I just don’t know what to make of Ohad Hammer’s “optimistic” post on NPU vendor EZ-ChipLanOptics Subsidiary EZchip In Bed with Cisco – Seeking Alpha. I feel like I’m getting sucked into the “timeless” NPU argument that started with the LSI ATMizer back in the mid-90’s, then moved to Maker, then to Sitera/Agere/C-Port, and now to Ez-Chip ( and Cavium ). The NPU’s past is checkered with “niche” successes that couldn’t support development costs. It is interesting that Cavium’s IPO has shown that there is “optimism” for NPU vendors with “difficult” pasts. Very curious stuff.
Hammer covers Ez-Chip’s “difficult” past history: lot of cool product evolution and > 100 design wins, but no meaningful revenue 👿 Then he introduces EZ-Chip’s view that the CESR market may be the “inflection point” for NPUs and provides a “wild” unit forecast.
This fast growing market, dominated by Cisco, Alcatel-Lucent and Juniper is expected to be worth 6.5$ billion in 2007, growing to more than 9$ billion in 2010. EZchip’s management hopes that the migration towards ultra-fast, converged networks will ultimately lead vendors to design and build their platforms based on programmable NPUs such as EZchip’s. In fact, management claims this shift is already occurring.
According to the company, each successful CESR platform from a leading tier-1 vendor can potentially generate sales of tens of thousands chips annually, while each successful platform from a smaller vendor can potentially generate sales of several thousands chips annually. Assuming an average of 25,000 units per 6 tier-1 platforms and 2500 units per 10 small platforms, the numbers of chips sold annually might be around 175,000. Assuming a price per unit of 250$, total sales could reach 43.75$ million.
I’m still not convinced the NPU vendors are at an inflection point in CESR space. I completely agree that NPU’s are an integral part of the new CESR platforms ( just like NPU’s were integral to FR/ATM Switches). These NPU’s provide the higher functionality specialty cards for the boxes. I wrote about the chip architecture of these boxes last week and noted that the “Carrier Specific” chips in this architecture have not being doing well financially. Hammer’s post appears to reflect the fact that NPUs will not be used on every card in a CESR box with the forecast of 25k units per tier-1 platform, but there aren’t 6 “tier 1” platforms that drive this kind of volume.
I also like the discussion on design win stickiness without a “peep” about how much it costs in “application support or software development” to win these designs and create this stickiness. Maker was very sticky in the ’90’s as well.
That said Ez-Chip have shown that they are survivors. They’ve built cool stuff, won significant technical designs, and won signficant customers. They are “for real”. Now they’ve just got to find a way to become profitable. This last point has yet to be done by any NPU vendor, and thus I continue to be skeptical.