The slow motion disaster movie that is Applied Micro Circuits Corporation (AMCC) is still in theaters even though investors stopped buying tickets long ago.
AMCC announced this morning an all cash deal to buy Quake Technologies for $69M net. AMCC has managed to spend nearly $1BB in cash on a number of acquisitions, none of which have provided a return better than simply sticking the dollars in the bank. We feel that Quake will be yet another failed acquisition.
The best thing we can say about this acquisition is that it’s an improvement over buying jet fuel for ex-CEO Dave Rickey’s Gulfstream to shuttle back and forth to Nice in order to oversee yet another failed acquisition, IBM’s Prisma switch fabric group.
Quake Technologies builds serializer/deserializers (SERDES) chips for 10 Gigabit Ethernet (10 GbE). They take multiple signals at 3Gb/s and turn them into signals at 10Gb/s. Most of these chips are sold into 10GbE Ethernet modules.
We estimate around 400k 10GbE Modules will be sold in 2006. Quake probably has around 50% market share and receives $50 per chip. Annual revenue for Quake is about 10-14M this year and that gross margins are in the 65%-70% range. There is a conference call this afternoon to review the details of the acquisition, let’s see how our guesstimate made no company data holds up.
On this conference call, expect to be wowed and wooed with tales of exploding 10GbE market size. Listeners should bring their hockey sticks in order to be adequately prepared. And it will all be true. 10GbE volumes are exploding, growing at 100%-200% for the next few years. From the release:
According to the Dell’Oro group, 10GE ports are predicted to grow from approximately 500 thousand ports shipped in 2006 to over 9 million ports shipped in 2010; a 17-fold increase in four years. 10GE is now one of the fastest growing networking technologies.
The problem is 10GbE isn’t going to explode until pricing drops. And the only way pricing drops is by eliminating components like Quake’s SERDES. Del’Oro left this little fact out when they wrote their report.
A Cisco 10GbE port that ships for $5k-$7.5k requires another $2.5k for an optics module (Typically XENPAK/X2). These optics modules are expensive because they integrate large amounts of components inside, components like Quake’s SERDES.
Cisco is the market for Ethernet and is pushing for a new optical standard called SFP+. (see ‘SFP+ – Yet another optical MSA‘ ) SFP+ is expected to cut the cost of optical modules by a factor of five by eliminating much of the active electronics (i.e. the Quake chip) in the module and leveraging the manufacturing volume of the existing SFP module form factor, which runs at around 13M units a year.
The SERDES function would then be implemented outside the module, most likely integrated into the Ethernet MAC or Ethernet switch that abutts the optical module.
AMCC will claim that this acquisition provided the needed intellectual property to enable such integration, but AMCC lacks the Ethernet products to make any use of this IP. A Broadcom (BRCM) or Marvell (MRVL) or Agere (AGR) could make ready use of this intellectual property but not AMCC.
Quake knew their long term dilemma and resolved the situation by cashing out. AMCC took the bait.
AMCC used to be a leader in SERDES technology and battled head to head with Vitesse (VTSS.pk) in the late ’90s. It is a sad commentary on the state of AMCC when they can no longer develop chips internally that used to be their core competency. Vitesse Semiconductor (VTSS) has an excellent 10G GbE SERDES – for $69M I’m sure they would have been happy to sell it to AMCC!!!
Here’s what AMCC really bought – a $10M a year revenue stream, maybe going to $20M next year with lower margins, then dropping to a few million within 4 years. I will bet $1 with anyone that AMCC never realizes $69M in product revenue from this acquisition, let alone gross margin or profit.
We’ve also heard rumblings of AMCC scooping up Intel’s (INTC) communication semiconductor assets. This could be a good deal but like all deals, it depends on the price. Given AMCC’s current aptitude at deal valuation, I suggest Intel ask AMCC to make an offer this week and cash the check next week.
I can hear the laughing from Quake headquarters all the way here in Boston. Don’t forget the banks close at 5PM.
Update – Several people have asked me if I own AMCC. I do not and have not for several years and I hold none in our portfolios.
2nd Update (4 Aug 06) – Listened to the call last night. Analysts asked all the right questions. AMCC provided revenue, margins, growth, market share guidance on Quake. I felt no need to ask questions, particularly because I am not a shareholder.
