After years of speculation about a possible IPO or acquisition, Force10 Networks finally made a move this week, announcing a planned merger with Turin Networks. Force10 has historically focused on the 10 Gigabit Ethernet market, and would have a hard time surviving the telecom bust had it chased after the Gigabit market dominated by Cisco (CSCO), Extreme (EXTR), and Foundry (now part of Brocade (BRCD)). But with a new IEEE standard coming out next year promising even higher line rates, many are now wondering if a startup will be able to grab share from the traditional switch vendors by focusing on 40 and 100 Gigabit Ethernet. However, this is unlikely to happen.
New Technologies Used to Create New Vendors
Historically, the advent of a new data networking protocol or a higher line rate for an existing protocol has created major opportunities for equipment startups. ATM and Frame Relay, for example, allowed Stratacom, Ascend, Cascade, and others to compete with large vendors, at least until they were bought out by their larger peers. When large corporations and carriers started deploying routers, vendors like Bay Networks (bought by Nortel (NT)) and Cisco were able to deliver products still unfamiliar to many of the industry's traditional names. MPLS allowed Juniper (JNPR) to take share from Cisco in the carrier routing market by promoting an alternative to the late 90s standard of running IP on top of ATM.
And within Ethernet, the standardization of Gigabit line rates opened up opportunities for Extreme and Foundry, while 10 Gigabit has allowed Force10 to find its footing over the last few years against much larger competitors. However, neither 40 nor 100 Gigabit Ethernet is likely to offer the same opportunity to a data networking startup, and this inability to gain on entrenched Ethernet vendors will have nothing to do with VC interest in the industry, the state of the overall economy, or the distribution strength of established suppliers.
Programmable Chips Have Changed Network Economics
One of the main reasons this will not be fertile ground for a networking startup is the growing use of FPGAs in networking systems, supplied mostly by vendors such as Xilinx (XLNX), Altera (ALTR) and Lattice Semiconductor (LSCC). This has changed the typical R&D process for a data networking startup, which over the last 20 years has been to develop a proprietary ASIC, build it into a proprietary box, and then surround it with standardized transmission and backplane protocols. However, with ASIC design costs rising from about $10 million at .25 microns (250 nanometer manufacturing process) to over $100 million at 65 nanometers, it is no longer economically feasible for a venture-funded equipment supplier to design the ASIC, write the network management system code, develop the backplane architecture, and sell the box.
When 802.3ae (10 Gigabit Ethernet) was ratified in 2002, .13 microns was still the state-of-the-art manufacturing node, when 802.3ba (40 and 100 Gigabit Ethernet) is ratified in 2010, 45 nanometers is likely to be the state-of-the-art node, and FPGAs with this technology are available today. This does not mean FPGAs will crush all other logic chips in delivering 40 and 100 Gigabit technology, but that any new fixed-function chip like an ASIC or ASSP will increasingly need multiple vendors over which it can recover nonrecurring development costs, which was not true five years ago.
Still Opportunity for Chip, Component, and Application-Specific Startups
While high fixed costs are likely to prevent a 40/100 Gigabit equipment startup from developing, the same is not true in the chip and component industries, especially with higher line rates introducing all sorts of new challenges to memory buffers, packet forwarding latencies, and transceiver designs. While there probably will not be another Force10, there might very well be another NetLogic (NETL), a chip vendor whose revenue increased rapidly with the growth of 10G four years ago as switch manufacturers made heavy use of its TCAM technology.
An equipment supplier that serves specific niches within 10, 40, or 100 Gigabit could find opportunities by developing products for a particular application. By focusing on data centers and blade computers, for example, Arista Networks and Blade Network Systems are not spending tens of million in R&D costs that would be needed to create a carrier-class Ethernet switch or router.
Old Faces in Familiar Places
Supercomputing clusters, electronic trading networks, and long-haul research networks were some of the first places 10 Gigabit appeared, and many are likely to get upgraded to 40 and 100G as new InfiniBand and Ethernet switches come to market. However, the nameplates on those switches will look very familiar to investors.
Disclosure: Some of the companies mentioned in this report may have purchased published research from my firm, however we do not offer consulting services to individual companies.