Software is eating the world, as Marc Andreessen, founder of venture capital firm Andreessen Horowitz, said, “but AI [artificial intelligence] is eating software”, according to chipmaker Nvidia, at its annual conference in May.
As a result, there has been increased interest in all parts of the AI ecosystem and particularly the chipmakers, such as US-listed Nvidia, in which SoftBank took a stake earlier this year through the near-$100bn SoftBank Vision Fund, and UK-based machine intelligence technology developer Graphcore, which just secured $50m in its series C round.
However, the backers of Graphcore and other nascent AI chipmakers indicate potential collisions of different approaches to venture capital and how they support startups in North America compared with other regions.
While venture capital firm Sequoia led Graphcore’s series C round, following peer Atomico’s lead in its $30m series B this summer, the round also featured the corporate venturing subsidiaries of industrial product manufacturer Robert Bosch, electronics producer Samsung and computer vendor Dell.
Robert Bosch Venture Capital (RBVC), Samsung Catalyst Fund and Dell Technologies Capital, therefore, invested alongside Sequoia Capital, Amadeus Capital Partners, Atomico, C4 Ventures, Draper Esprit, Foundation Capital and Pitango Venture Capital.
RBVC had previously led the company’s $30m series A round in November 2016, just after it was spun out of chipmaker Xmos. In turn, RBVC had lead Xmos’s previous D round in 2014 a few years after it spun out of University of Bristol in the UK, before strategic investor Infineon Technologies led its $15m series E round in September this year.
By contrast, the Graphcore co-founders’ previous business, Icera, had financial investors. Nigel Toon, Graphcore’s CEO, and its chief technology officer, Simon Knowles, were previously co-founders of Icera, a 3G cellular modem chip company sold to Nvidia in 2011 for a reported $367m, after $250m of VC equity and bank debt.
But rather than an aside, the source of funding for nascent AI-focused chip companies could be important if it brings strategic options beyond the provision of capital.
A check through the funding sources of some of the most-mentioned nascent AI-focused chip producers around the world shows a pattern of VCs backing North American startups and a more diverse syndicate base backing those in other regions.
Four of the 17 AI chip startups based primarily in the US – Beyond Limits, ThinCi, Adapteva and AImotive (formerly AdasWorks) – have disclosed corporate or strategic venturing backing, compared with eight out of nine based outside North America with strategic backers.
As Qi Lu, chief operating officer at China-based search engine provider Baidu, said in a podcast with Y Combinator this month: “In the AI era, my view is that data will become a primary means of production. So, harnessing data becomes key.
“And that comes back to China, because China has a different social economical policy around it. With that, it creates an environment for developing AI technologies, and then commercialising those technologies towards market-oriented applications or social applications. It is in that context that China has a structural advantage.”
Kai-Fu Lee, founder of VC firm Sinovation, described AI in May to news provider Forbes. “Whoever has the most data wins,” he said. Goldman Sachs and a host of others have pointed out that China has the data and is offering it to companies, aided by strategic funding.
If the AI industry proves to be as important to economic growth and societies as predicted by SoftBank and other investors, then having a handle on the hardware underpinning it will be important.
The startups certainly hope so, and a host of established companies including Apple, Google and Facebook, as well as incumbent chipmakers such as Nvidia and IBM, are betting the same way and investing in AI chips, such as the Bionic chips in Apple’s iPhone X, and entrepreneurs in the AI ecosystem.
Still, AI might be eating software but the software could help shape which hardware platform succeeds.
As Jeff Hebst, vice-president at Nvidia and head of its GPU Ventures unit, told Global Corporate Venturing: “The market potential for AI is extremely large, so makes perfect sense that companies will form to build acceleration chips.
“However, most grossly underestimate the software efforts that are required – building the chip is maybe 10% of the job at best. We support all the frameworks and all the clouds. And we invest billions in every generation of our platform, which we iterate on a regular cadence.”
An analysis from Ark Investment Management in September this year found Nvidia backed the main programming codes supporting the machine or deep learning methods underpinning AI, and more than its main peers, although Ark noted that consolidation around Google TensorFlow could help level the playingfield even as it potentially consolidates.
