Infinera Inc. remains cloaked in mystery after raising $53 million more this week, but pieces of the startup's strategy are coming to light.

The Series D funding, announced yesterday (see Infinera Scores $53M), brings Infinera's total haul to at least $143 million, based on the company's earlier statements that it had raised more than $90 million. With the extra funding, according to one source, Infinera is shifting strategy to build entire systems rather than chips and subsystems.

Infinera's valuation almost certainly fell with this round, to $150 million from possibly as high as $300 million, according to the source. But it's also likely that the company didn't suffer the massive dilution that's hit other Silicon Valley startups (see Washed Out in the Valley).

The funding jibes with earlier whispers that Infinera was trying to raise a large sum (see Infinera Shoots for the Moon), but concurrent rumors of a nine-figure backlog now seem exaggerated. In fact, it's uncertain whether the chip at the heart of Infinera's technology has made it to sample volumes yet. (To be fair, Infinera did laugh off those numbers. Infinera declined to comment for this story.)

Infinera has been a tantalizing puzzle, racking up large venture sums in 2001 and making the cover of Red Herring in 2002. But the company hasn't said much about its plans. Its founders include Lightera Networks founder Jagdeep Singh and SDL Inc. research manager David Welch; board members include prominent names such as Vinod Khosla of Kleiner Perkins Caufield & Byers (see Zepton Has an $86M War Chest and More on Infinera (née Zepton)).

It appears Infinera has been working on a way to build 10-Gbit/s long-haul links that don't require Dispersion compensation or Optical Amplification.

According to one source requesting anonymity, Infinera was founded to build an indium phosphide (InP) chip that regenerates 10 channels of 10-Gbit/s apiece. Complex enough to prod Infinera into building its own fab, the chip integrates 10 lasers, 10 receivers, and even a crossbar. It would receive an optical signal, convert it to electrical form, and pump it back out as optical -- the classic OEO regeneration.

The result is an add/drop multiplexer condensed into a chip, allowing adds, drops, and express-lane regenerations on ten lanes carrying 10 Gbit/s apiece. Using compact OEO systems based on this chip, a carrier could build a long-haul network without amplification or dispersion compensation. And because the chip is so highly integrated, it reduces the size -- and therefore the cost -- of those OEO conversions.

Reportedly, carriers one after another are falling in love with the technology, which is being demonstrated privately. Their only concern is with its manufacturability, because the chip at the heart of Infinera's machine would be quite complex.

But to sell the technology, Infinera might have to build its own systems.

Infinera originally wanted to sell just the chip, or possibly subsystems, according to the source. But optical networking OEMs increasingly are seeking out edge and even enterprise business rather than the long haul; moreover, much of the effort in long-haul networking is going into providing new line cards rather than new systems. Both trends mean less OEM attention for Infinera, possibly forcing the company to build its own systems and sell to carriers directly.

Who stands to lose if Infinera makes it? Obviously, makers of amplifiers and dispersion compensation modules would be hurt, but not immediately, as Infinera would most likely sell into greenfield networks. A tougher blow would be to vendors still pursuing all-optical networks, because Infinera may have found a way to keep the OEO alternative practical through its 10-Gbit/s generation.

— Craig Matsumoto, Senior Editor, Light Reading