Monday, June 02, 2008

Will LSI buy Chelsio?

It’s a reasonable speculation from where I sit in the blogosphere. In a recent interview with CEO Ahbi Talwalkar (left) I found out the company lacks standard Ethernet chips to support its nascent drive into networking, but it does has an investment in the startup that was early to the long-coming 10G party.

It’s clear a good 10G capability would be key for the company that wants to live sat the space where storage and networking meet. However, there are plenty of available startups out there, LSI has some custom capabilities that could spawn products and there is always the question of when the timing is right for the 10G ramp to really start.

I’m thinking pretty darn soon. But I’d love to hear any good insights you may have in a comment or at rbmerrit@cmp.com

8 comments:

Anonymous said...

I would doubt this will happen soon based on the current size of the 10G NIC market and IC development costs required...

Rick Merritt said...

But Chelsio has the chips ready to go, and the market should...ahem...starting ramping soon, yes?

Anonymous said...

The 10G NIC market is in for a brutal shakeout. It's a crowded space (Chelsio, NetXen, Neterion, NetEffect, Mellanox, Intel (!!), ...), and getting a bit long in the tooth. (Chelsio has been around since 2002, and has reportedly raised close to $100m.) It's time to head for the exits before 10G NICs are included on the motherboard, because then, it'll be game over for vendors of $1k-a-pop NICs. Costs of the rest of the solution are coming down: it's hard to spend more than $2k-$3k on a server these days, so $1k for the NIC represents a 33%-50% cost increase over the 1G NICs that come for free these days. On the other end of the 10G link, high-density, low-cost 10G switches are now coming to market, driving down the costs there...

Rick Merritt said...

Wow, the 10G startup battle sounds even bloodier than I thought. So maybe LSI's best move is to develop in house a low cost 10G alternative for the motherboard and let the others die a painful death...if LSI really has the IP they claim. It's always ;possible they are just laying out chips for customers who have the 10G IP.

Anonymous said...

What a doom-and-gloom attitude. When server vendors start shipping 10G Ethernet as a LOM solution, then the industry and customers alike will realize that 10G Ethernet has finally arrived. Sure, this means that price pressures will increase but for those that win the LOM business, they make up for it in volume offerings.

Does the industry need a dozen 10G Ethernet suppliers? Likely not but that does not mean it will boil down to just two like it largely did for 1G Ethernet.

As for actual pricing, who ever pays list for anything these days? Volume pricing effects always apply and these will apply in the 10G Ethernet market as well.

Also, don't forget that these various Ethernet providers vary in terms of what their products actually do. Some are just layer 2 Ethernet, some are full-blown TOE, and some provide iWARP / RDMA as well. One would also expect to see FCoE supported longer term with more players coming into the market from the traditional FC suppliers. Value pricing will also influence just how rapidly the price declines occur or how many vendors remain at the end of the day.

So, life isn't so simple as the commentary makes it out to be nor are these vendors going belly up just because LOM solutions appear. Don't be so doom and gloom. It is going to take some number of years for this to all shake out. Pricing will play a major role in the end but so will profit margin / volume shipped.

Rick Merritt said...

Good points. 10G may not ever be a vanilla offering but have multiple flavors and thus vendors. It certainly seems to be evolving in a way that keeps layering in new functional options.

Anonymous said...

(I'm the poster who said the 10G NIC market is in for a "brutal shakeout".)


Just to clarify: I was referring specifically to the 10G NIC market. I think there are plenty of opportunities in the 10G space, e.g., switches. Cisco is vulnerable, for example, because their stock price would implode without high-margin Ethernet switches: they sell about $12bn worth of switches each year. Even if Cisco's margins were "only" 50%, that's $6bn/year in profits, or $1.5bn/quarter (75% of their profits!). If switch ASPs dropped by, say, 20%, that would be -$2.4bn straight to their bottom line, or $600m a quarter: their profits (and, thus, the stock price) would drop by 30%! Ergo, they can't lower their prices. Instead, they have to play the feature game, trying to convince customers that they should fork over the same amount of $$$ from now until eternity, getting more and more features (which nobody needs, but Cisco manages to make them think they do) in "return".


But back to NICs: I agree that the 10G market will really take off when the NIC is integrated onto the motherboard ("LOM", or "LAN-on-Motherboard", as the other poster noted), i.e., when it is basically "free". At that point, I'd expect margins to be compressed severely, and I really don't see more than a small handful of NIC chip vendors surviving...


On the physical layer, it remains to be seen if 10GBASE-T will achieve low-enough power and cost before SFP+/Twinax passive copper cables obsolete it before it has even taken off... (On the other hand, betting against xBASE-T has historically not been a bad deal...)

Anonymous said...

Yikes -- now that my comments on 10G NICs have been quoted elsewhere (see "links to this post" below -- a Google News alert picked this up for me), I should correct a typo in my final comment:

When I said "betting *against* xBASE-T has historically not been a bad deal", I should have omitted the "not", i.e., it should say "betting *against* xBASE-T has historically been a bad idea."

(Just for the record... :)

 
interconnects