My estimates of $10M was low for ’06, and $20M for calendar ’07 is closer to their guidance of $24M.
My favorite question was why AMCC simply couldn’t build this themselves…. kudos to you if you thought that one up yourselves.
My conclusions remain the same. Quake is a good company that AMCC severely overpaid for. What expected rate of return is AMCC using when they pay 69MM? What does this acquisition contribute to profits in ’09? ’10?
You may be right, but it would be doubtful that you are right for the reasons in your post.
And the only way pricing drops is by eliminating components like Quake’s SERDES.
Huh? The components are not eliminated, otherwise the function is entirely different. Various components would be combined, at least. During this combining, you need a SERDES core, and putting one together @ 10G in a cheap Si technology is pretty damn difficult.
The SERDES function would then be implemented outside the module, most likely integrated into the Ethernet MAC or Ethernet switch that abutts the optical module.
And then run the 10Gb/s line to a pure optics module? Huh? Do you think there might be a reason why the computer industry doesn’t do this already?
It may be that your post was hastily written, but it seems to me that you fail to understand the technical purpose of an optical module:
Since running high-speed electrical signals is difficult and costly on a PC board, you break the electrical signals into several slower-speed signals running in parallel. You can wire this bus around your box (router, switch, computer) with greater technical ease.
Eventually, you need to come out of the box. At this point, you must conform to an existing standard, and single-signal optical is the predominant choice. That means that you must combine the slow-speed parallel signals.
If you combine them @ the MAC or away from the optics, then you are stuck with running the high speed signal around inside your box. Avoiding this routing is why the lines are parallel to begin with.
You show a stunning lack of knowledge of this space.
Where is cost in 10G optics? Not in the SerDes as you say, but in the transmit OPTICS. The lasers, buddy, that’s where the cost is. This is why XFP is *still* expensive, even though the SerDes (PHY) is on the system board and not in the module.
Does the SerDes go away as you say with SFP+?
Well, yes and no.
The Serializer/Deserializer may get integrated into the switch chip on the system board (over time, but not for a few more years), but to do this requires serious signal processing on the system board. Who does this signal processing?
That would be Quake.
Look at who the biggest contributor to the SFP+ standard is…it is Quake. If it is so bad for them (as you say), why would we (yes, I am a Quake employee) push it? [Did you bother to *check*?]
Quake recently announced a PHY chip for SFP+ (guess you missed that one, shocking based on the lack of research here) that includes this signal processing capability. We’re the only ones with this technology today and expect to be for some time.
Back to making 10G cheap:
Our signal processing technology allows the use of MUCH cheaper transmit optics (optics normally used for 4G, for example). Since the laser side of the cost is 10 times anything else involved. You make the laser cheap, you make the solution cheap. And cheap meand high-volume.
But you’d have known this had you done any research.
-MAK
Steve – I know what a SERDES is. I’ve designed them in GaAs. And I’m well aware of the purpose of an optical module.
SFP+ proposes exactly what you say is not possible – using 10G serial to talk directly with the module. XFP already does this today.
As for you Mitch, there are a number of companies out there that can do the retiming and processing required for SFP+ EDC.
I am aware that you are a big contributor to SFP+. You would have to be in order to survive the transition away from standalone PHYs. There are also a number of companies that have the same retiming technology and I don’t put a big price tag on that IP. How much did Big Bear get for their EDC IP? A million dollars? Less?
Face it, you got lucky when Cisco stayed with XENPAK/X2 as long as they did and didn’t move to XFP. Your luck is running out.
Remember, this isn’t an indictment of your technology. Just the price tag.
SFP+ proposes exactly what you say is not possible – using 10G serial to talk directly with the module. XFP already does this today.
Your reading comprehension is not winning you points. Presumably, this is indicative of your general reading comprehension?
I didn’t say it was not possible, I said it was difficult and costly. Since you’ve done III-V designs, you understand what I mean by that.
Customers that bought your III-V designs paid a premium for the function because (presumably) they either needed the additional performance that III-V could provide, or perhaps your designs were in a period of time before Si technologies could handle the speed.
Either way, that experience obviously did not prepare you to the total system design issues (both technical and cost).
Face it, you got lucky when Cisco stayed with XENPAK/X2 as long as they did and didn’t move to XFP.
Hmmmmm. Might be a hint here. What do you think?
