Article Info

The Madness of the Carrier Chipset Market

image There is an excellent editorial today by Lee Goldberg that explores the lack of new R&D in SONET/SDH and PDH chip sets. While I don’t agree with the conclusion it is a worthy topic of exploration and he highlights something missed by the mainstream tech media.

The networking industry may be about to hit a hidden speed bump as the number of semiconductor companies actively involved with developing products to support SONET/SDH, PDH, and other TDM-based technologies can now be counted on one hand.

Lee thinks this is a big problem. This is not a problem at all. It is the only logical solution to the madness of the past 6 years.

The previous situation of having 10 companies all supplying a market growing at single digit rates was the problem.

Each of these companies sunk huge amounts of R&D (often 40-50% of annual revenues) into developing SONET/SDH or PDH chip sets only to find that five others had done the same. Each company then priced products at marginal cost (not a smart thing to do when R&D runs at 40-50% of revenue) in order to achieve any of the following misguided goals:

  1. Be a market share leader (lose money on each part but make it up in volume)
  2. Establish a strategic vendor relationship (as in ‘Let’s outsource our R&D to this desperate chip company‘)
  3. Win additional business in horizontal products (bomb the price on the high volume business to win low volume/profit business)

These ideas formed the cornerstone of PMC-Sierra (PMCS), Vitesse (VTSS.PK), AMCC (AMCC), and Mindspeed (MSPD) business strategy between 2001-2007. Equipment vendors loved it and learned to play each desperate vendor off of another. Unit volumes today are at levels above where they were 10 years ago but profitability has been completely destroyed.

No longer. Companies are refusing to invest R&D in the SONET/SDH and PDH area because the history of returns is so poor.

Either through consolidation, or self selection, fewer players are bellying up to the networking semiconductor market. Witness the destruction of the PMC-Sierra wire line telecom R&D behemoth, or as Lee Goldberg correctly points out, the depreciation of the once mighty Agere (LSI) (aka Lucent Microelectronics).

The result is a classic last-man-standing winner take all scenario. The few companies remaining face an end market with fewer competitors, better pricing, and – if you believe the endless ballyhoo of video – substantially better volumes. This is a positive development and is why we own companies like AMCC, Transwitch (TXCC),  Zarlink (ZL),

But all is not up-and-to-the-right. The damage is permanent. Big iron telco Equipment vendors are increasingly looking to chipsets used by Enterprise and even Consumer applications. Why would a vendor rely on a boutique technology like SONET/SDH when  they can slipstream behind 1.21 gigports of Ethernet shipped every month? This is why Broadcom (BRCM) will emerge as the big winner in the Telco market without even trying.

The lessons learned in the SONET/SDH area will be relearned by others. L2/L3 NPU vendors like EZ-Chip (EZCH), Xelerated, and Bay Micro as well as vendors of custom switch silicon are traversing the seven stages of denial. Virtually every IPDSLAM today uses an Ethernet switch identical to the one sitting in your office wiring closet. This trend will accelerate. Optical module vendors face an even worse situation, with Cisco purchasing the majority of devices that have little to no technical differentiation. Consolidation is the only solution.

The move to Ethernet in core networks was never about technology, it was about Economics. The carrier folks who fought the technology battles in the late 90’s are toast. It is Ethernet’s operational cost, deployment cost, and now, lack of supply chain risk that makes it so attractive. The growth in carrier traffic and massive bandwidth price erosion forced the change – not the wishes of 1999 RHK punditry.

We will see a spectacular revenue bloom as the misguided strategies of the past are buried. As Avici (AVCI) recently showed, the market misprices the effects of an End-of-Life event.

When the meteor hit, the Dinosaurs evolved into birds. The roaches thrived. It remains to be seen what will happen with the existing SONET/SDH vendors who bravely move forward. But I wouldn’t bet against Ethernet. It will outlive us all.

Hat Tip to Osxman in the Nyquist Forums for finding this great link.

Author owns positions in ZL, AMCC, and TXCC


Comments are disallowed for this post.

  1. First, I would like to point out the link in the editorial to Senator Ted Stevens’ speech on the internet. Yes, folks, this is the man charged with regulating the internet. This is why we’re all in trouble.

    Posted by osxman | July 23, 2007, 7:25 PM
  2. When I need to field a really important phone call, I ask people to call my landline. The gold standard for voice is still POTS and I’m willing to pay the difference to keep it that way. I suspect businesses keep a good many POTS lines in place even after making the switch to VoIP.

    In most places, by law, POTS tariffs are closely regulated. The political deal that incumbent carriers should be willing to make is to open up their infrastructure (such as fiber) in return for an increase in POTS rates. Instead we see them trying to usher in PON based fiber access under the cover of incumbency, when a technology neutral fiber access choice is the only way to meaningfully exploit the celebrated volume pricing of Ethernet.

    The incumbent carriers have long known that data remains the bleeding edge of communication. They should look to make money both ways – old and new, instead of losing it both ways. Investment positions in semiconductor companies are only of provincial significance against the backdrop of business opportunities on the table.

