I spent 12 years in the microchip business, specifically from 2000 to 2012. My friends and I built our own chip startup in 2004 and sold the company in 2008, before the Lehman shock. By the time of the exit of my startup, semiconductor startups were disappearing, due to the departure of VC investments. I remember the last massive wave of VC semiconductor investments, into a now largely forgotten standard called UWB (Ultra-wideband). All VCs got burnt. That means all founders got burnt too. In 2011, Qualcomm acquired (for $3B cash) my then-employer Atheros, ushering in the long-awaited M&A age of semiconductor firms; an era that had been long anticipated due to the escalating development cost. In 2012, Facebook hit IPO with virtually zero hard IPs. I thought I would never see VCs funding chip startups any more. I could not have been more wrong. After years of chasing one lean software startup after another, throwing hundreds of millions of dollars at them, and making the term “unicorn” more popular than Her Majesty’s favorite grandson’s first name, VCs are suddenly, back in funding chip startups en masse. This time the main theme is AI chips, or specifically, all the kinds of chips performing some sort of machine learning functions. In recent years, US and Chinese VCs (alongside corporate investors) have pumped at least $700M in total into several (relatively) high-profile AI chip startups: Mostly centred in Silicon Valley and Beijing, with a notable exception in Bristol, where Sequoia led the $50M Series C of Graphcore. In the table above, I listed the 8 most funded AI chip startups — at least those known publicly. Note that the Chinese ones are in red while the US ones and Graphcore are in Blue, simply to make it easier to read. The two groups are also ranked by the total amount raised. For those of you more familiar with lean startups, i.e. the Airbnb’s and the Uber’s, the amounts raised here don’t seem much. However, for semiconductor veterans, this has been a huge break from the market norm of the previous couple of years. Before this sudden frenzy in AI chip investments, it was very hard for a chip startup to get funded. A common misconception was that this difficulty was down to the high starting cost; This was only part of the problem and was readily contradicted by the takeoff of the 2015 wave of lean startups, who started off with more capital. The main problem was in the exit: the public market was trading most profitable semiconductor companies at a market cap-to-revenue ratio of roughly 3:1. Most hot software companies were trading at least twice of that. This meant that a $1B exit would require a chip startup to achieve $333M in revenue, while keeping all other key metrics healthy, e.g. 50% gross margin and >15% net margin. This seemed daunting and not at all attractive. The VC money, therefore, followed the exits elsewhere, chasing the next Travis-Kalanick-wannabe straight out of college instead of funding a two-time chip founder of 40 years old with a 20-page business plan.