Your luck is running out.
Oddly, this is the only thing I agree with, but it is not for any of the reasons that you posted.
FYI: I am an employee of neither quake nor amcc, if it matters.
You say “…you would have to be [a contributor to SFP+] to survive the transition away from standalone PHYs…”, but I guess you’re not aware that we have been evangelizing SFP+ for TWO YEARS.
We must be doing this because we are insane and bent on our own destruction (according to your logic). Or maybe, we know a bit more about the market than you do.
And sure, it *must* be luck that kept Quake in their market position for the past 4 years. We were lucky that Broadcom came in and we beat them, Marvell came in and we beat them, and your illustrious alma-mater Vitesse tried to come in – and we beat them too. Damn, we’re lucky guys.
And where are all those guys who can “do the same thing” we can do on the signal processing side? I don’t see any other SFP+ PHYs out there, just some old semi-functional EDC chips that don’t even do 10% of what is needed. Vitesse got what they paid for in BigBear: a SiGe BiCMOS monstrosity with no business prospects.
Guess they just weren’t lucky.
– MAK
Talk to me in a year Mitch. And take my $1 bet.
I’ll take a $10K bet. Put your money where your mouth is.
Mitch-
FYI: The BigBear EDC went to Infinera, not vitesse. It’s in production there.
Steve
Steve,
The single mode EDC went to Infinera, they dumped the multimode on Vitesse where it is today.
Mitch
I do Mitch. Every Day. That’s my job. Send me specific terms since technically, with my proposal, I can never win. Never implies I never get to collect. That’s OK for a $1 but not for $10k. And I want cash in escrow, not AMCC stock.
Steve – Big Bear split the technology between Infinera, Finisar, and Vitesse (not announced). Readers can fill in the details.
http://www.infinera.com/news/2005-12-06.html
Propose your terms, bearing in mind AMCC may not break out our contibution. If we can sort that detail it must include revenue from Quake’s current and roadmap products and products derived from Quake technology.
But you might want to do a little math on ASPs, market migration times, market shares and volumes before you jump into a bet. Post your results here so we (because it won’t be just me), can tell you why you’re wrong.
FYI, start with Quake’s market share. You’re way low. Next go to market size…low again.
I think that the value in quake is in their current revenue stream. They did a great job getting a part out winning xenpak biz, but aeluros has been taking some market share for the last couple of years. The was pretty clear at OFC in 2005 when aeluros had a bench setup in in the opnext booth in plain site of the quake booth. ouch!
Moving outside the module, it is going to be hard to win share. broadcom and marvell have macs and switches that give them a lot of leverage. vitesse has their 300m (bbn) edc for sfp + and aeluros still has a better solution wrt power.
Where is the quake competitive advantage? How will they win new sockets?
There would seem to be big problems doing any kind of true merger given differences in technologies (design flows,fabs are different), and markets. I can’t believe that amcc went into this thinking that they would maintain a staff of 50.
Any thoughts on the future of the ‘ottawa design center’?
Mitch:
I’ll dial into the call tonight. I’ll ask questions to clarify market share, revenue, ASPs and margin. I appreciate your feedback on my estimates and look forward to company guidance.
Make sure you tell the folks at AMCC to have operator to take my call. I see you are commenting from an AMCC ip address, so either the WAN integration is complete, or you are at an AMCC site.
You sure about the $2M a quarter. I listened to the call (hoping to hear you, frankly) and heard $5M a quarter; more like $2m/month. This number showed up in other analyst reports I saw as well.
Big difference, so you might want to check it (lest you get flamed some more.)
Pulled the replay. $2M in September, not Sept. quarter. Indicated this should be 40% of quarterly revenue, $2M/.4 = $5M. If you believe the 40% number, you are correct. I’ve edited my notes above.
The quarters are backloaded only because semiconductor companies use the last quarter to shovel product into the distrubution channel in order to meet guidance.
I don’t know why they didn’t just answer the direct question about quarterly revenue.
The question I wish I had asked, now that I think of it, is what rate of return AMCC expects on their $69MM investment.
“Talk to me in a year Mitch. And take my $1 bet”
who won this bet?
It has been more than a year now, guys! Face up the fact.
Yeah who did win this bet? Stock price is showed strength for a while but is back where it was in Aug 06.
Trackbacks / Pingbacks