to work in a standard cubicle: https://en.wikipedia.org/wiki/Andrew_Grove#Egalitarian_ethos
Because of Grove's management style, information is ammunition--you do not share it. This is anathema to problems that need heavy R&D to solve.
The corollary to this is that if you know someone is making a mistake, you let them as it improves your political position.
"While Grove supported helping technology startups, he also felt that America was wrong in thinking that those new companies would increase employment. "Startups are a wonderful thing," he wrote in a 2010 article for Bloomberg, "but they cannot by themselves increase tech employment."[40] Although many of those startups and entrepreneurs would achieve tremendous success and wealth, said Grove, he was more concerned with the overall negative effect on America: "What kind of a society are we going to have if it consists of highly paid people doing high-value-added work and masses of unemployed?"
Even Google which is very hesitant to enter businesses that require lots of labor employs hundreds of thousands of contractors.
You can't just look at revenues without looking at expenses.
https://www.cnbc.com/2020/11/19/walmart-and-mcdonalds-among-...
If a poor person has two income sources, either source is less essential than if he had just one of the sources. The less essential the pay from the Uber gig, the less Uber can squeeze the worker.
Without the work requirement, you're actually helping people.
The reason we have it partly that welfare recipients literally don't like getting welfare without strings attached because they think it's embarrassing (or that the other people getting it don't deserve it), and part that states are hoping to save money by kicking people off for doing the paperwork wrong.
EITC is a way to do "work requirements" without the downsides.
In this way welfare is actually subsidizing Uber employment. Assuming of course, that welfare is also insufficient to survive.
You're actually arguing that welfare is bad - if giving people welfare allowed Uber to lower their wages that means people would be better off without it.
How many have Uber destroyed?
For Amazon the claim is more obvious - picking up the "killing local business" and "killing US suppliers" torches from Walmart before them - but I would be a bit suprised if there were any cities where there were fewer transportation drivers now than pre-Uber in the US. Taxis in the vast majority of the country were pretty few and far-between.
They are very different types of tech company; not as completely different as something like SaaS with no cost of physical goods at all, but Uber (while hardly an example of a good citizen company) is not really the sort of "some people in one city have jobs and nobody else does as a result" tech company as some others.
Which goes to show how quickly everything can change. Everything’s built on sand, literally.
What makes you think so? They're about to deliver 18A in 6 months
that was firmware issue, not the node, yup?
There were many steps at which Intel could have avoided this outcome, but it started from the fact that Intel had to push voltages and clocks beyond reasonable limits to make up for having a microarchitecture with poor enough performance per clock that Intel needed to hit 6+ GHz to be competitive.
So Intel sold the Xscale family, just before iPhone and Android happened… To Marvell.
NVidia's P/E is currently about 120, that is, the market values NVDA as 120 years of its current earnings. I'm not even certain that NVidia will last as long, or will be in an equally good shape. E.g. IBM has lasted 120 years, and it had wild fluctuations of its market position in these years; toda it's hardly comparable to what it was a century ago.
I would say that Intel is now easier to buy in a large acquisition, without using words like "valuable".
Intel is no longer relevant and should be shut down, and the money invested in other more promising ventures with a future.
Marvell: 6.5k employees
AMD: 26k employees
For Intel, the number of employees is a liability, not an asset.
Fab + design... its apples to oranges.
TSMC for example has 77k employees and looking to add 23k more.
Packaging and testing are labor intensive, and require folks to be added in different geographies.
SMIC: 20k employees
UMC: 20k employees
TSMC only has that many employees because they have a nearly 2/3rd market share of all ASIC production.