Want to Start an AI Company? Here’s What VCs Are Looking For Now

How can you get in on the AI startup boom? At the Web Summit conference in Vancouver, two seasoned VCs, Salil Deshpande, general partner at Uncorrelated, and David Cohen, co-founder of Techstars, attempted to answer that question.  

Is the AI boom winding down? Not at all, the investors said. “It’s pretty early,” Cohen said. “What is it, 1 percent of the world that’s actually using these tools? That’s going to be much more mainstream.”  The way we use AI will change too, he said. “We’re going to shift from a chat-based understanding of AI to the actual implementation of intelligence more broadly. I think it changes everything over a very long road.”  That long road ahead makes it hard to guess which startups will succeed and which will fail, the VCs said. But there are some areas that seem ripe for growth where these heavy hitters are currently investing.

1. Infrastructure software.  The lower you go in the software stack, the better, the VCs agreed. Other than that, Deshpande said, he asks a series of questions. First, is your product built on a database? “If it’s not, then I think you’re good, because you’re probably doing something lower. You’re probably not application software, you’re probably doing something lower.”  The problem with application software is twofold, the investors said. For one thing, the underlying AI software changes so frequently that you’re in constant danger of being disrupted. And, the VCs said, you’re always in danger of competition from what they termed “a 15-year-old with a laptop.”
2. Software where you own the data.  If your product is built on a database, all is not necessarily lost, Deshpande said. “The next question is: Whom does the data belong to?” In some cases, he said, you may generate data but not own it. For example, if you drive a Tesla, your driving data belongs to Tesla, not you, he said. On the other hand, Salesforce doesn’t own users’ data, but it’s hard for those users to migrate their data out of Salesforce because of the insights Salesforce brings to that data. “So that’s worth something,” he said. “There are other types of data schemas that are much harder, and data that’s more valuable, more voluminous. So I think that’s probably the key.”  In today’s world, the VCs agreed, data is much more valuable than code. “If you’ve just got software, it’s probably going to be easy to compete with,” Cohen said.
3. The infrastructure that supports AI.  “AI is going to be a very good customer of infrastructure for the next 20 to 30 years,” Deshpande said. “So that’s racks and machines and chips and power and cooling. All of that feels like it’s going to be good.” Anything AI needs is going to be a good investment, he added.  Because software startups have been under attack from the giant AI companies, Deshpande said he’s been focusing on hardware instead: “I’ve made half a dozen hardware investments now.”

There’s one other thing, Cohen noted. Entrepreneurs willing to bootstrap or otherwise bypass the VC ecosystem have a huge opportunity to build successful companies on a smaller scale using existing AI.  “Most investors say the thin wrapper around AI that just solves a particular problem—those all get wiped out, and we’ve seen a lot of that,” Cohen said. “But if the context is that you’re an individual who understands this technology and wants to go to the dry cleaners around your city and automate them, you could probably do pretty well investing in that. For venture capital, that’s going to be less scalable. That’s not going to create a billion-dollar outcome.”