    Posted by Bandgap | July 24, 2007, 8:33 AM
  3. I think what is going on here is a good example of what Churchill said, something like capitalism is the worst system, except for all the rest.

    In the end, capitalism will get it right, but from time to time it overshoots, sometimes in a big way. I think it is important to understand that there has been nothing normal about the telecom market for almost a decade.

    Your situation of ten companies supplying a market growing at a single digit rate is a direct result of the exuberant times of the 1999-2000 period when the industry was exploding, profits were absurdly high, and venture capital money was chasing every opportunity. Under almost any circumstances, those ten companies will experience subpar returns on investment. Given the bust that followed, they all almost got wiped out.

    Today, the situation is just the same, except at the opposite end of the spectrum. The large losses of the previous five years have taken all supply out of the market. No one is doing anything, and those who do persevere are likley to experience significantly better investment returns.

    Posted by Herb Chen | July 24, 2007, 8:56 AM
  4. “Lee thinks this is a big problem. This is not a problem at all. It is the only logical solution to the madness of the past 6 years.”

    I am not sure logical is the word I would use. In either case, it probably is a problem for the industry, but it is a good thing for the chip suppliers who have hung in there. With the laws of supply and demand, nothing stays down forever.

    I don’t think the “madness” was in the last six years. I think the madness was 1999-2001.

    “Each of these companies sunk huge amounts of R&D (often 40-50% of annual revenues) into developing SONET/SDH or PDH chip sets only to find that five others had done the same. Each company then priced products at marginal cost (not a smart thing to do when R&D runs at 40-50% of revenue) in order to achieve any of the following misguided goals.”

    R&D spending was only 40-50% of revenues because revenues had declined so low. When, as in the case of TranSwitch, revenues go from $10mm to $50mm per quarter, it is hard to know what is real and what is not. Once expansion plans are set in motion, it is hard to reverse them overnight. Remember, as well, for the first year of the decline there was a sense in the industry that this was just a correction, and that growth would resume in a short time.

    As for what you say on pricing, I am not sure it is borne out by the numbers. From what I can tell, gross margins remained healthy even as volumes dried to nothing, which would imply that pricing did not decline materially. (Of course, the inclusion of previously written off inventory would also tend to inflate the gross margin line, sometimes significantly.)

    Posted by Herb Chen | July 24, 2007, 10:10 AM
  5. Andrew,

    In the end, your blog post does not really address the thrust of Lee Goldberg’s issue, and I would be interested in your thoughts.


    Will there be a significant market for continuing generations of TDM infrastructure chips? What kinds of functionality will be required and who will supply it? Will TranSwitch and the few others who are pursuing this niche have the field all to themselves? Will they be attractive takeout targets as companies realize that this can be a good business again? Or, perhaps because they need crucial elements of intellectual property? Will the economics, as bad as they have been for the last five years, be that good for the next five?

    Posted by Herb Chen | July 24, 2007, 10:16 AM
  6. Gross Margins need only drop 5-10% to have a dramatic impact when the absolute level is at 60%. Assuming COGS is flat…


    65% Gross Margins on a $100 part
    COGS = $35 Gross Profit = $65

    Assume Margins erode from 65% to 60% (not much)

    Margins now 60%. COGS still $35
    ASP = $87.5 and Gross Profit = $52.5.
    Margins dropped 5% but gross dropped 19%

    Gross Margins are a poor measurement when the absolute level is high. A few % points make a big difference.

    A true test of the theory would be a measurement of aggregate gross margin dollars for the industry, over time. Pretty easy to do too. This does not work as it ignores underlying changes in unit volume. If pricing was cut by 50% but unit volume doubled, Gross Profit stays constant. Anyone have ideas here?

    Obviously, in a high fixed cost business just a few points of margin makes a big difference.

    I do not think there will be a market for continuing generations of TDM infrastructure chips. What exists today will be good enough.

    The exceptions are transition technologies like PDH over Ethernet, VoIP trunking, Ethernet over SONET/SDH, and perhaps Ethernet over PDH. The big growth will be in FEC/OTN, ‘Smart Optics’ like EDC, and most of all, Ethernet switching. The most interesting carrier chip company right now is Fulcrum Microsystems.

    That being said, the existing generation of chips will ship for some time, with significant pricing leverage (for those brave enough to use it) capable of being applied to equipment vendors who simply cannot redesign legacy equipment.

    Posted by Andrew Schmitt | July 24, 2007, 12:05 PM
  7. “The exceptions are transition technologies like PDH over Ethernet, VoIP trunking, Ethernet over SONET/SDH, and perhaps Ethernet over PDH. The big growth will be in FEC/OTN, ‘Smart Optics’ like EDC, and most of all, Ethernet switching. The most interesting carrier chip company right now is Fulcrum Microsystems.”

