Article Info

Cisco’s Fear of a Broadcom Planet

brcm There are few franchises in the silicon business where one vendor so completely dominates the market. Everyone can name Intel’s CPU business as one. But can you name another?

Infineon/TI (IFX) in DSL is a contender. Conexant (CNXT) in PC modems (but little profit to show for it). Netlogic (NETL) in exotic CAM’s (but beholden to one sugar daddy customer – Cisco).

But nothing approaches the complete and total dominance of Broadcom’s (BRCM) grip on Ethernet switching silicon.

Broadcom and Marvell (MRVL) are the Intel and AMD of this market, with Marvell occasionally scoring tactical victories as an insurgent force but never toppling the regime. Broadcom pulls in about $1B a year in Ethernet switching and are the Microsoft of switching silicon.

As big as Broadcom’s Ethernet switching business is, it would double in size if Cisco (CSCO) moved to use outside silicon. Cisco represents about 60% of the market for Ethernet switching equipment and has doggedly pursued ASICs for the majority of it’s systems. The reasons for doing in house silicon are debatable, but the general consensus is that it prevents companies like Huawei from easily cloning their systems.

I previously highlighted the risk Huawei/3Com (COMS) poses to Cisco based on their strong success in China (see Dr. Strangelove, Or: How I Learned To Stop Worrying and Love Huawei). Cisco has about 60% market share everywhere except China, where they have 40%, and the difference is Huawei/3Com. It is interesting to examine what threat Huawei/3COM poses on a macro scale.

Cisco can afford to fund their own ASICs because of the premium prices they charge. It is generally believed that buying standard silicon on the open market would be a cheaper solution for Cisco but would open the risk that their products are cloned by vendors such as Huawei.

It would be very positive for Broadcom if Cisco faced pricing pressure by vendors like Huawei/3Com, and lost market share as a result. Either Cisco would be forced to move to standard Broadcom silicon for cost savings, or concede market share to vendors that use Broadcom silicon. Either way Broadcom sells more product, and if the market fragmented away from Cisco they might even command more margin.

This realization came as a shock, as I always felt Broadcom was a good company but a fairly valued stock. But if Cisco’s shift from ASICs to outside silicon takes place, which I think will in the next 3-4 years, it has the potential to add another $500m-$1B in revenue at Broadcom. I am sure Cisco would manage such a transition delicately, and do their best break the Broadcom dominance, so perhaps Marvel captures a fair share of this windfall.

A failure by Cisco to move from in-house to standard product silicon already has a historical proxy. DEC failed to move from in-house processors to Intel standard products and ultimately their dominance of the computing market was toppled.

In short, as dominant at Broadcom is in Ethernet silicon, the reality is that 50% of the market remains untapped. Eventual market or technological changes will force the inevitable conversion of Cisco’s Ethernet business to outside silicon suppliers.

Author owns shares in COMS. Public Enemy album cover modified under fair use provisions. Modifications copyright Nyquist Capital LLP


Comments are disallowed for this post.

  1. Thanks Andrew, interesting article.

    Can you comment on the role EZChip plays in the context of these dynamics?

    Posted by mark seery | August 20, 2007, 12:09 PM
  2. At least for public discussion, I’m saving my opinions for October at Gilder’s Conference.

    Posted by Andrew Schmitt | August 20, 2007, 1:36 PM
  3. Great Analysis.

    I suppose that the impressive $1B a year in Ethernet switching is roughly 25% of Broadcom’s revenue. How do you view their position in the broadband business?

    Posted by doron tal | August 20, 2007, 2:22 PM
  4. There is a small company called Dune which is actually taking over most of the big designs. I think they partner with Marvell in some way, providing them access to their technology.

    Posted by Andrew | August 21, 2007, 1:19 PM
  5. You are not speaking to a big piece of the puzzle, which is phy’s (the physical layer transceivers).

    Posted by John Blanc | August 21, 2007, 10:58 PM
  6. With 50% market share, Cisco has more ethernet switch chip volume then BRCM. With these volumes, ASIC development cost is a very small percentage of device cost. You imply BRCM ASP is lower then Cisco cost. What makes it so?

    Posted by Jon Loewen | August 22, 2007, 12:23 PM
  7. Broadcom does well in PHYs. Vitesse had a better PHY for a long time and in the end it made no difference.

    You are correct Jon in that the fixed costs are small, but I would venture Broadcom’s are less, primarily because they have scale from other businesses they participate in.

    I am not trying to imply BRCM ASP is less than Cisco cost though it may be. Once you roll in the fixed costs Cisco runs then their solution must be more expensive.

    BRCM, if it chose to, could sell at a steep discount to Cisco since it’s marginal cost of doing so would be low.

