Among the analysts attending, were there any new ones from major investment houses (apart from CSFB, Prudential and Lehman)?
I will try to get back to you on this.
Was the "Linux threat" the only major concern among them?
It was, as far as I recall.
Is it reasonable to assume a 230% growth rate for 4 years (to 23 million from 200 K units)?
Over 20 million is the market research I recall seeing before the conference. I think it was mentioned in the Q1 CC, for example. And it agrees with projections for DSL I am familiar with. On a global basis, it actually may be reasonable, especially since DSL will be used for SOHO as well as pure home. In five years, DSL, particularly 2nd or 3rd generation DSL, may become one of the primary Internet, communication and entertainment gateways to the home or office, not just a way to connect the PC to the Internet. Also, you should know that Kevin Landis sees DSL as a major investment theme this year.
If Wind currently owns 90% of the market, is it not likely that competition will start to heat up and erode Wind's share? 90% seems almost impossible to maintain even if Wind is the gorilla.
WIND has a huge first-mover advantage, which is the most important indicator of market success in an emerging market. WIND has the vertical solution software stacks, through Epilogue, RouterWare, Xact and 500 partners. WIND has reference designs and VAR agreements with relevant semiconductor companies. And WIND has VAR agreements globally with Cisco for DSL, as well as major working arrangements with most other major players in the space, like NT and Intel. This market position is well beyond challenge for any other RTOS vendor. If WIND were to lose share, it would have to be somehow to Linux, but not before someone assembles a competitive DSL solution. $70 million five years from now is attractive, but a competitor will have to live with a single digit percentage of less than $1 million this year -- hard to justify the huge investment necessary.
To reach your $70 million would require the average unit revenue to be $4.7 - rather high don't you think, given that many will be $1 revenue units?
My calculations don?t agree with your calculations. In CY 2003, I ended up with 21,437,803 DSL units paying royalty to WIND. I assumed 25% would be 1st generation DSLs, with the remainder 2nd generation split between the low and high royalties. Total royalty rounds to $70 million, for an average ASP of $3.25.
Was there no reference to I2O at the meeting? I am rather concerned about this. You have always been highly optimistic about I2O in the past ("I2O can alone justify Wind's current price", and during one of last year's early conference calls, Jerry F (or was it Ron A?) mentioned that I2O was finally beginning to ramp up and that this was one of the instances where the ramp-up was later than anticipated but the potential volume had meanwhile increased significantly. Now in the latest conference call, I2O was mentioned as being insignificant. What has changed?
It was Jerry Fiddler who made the remark you remember. He described the equivalent of the lily pond metaphor.
I2O was a big part of the WRS Networks presentation. I apologize for thinking that was clear from my summary. I'll clear it up now.
1. The network storage product WRS Networks is developing includes the iRAID building blocks announced January 19 by WIND and Intel. Those blocks are built on top of IxWorks and I2O.
2. InfiniBand was mentioned by the Dave Fraser, the Networks GM, as being a driver for future business. From WIND's point of view, InfiniBand implies Target Channel Adapters based on I2O IOPs. (It is possible to connect with so-called fixed-function I2O chips or conceivably by some other, proprietary means, but based on work with Intel and the I2O-SIG (which very much targeted InfiniBand), Dave believes the I2O IOPS will be the preferred connector.
3. I mentioned the off-loaded NICs, which I call iLAN connectors, or iLAN for short. Dave calls this the killer app, which makes I2O mainstream. This is the logic. High speed LANs overwhelm both servers and workstations. I think everybody agrees the only feasible solution is to offload as much of the TCP/IP stack and related processing as possible. Proprietary solutions are possible. There is even an ASIC solution that helps, but still under-performs I2O/IOPS solutions because of its limited intelligence.
The WIND/Intel iLAN solution performs at wire speed, taking most of the load off the CPU. It works with all the major OS?s, although disentangling the TCP/IP stack from Windows NT required deep tweaking. In other words, iLAN works and works well.
Intel and 3Com share the majority of the NIC market. Consequently, Intel has the market power and the solution to make iLAN the de facto standard. Intel is hugely motivated to push iLAN. (a) to sell IOPs and iLAN NICs, (b) there is synergy between iRAID storage and iLAN NICs because of peer-to-peer (the intelligent NIC can selectively retrieve/store network data without involving the host CPU) (c) iLAN NICs can house IXP 1200s giving rise to complete new categories of inexpensive NIC switches that do the work of $40K switches. (d) iLAN provides the perfect platform to house a host of vertical markets.
