Some of you were along for the rollicking 1995-96 ride when startups tried their hand at Gigabit Ethernet, merged with Layer 3 switching. Others joined in for the 1998 craze, gigabit and terabit routing in the carrier backbone. The latter market even attracted a coalition of companies from Japan.

Well, fasten your seat belts, Tower of Terror lovers, because it's time for the latest batch of too many startups chasing too few seats: hybrid packet/circuit switches for service mediation. Hey, wake up back there, this really is a cool concept for optimizing packet and circuit performance when you don't have end-to-end ATM.

Castle Networks teased us with some of its plans back in October, and you'll soon be reading about some very intriguing new ideas in packet gateways from newcomers like Dynarc and TransMedia. And there's plenty more startups where those came from, some expected to spill the beans before ComNet, in Washington, at the end of this month.

Still, there's something amiss when so many brilliant young minds come up with very similar ideas, chasing a small customer base. We're already seeing some bloodletting in the Layer 3 switch and virtual-private-net box markets, even though OEMs there can depend on relatively high-volume customer-premises sales. Gigarouter and terarouter vendors have to rely on carrier central offices and Internet service provider points of presence, and there aren't many COs and POPs to sustain a lot of startups.

Hybrid-switch vendors have it even tougher, chasing after the limited number of gateways where packet- and circuit-switched systems meet. The only winners will be semiconductor vendors that sell specialized processors for accelerating packets and circuit emulation-newcomers like T.Sqware and Softcom Microsystems Inc., and old familiar names like Motorola, Analog Devices and Conexant (you might know that last one as Rockwell). An OEM trying to appeal to a carrier with a special switch architecture, while five other startups clamor for the same space, is going to run into some numbers trouble-just do the math.

At last November's Next Generation Networks meeting, an investment panel concluded that only software and semiconductor vendors will make money in the WAN. We know the former is true from the end-of-year Internet stock frenzy, and the chip vendors have the advantage of being neutral arms merchants, selling to all sides. The more the hardware OEMs tailor systems for broadband traffic, the fewer systems they will sell.