A post-fabless era?
"These days it's very difficult to build a fabless company making a device with any digital complexity for less than $100 million, and a number have exceeded 200 million," said Rappaport, principal of August Capital and a founding investor in companies such as Actel and Atheros.
Tape out costs can run as high as $20 million, but they are dwarfed by verification costs that can soar five times as high, he said. The payroll for verification teams alone can cost $2 million a month and silicon support teams to serve a couple OEM customers can run another million per month, he added.
Ten years ago, it was not unusual for VCs to fund a startup through a second-generation chip if the market turned bad. Not today.
"It is so expensive to do anything now that no one wants to fund a mulligan," Rappaport said. "What you conclude is the number of chip companies you can fund is a decreasing number even as the overall semiconductor industry continues to grow," he said.
Indeed, Rappaport gave a presentation last year to two groups of executives from fabless chip makers on semiconductor value in what he called the post-fabless era. "No one has disagreed, but there's a lot of frustration in the fact the data is somewhat irrefutable," he said.
Today August Capital has just one silicon startup left in its portfolio, the result of merging multiple companies it has funded. "I think there will be a handful [of chip startups] worth doing over time," he said.
"I tend to agree with Andy," said Mark Stevens, a principal at Sequoia Capital and a member of the venture capital council at the Global Semiconductor Alliance, the trade group of fables chip companies. "We've done just one new early stage startup in the last two or three years, and we've done one later stage deal," he said.
Stevens estimated it takes $40-$100 million and six to eight years to get a major chip startup to a breakeven point. "That's much larger than it was in the '80's or even the '90's," he said.
|Sequoia Capital, Partner|
Making money on such investments is tough because "the IPO window has been largely shut for startups for most of this decade," he said. "When you put nearly $100 million in a company and Broadcom or someone buys it for $200 million, if you are lucky, the math doesn't work," he added
Making matters worse, established chip companies are paying less and less for startups. The average acquisition price for a silicon startup in 2007 was $160 million. That dropped to $95 million last year, and it slid down to $65 million so far in 2009, said Lip-Bu Tan, chairman of Walden International and a veteran investor in chip startups.
Stevens pegs the falling prices on the rise of the Internet.
"The big companies are much more agile than they were, so when there's a new networking or video standard the Broadcoms and Intels and Nvidias can jump on it much quicker," he said. "They are not surprised as much by startups, so the startup window has shrunk," he added.