Just to make this absolutely clear, let me describe one I have been toying with intellectually. Erwin, please strike this from your mind after reading it since the idea is proprietary with me, and please don't tell anyone else.
First some background. Traditional Intrusion Detection (ID) is host-based or network-based (i.e. in gateways or other network devices). ID is signature-based (known virus DNA like "I Love You" in an email) or anomaly-based (perhaps a configuration or execution file changed erroneously, or something else very subtle is judged to be abnormal -- I'm leaving this vague on purpose, and also because there is a lot of very sophisticated research on-going in this area.) You can imagine that every system administrator would prefer to detect an intrusion before it reaches any host computer, thereby eliminating any chance of the virus defeating defenses on a host computer. This means an efficient network-based detection scheme would be treasured by enterprises in general and eCommerce sites in particular.
Network-based ID of either type generally is limited to pretty dumb stuff like firewalls. One reason for this is that packets, being just pieces of messages, carry less information than a fully assembled message. Since an off-loaded NIC assembles messages, it follows that the best place to conduct network-based detection would be at an offloaded, intelligent NIC. By maintaining a constant software platform in all iLANs, the market for a powerful, subtle ID vertical solution for iLANs would be huge. This observation is so obvious, so important, and so practical to implement, that you can count on this application alone to drive the iLAN market.
4. More than one and a half years ago, Intel decided absolutely not to push I2O as a "thing", thinking of it as a technology instead. A PCI bus is a thing, and IOP is a thing, but I2O is a software abstraction, a technology. Since this makes sense to Intel, if not to you and me, Intel is no longer branding I2O at all. Intel makes IOPs, RAID building blocks, intelligent NICs, and Target Channel Adapters for InfiniBand. What Intel no longer bothers to emphasize is that Intel carefully implements the full I2O standard in each of these things. That means each of these things houses an IOP running IxWorks and is plug compatible with any host computer with an OSM and, currently, a PCI bus. It is not a coincidence that the IXP 1200 is I2O ready (i.e. has an OSM) and supports the PCI bus. Even though InfiniBand presumably obviates the need anytime soon for a faster bus than the PCI bus, in the future, IOPs could be made to work with just about any hardware bus. That's one advantage to not being a "thing".
Intel's original's decision to downplay the branding of I2O followed the premature rollout for I2O. I2O wasn't ready for prime time and Microsoft was foot-dragging. But mainly the strategy reflected Intel's new infatuation at the time, NGIO, which became InfiniBand. In the bus war with IBM, HP and Compaq, Intel became defensive about being represented as pushing expensive IOPs for I/O connectivity, so Intel killed I2O branding in order to strengthen the chances for NGIO. I found out about the decision about six months after it was made. The decision-maker and technical marketing director, of the division that owned I2O at the time, told me that in five years if you asked an IT manager what she thought of I2O, she wouldn't know what you are talking about, not realizing that virtually every I/O device in her shop used I2O.
I2O itself was not killed, but simply went under wraps ultimately rationalized as a "technology" versus a "thing". Even WIND no longer emphasizes I2O, following Intel's lead while also preferring instead to emphasize the higher level solutions with much higher ASPs -- which is how the WRS vertical business units are incented to think.
Call me old-fashioned, but I still use the expression, I2O, much like I use the metaphors gorilla and lily pond. A gorilla is not just a large company. A number of attributes are required, and that's why its handy to be able to lump them all together in one word. I like the lily pond metaphor for the same reason. I don't want to say, "a market segment that enjoys sustainable growth of at least twice the normal rate of the company as a whole, in which the company holds at least a 50% share, and in which the ASP either is expected to stay constant or even increase." To me, I2O means, "a software I/O set of specifications, implemented using IxWorks running an IOP that is the foundation for an appliance or TCA for a server or workstation (or possibly some other network device)." When I talk about I2O revenues, I mean the royalty on IOPs and all licenses and royalties for verticals layered on top, hardware as well as software.
I view I2O as WIND's most advanced lily pond, but there are rapid developments occurring in broadband optical switches and possibly other market segments that happily could make me wrong about this. Nevertheless, I remain convinced that I2O can justify WIND's entire market cap with plenty to spare.
Allen |
