The winners and losers in the market for switch fabric chipsets have almost been decided, according to The Linley Group, which published a report on the topic today.

The winners are likely to be Applied Micro Circuits Corp. (AMCC) (Nasdaq: AMCC) and Agere Systems Inc. (NYSE: AGR.A); and the losers are likely to be Broadcom Corp. (Nasdaq: BRCM) and Vitesse Semiconductor Corp. (Nasdaq: VTSS), according to Linley senior analyst, Jag Bolaria.

Among startups, only five companies still have a chance of success, according to Bolaria: Dune Networks, Erlang Technology Inc., Sandburst Corp., TeraChip Inc., and ZettaCom Inc..Linley’s reasoning for this division of likely winners and losers goes like this: System vendors are gearing up to develop the next generation of packet switching equipment, ready for launch in 2006. They’ll start evaluating silicon by the third quarter of this year, according to Bolaria, so whoever has the best chipset right now will end up as the winner. The vendors with technology that isn’t fully baked aren’t going to make it. "No one is going to design equipment with a chip that exists only on paper," he says.

AMCC appears to have recognized this some time ago and acquired no fewer than than three switch fabrics from other vendors to get itself into pole position. Its coup de grâce was its recent purchase of IBM Corp.’s (NYSE: IBM) PowerPRS product line (see AMCC Switches On IBM's Fabric).

According to analysts everywhere, IBM's PowerPRS product is tops. In Heavy Reading’s latest report, the 2004 Communications Chips Market Perception Study, IBM was ranked first overall in the packet switch category, gaining top marks across all measurables, including price, product performance, product quality and reliability, and service and support.

IBM’s superiority is reflected in its market share. Figures from iSuppli Inc. indicate that IBM had switch fabric revenues of $50 million in 2002 or nearly half the total market.

"AMCC has a decent product from IBM, and $1 billion in the bank," notes Bolaria. "I view them as a long-term winner."

The other major player tipped to be a winner is Agere with its PI40 chipset. "In Korea, Agere is a leading supplier," Bolaria notes. "I'm surprised they aren't more successful [back home]." He thinks a reason for this may be that the PI40 has proprietary interfaces, so customers have to develop FPGAs to link the chips into their system. Agere is addressing this problem with a range of bridge chips, so it won't hold them back.

In the Heavy Reading Market Perception Survey, Agere did reasonably well, finishing in third place after AMCC/IBM and Broadcom.

Broadcom was ranked first for price and product performance jointly with IBM. Yet Linley considers Broadcom a potential loser. "Its Gemini product is tailored heavily towards a single customer [unnamed]," says Bolaria. "It probably isn't suitable for other potential buyers."

Scott Harlin, spokesperson for Broadcom refutes this. "Gemini features a standards-based NPForum CSIX interface and is suited for general availability," he says. "When compared to competitors who have switch fabrics with a CSIX interface, we believe that our silicon is heads and tails above these competitive offerings."

Linley also pinpointed Vitesse, once a high-flier with its GigaStream product, as a likely loser. The company has more or less dropped out of the game because of delays with its next-generation chips (see Vitesse's Balancing Act ). "[Vitesse] had such tremendous delays in bringing TeraStream to market, that they ended up losing design wins. The product risks being obsolete by the time it comes out."

The Heavy Reading report results reflect this. None of the 57 people that rated vendors in this product category considered Vitesse a market leader on price or product quality and reliability. It also got middle-of-the-road ratings for product performance and service and support.

Vitesse says it’s not shipping next-generation switching fabrics yet because there’s no demand for them. "We are in a wait-and-see mode (market wise) for larger TeraStream fabrics (> 160 Gbit/s)," writes Simon Keeton, director of marketing and applications, intelligent switch fabrics, in an email to Light Reading. "But the TeraStream product (160 Gbit/s) is ready to go... and is scaleable up to 640 Gbit/s."

Keeton adds that the older GigaStream product is doing exceptionally well: "We are seeing both product ramp at existing customers and new design wins at Tier A customers."

The switch fabric market is particularly tough on startups, as they need enough funding to wait until revenues ramp in 2006 or later. As noted, Linley considers Dune Networks, Erlang Technology, Sandburst, and ZettaCom as the only startups with a chance of surviving that long (see Dune Digs Up $24M , Erlang Preps Switch Fabric Push, Intel Backs Another Switch Chip, and ZettaCom Lands $19M). TeraChip has "a decent product" but will probably need more funding (see TeraChip's Retro Switch Silicon).

Startups that fell by the wayside through lack of funding include PowerX Networks (see Power X Powers Down); PetaSwitch (see PetaSwitch Kicks the Bucket; TeraCross (see TeraCross Shuts Down); and Zagros Networks. Bolaria also says that Internet Machines Corp. has cancelled its switch fabric and is currently trying to sell the intellectual property.

Internet Machines could not be reached.

— Pauline Rigby, Senior Editor, Light Reading

Archives of Related Light Reading Webinars: