Korva is a scroll of the internet’s best ideas.
Skeptics are right, until they aren't
Young people can't build chips. Don't start a semiconductor company. Experts in any hard field warn that it's too hard and the base rate is too low, and those warnings are correct almost every time. A saying passed along to one investor sums up the logic: if it doesn't work, it will be for all the reasons your partner gave, and if it does work, it will be because those reasons didn't matter.
That makes the working question what could go right, and whether a path to it is visible. Cerebras fit that question: it set out to build something people had tried to build for 50 years without success, and the team saw a reason and an application that made it worth trying now. Most of the time the reasons a company won't work are right. Sometimes they just don't matter.
How low are startup success rates?
Harvard Business School's Shikhar Ghosh studied venture-backed companies and found that roughly three quarters never returned investors' capital. The Wall Street Journal reported the study in 2012. Venture investors accept those odds because a few large winners can cover many losses. Most new businesses of any kind fail within their first decade, according to US Bureau of Labor Statistics data.
Why are chip startups so rare?
Incumbents like Intel, NVIDIA, and Qualcomm hold large customer relationships and deep engineering benches. A mistake found after manufacturing means a costly redo. Venture funds have long treated semiconductors as capital-heavy for these reasons. The analyst firm International Business Strategies has estimated that designing a chip at the most advanced process nodes costs hundreds of millions of dollars.
What earlier wafer-scale attempts failed?
Trilogy Systems, founded by computer architect Gene Amdahl in 1980, tried to build a computer from whole-wafer chips. The company raised more than 200 million dollars, an enormous sum for the time. Defects in the silicon and other manufacturing problems sank the effort, and Trilogy gave up wafer-scale in 1984. Amdahl is also known for Amdahl's law, which describes the limit on speeding up a program by adding processors.
More from Invest Like the Best with Patrick O'Shaughnessy
