Why AI Won't Be the Investment Opportunity Everyone Thinks It Is
The other day, I was watching an episode of Seinfeld and was struck by how dramatically different location-aware mobile phones have made our day-to-day existence. The entire plot was based around no one having a phone in their pocket. Watching the characters leave answering machine messages for people (and miss them), use paper maps, and get lost on the way to a cabin made it all seem pretty dated.
Once cell phones became ubiquitous, it became noticeably harder to date TV shows. Was this episode from five years ago? Ten? People had iPhones 15 years ago so you have to go back quite a while before a flip phone gives away a show’s original air date.
When I think about true disruptions in tech—the ones that enable huge investor outcomes because they create generational behavior change, entirely new markets, and populate whole business ecosystems out of nothing—location-aware mobile devices stand out to me as right up there with the web itself.
The venture asset class seems to have already decided that AI is the next great investment opportunity, but I’m not so sure it’s going to disrupt business and create the across-the-board wealth that has been predicted.
Back in 2004, I was working for the General Motors pension fund, which had been making limited partnership investments in venture capital since the early 1980’s. I got to see all of the top VCs pitching their funds.
What was notable was how similar they all sounded—that is, until I got the pitch from Brad and Fred at Union Square Ventures.
That was the first time anyone I’d heard anyone talk about long-term cycles of disruption, not just individual technologies, and what new business models were going to be possible because of it.
USV came in pitching digitally native businesses models that could not exist until the internet connected everyone. One particular example that Brad brought up was how Amazon was simply a physical store but on the web. The analog of Amazon was Sears—also a place where you could buy a lot of different stuff, but not something particularly groundbreaking.
Google, on the other hand, was a digitally native business model, with no offline analog. It was built on the data of crosslinking, making search quality and efficient information retrieval possible at an enormous scale.
I’ve been thinking a lot about whether AI represents that same kind of game-changing investment opportunity that stems from cycles of disruption. Despite how impressive it is to have a conversation with a seemingly all-knowing chatbot or to watch it create a Wes Anderson take on Star Wars, I keep coming back to two things:
Most AI business models are simply “better, faster, cheaper” models—iterations on existing models currently done or potentially done by humans, but perhaps not profitably.
Given that these models require training data, companies that have lots of proprietary data are at a distinct advantage, leaving everyone else to deal with what’s available in public.
Technology has already made the world pretty efficient. The idea that you’re going to get a giant internet or mobile-sized economic opportunity across the board by going better, faster, cheaper on human-based models seems a little tough to swallow—especially when you take into consideration the hype cycle, which is greater now than it’s ever been. Other than the very first checks, no one paid a remotely reasonable risk-adjusted price for shares of OpenAI unless it winds up being a FAANG sized outcome—and the same is true for a lot of AI companies these days.
On top of that, the zero-sum game of the training data prerequisite feels like a limiting factor for the asset class as a whole. While a particular one-off company might have proprietary access to some great dataset—innovation in AI doesn’t seem to be a particularly level playing field. Companies like Google, Apple, and Meta should have had a huge head start in building large language models and I suspect they’ll catch up soon. Bloomberg announced that it had built a financial version of a language model—seemingly snuffing out a lot of startup investment opportunities in that space even before it got started.
In addition, all of these better, faster, cheaper models are mostly to be around current ways of doing things—creating, scheduling, etc.—things where the stack necessary to complete the task is perhaps 10% AI and 90% a lot of other “table stakes” parts. The best place to draw with AI is very likely to still be an Adobe app using AI plugins built either by themselves, or by other companies, but those other companies aren’t going to want to rebuild all the various layering, coloring, editing and other types of basic drawing tools Adobe has perfected over the years. Again, this doesn’t feel like it presents a game-changing investment opportunity across the board the way the web itself was a pretty blank business canvas when it started.
Don’t get me wrong, there will definitely be great outcomes for companies that integrate AI into their offerings, and we might be watching today’s sitcoms 30 years from now marveling at how all this took place before AI changed everything, but I’m not quite sure the dollars are going to be there from the venture return side.
