A tweet from Brian Fuller led me to a new blogger that had a great post on why FPGA start-ups fail. Not only was it a good review of the basic problem from a technology it also gave a good overview of the business challenges. Oliver Coudert, who has been banging around the EDA world for quite a while, including chief technologist at the infamous Monterey Design (arguably the most spectacular flame-out in the EDA industry) pointed out that no FPGA start-up provides a significant enough advance in FPGA technology that Xilinx and Altera can't overcome it at the next process node. He also points out that with those two holding 87 percent of the market, it's tough for any company to get a serious toehold against the giants.
One thing is missing though.
Neither Altera nor Xilinx really compete directly against each other as much as they do against the ASIC world. In every presentation I have seen from either company, neither ever mentions the other. What they do talk about are the misconceptions about the limitations of FPGAs vs ASICs, and the inherent advantages of FPGA design. On the other hand, every FPGA start-up pitch I've ever heard target both the market leaders. The cost of developing a new ASIC is killing SoC startups, even when they provide a giant leap forward in performance, while the cost of developing an FPGA system is fractional by comparison. Both Xilinx and Altera attack that market and are successful.
What seems to be a constant in the FPGA start-ups is a preponderance of EDA marketing acumen. EDA start-ups target the market share of of established EDA companies, not the unserved market segments of the semiconductor and electronic systems world. The underlying belief of this practice is that they can be successful by stealing away just a few customers of the market leaders. On the other side, the EDA leaders work very hard at undercutting the startups, so they don't grow their own market either.
So the basic failure I've seen in both EDA and FPGA start-ups is more of hubris than of technology or market share.