    The real issue here is Cisco thinks they can wall themselves off from the world and out-innovate Broadcom. And then maybe they don’t. Either way, eventually Broadcom or someone else is going to leap ahead of the state-of-the-art and Cisco’s ASIC team will be caught flat footed.

    There are a number of scenarios where the ASIC business goes to ASSP. But it is hard to see what scenarios allow the business to stay ASIC in the long run.

    Xenophobia is ultimately a terminal disease.

    Thoughts welcome.

    Posted by Andrew Schmitt | August 22, 2007, 2:24 PM
  8. I am not sure that Cisco will transition to Broadcom-based silicon in a significant way anytime soon, for two reasons:
    1) Cisco is able to command premiums due to a premium feature set that includes both legacy support as well as advanced features that are not matched by Broadcom or Marvell

    2) The amount of software and QA investment required by Cisco in order to overhaul their entire switching product line has a very poor (negative?) ROI. Not to mention the immense support costs that would be incurred supporting the switch over.

    While such a wholesale change is not quite like making the leap from IPv4 to IPv6, if you dive deep into Cisco L2 / L3 switching, you’ll see that there are enough proprietary, inter-device protocols that will require wholesale changes to product families, which means Cisco cannot gradually make the shift.

    The transition could have happened 5-8 years ago, but it has not because of the reasons I mention above. I also spoke with a former Cisco product manager for stackable Ethernet switching a few years ago about this very topic, and he echoed these comments as well as mentioning that at the very low end (Cisco, not Linksys), Cisco does use some off the shelf silicon. Linksys utilizes both Marvell and Broadcom silicon, of course.

    Posted by Albert Lew | August 22, 2007, 10:46 PM
  9. Albert – you make good points about why it would be difficult to switch. My point is that Cisco won’t switch because they want to – it will be because they have to.

    Linksys uses Marvell, Broadcom, and Vitesse.

    Another thing to consider – the growth of market volume will take place at the low end and drive more share to Broadcom/Marvell as a result.

    Posted by Andrew Schmitt | August 23, 2007, 11:14 AM
  10. You are simplifying this issue when in fact there are many more variables.

    First of all, Vitesse may have a better Phy than Broadcom NOW, but that is irrelevant. In 10/100 and n Gigabit, Broadcom was one the FIRST two vendors with a working Phy. That, and the ability to hold on to the business is all you need.

    Secondly, the switching function is essentially not an analog or mixed signal function, which is BRCM and MRVL core compentencies.

    Where has BRCM or MRVL shown that they can excel at a non-analog software intensive market.

    BRCM has done well because they integrate their phy’s in the switches. That low end switching business is a cost per port issue, thats all. The switching functionality provided is not innovative at all.

    As you pointed out, getting good phy’s, ie Vitesse, can be done. Cisco does not integrate their phys in their switching silicon. It is a toss up if there is an inherent cost advantage to doing this. The delineation point is IEEE standardirized.

    Thirdly, the development cost of ASICs is expensive. Which is why you see the use of FPGAs and ASIC migration path silicon being used more and more.

    Lastly, I agree that BRCM/MRVL will creep up from the bottom. As the industry matures, what is needed at the bottom is strictly a low “cost per port” and not “networking” innovation.

    Posted by breedan | August 26, 2007, 4:30 PM
  11. Good points Andrew.

    Here are a couple points that I would add.
    * Product Segmentation — fixed format and modular boxes.
    – modular – Cisco’s owns the high margin modular box business ( probably > 80% share). Historically the ratio for these boxes is something like 80% of the revenue and 20% of the volume. Cisco killed everyone way back in the late ninety’s and the Cat 4k/6k series shows no sign of giving back share. FFIV, Foundry, Extreme try to go higher end, but struggle.
    – fixed format — Cisco is a big player, but it doesn’t have enormous share. This is the market that Broadcom dominates. The boxes are all from Asia. It is a huge volume market and prices/margins are very low. For example, I bought a MRVL based 16 port 10/100/100 fixed format box from Dell for ~$200.
    * Volume — it is my understanding that cisco is one of the biggest ASIC developers in the business. They develop more chips than most of the comm semi vendors combined. If Cisco can’t afford to develop ASICs for their flagship products then no one should develop chips for this market.

    Posted by iain | August 30, 2007, 11:36 AM
  12. maybe Vitesse is trying to take advantage of fear. Noticing that they has a job opening lately that is specifically targetted to CSCO account.

    Posted by yf | September 2, 2007, 1:39 PM
  13. Trackbacks / Pingbacks

  14. Google’s Secret 10GbE Switch at Nyquist Capital | November 16, 2007, 3:27 PM