    The question, as well as Lee Goldberg’s editorial, was really directed at the exceptions. I think it is probably pretty well established that there is no need for a next generation of straight framers. But is there a need for further generation of EoS framers? TranSwitch and Galazar both have Ethernet-over-PDH devices. Is there a need for another generation of these devices? TranSwitch and Zarlink have circuit emulation chips. Do we need open ended development here as well?

    Posted by Herb Chen | July 24, 2007, 1:15 PM
  8. The gross margin issue is hard to address because most of the companies one might be interested in have changed fairly radical since 1998, the year I would use for a comparison of normal margins. PMCS and AMCC have made significant acquisitions in lower margined areas. Agere didn’t exist as a separate company in 1998 and doesn’t now. VTSS is not in a reporting mode.

    For TXCC, 1998 saw gross margins of 62% or so. Today, that number would be in the 70-73% range. These numbers are not directly comparable because the company has upscaled in terms of the relative complexity of their devices.

    Posted by Herb Chen | July 24, 2007, 1:21 PM
  9. Andrew,
    I totally agree with your observations. This is simply supply and demand at work.

    However, for the future, as you said, I would not fight Ethernet. That fight is over. However, The time constant for change is high because these are changes to public infrastructure (privately owned maybe) networks and equipment and therefore any transition technologies such as those described above will last for a long time, albeit with long term modest growth that comes in bursgt mode.

    Having said alll this, Ethernet leaders like BRCM and MRVL have to make the necessary increases in complexity for public infrastructure. Such as the network maintenance, service levels, guarantees etc. This is not in the DNA of the mass market Ethernet switch companies. it is acutally in the DNA of the telecom companies.

    This leaves the question, in my mind, of who wins – still open. Companies like TXCC may be too small, but PMCS, LSI/Agere and others may have a chance if they still chose to play in this market. I am not convinced that BRCM will win or even has a prayer. If they acquire a WAN company, then that changes the game.

    Posted by Joe | July 24, 2007, 3:37 PM
  10. Andrew,
    Couple issues with analysis of “5% margin decline = 19% gross decline”

    1) This was not apples to apples, the ASP declined 15%
    2) I will give you ASP decline assumption if you accept a similar COGS decline assumtion over the some period (“things” travel downhill).
    3)In many cases prices (ASP)have gone up in the mentioned chip markets (per Herb’s commens on TXCC margins) as volume has not met expectations (jusy ask ALU after their annual “vendor negotiation” meetings in NJ recently.
    4)Outside of those mechanical issues, your points on the impact of margin on chip co’s are on target.

    Posted by JH | July 25, 2007, 10:20 AM
  11. nice discussion folks.

    one of the point that is missing so far is that FPGAs from Xilinx and Altera have been filling in a lot of the gaps left by ASSP companies.. as such FPGAs have been beneficiaries of lower volume, reduced R&D by PMCs of the world.

    As cost of development growing with Moore’s law, FPGAs will continue to hold strong in this space even with revived spending by telcos and therefore increasing volumes.

    On another note, Andrew, I am curious to know what you think about the PON ASSP gold rush – at this point, there are at-least 6 companies (probably 8 or more as rumor for new one shows up almost every month) going after the PON MAC space. Is there enough room for so many players? specially when you think of at-least two of them working on central office chips as well!


    Posted by five_whys | July 25, 2007, 10:07 PM
  12. The general trend in telecom L1/2/3 is towards commoditization. Things are more static, more quantifiable. This is not good for NPUs or FPGAs.

    At the same time there is an explosion in what I call ‘exception processing’ with deep packet inspection capabilities. This is where the NPUs and FPGAs live. The overall volume here is lower though the value is much higher.

    Posted by Andrew Schmitt | July 26, 2007, 6:54 AM
  13. great post.

    Posted by iain | July 30, 2007, 2:00 PM
  14. There seems to be the beginnings of a bubble demand for EoPDH and EoS for the wireless backhaul market. This will remain until new fibers are laid to the cell towers at which time the system will switch to Ethernet backhaul with some PDH and possibly SONET CES.

    Current chip vendors only have the higher volume cell site sized EoPDH solutions. They seem to be resisting the R&D spending to create the chips for the aggregation port (GigE) to hand off to the packet networks. This seems to leave an FPGA IP vendor, TPack as one of the few ways to create this aggregation port. The costs seem rather high for this solution (Large FPGA cost plus large IP cost = expensive solution).

    Marvell and Broadcom are VERY BUSY just competing in the enterprise markets and have little bandwidth to apply to the MUCH lower volume Telecom requirments market. They will try to put in support in their standard products but they are not looking to become PMCS (and why would they).

    Just My Personal Opinion
    This may or may not represent my employer’s position.


    Posted by jmunn | August 9, 2007, 1:05 PM
  15. Doesn’t TranSwitch have some solution with TPack for EoPDH?

    What do you make of that?

    Posted by osxman | August 9, 2007, 1:47 PM
  16. Trackbacks / Pingbacks

  17. Comments on Lee Goldberg’s - Death of the Telecom Centric skills in Chip industry « Iain’s Chips & Tech | August 26, 2007, 8:27 PM