This post from Seth Rosenberg at Greylock really hit me where I live. For the last several months I’ve been in the trenches fundraising for a startup that is building an entertainment platform on an AI foundation.
After many conversations with VCs, I had independently found myself using the same analogy that Seth uses in the headline of his post. Namely, that investors seem to be most comfortable investing in pick-and-shovel plays. Which is to say, building platforms that other companies (presumably funded by other investors) presumably will make use of in order to actually bring in revenue.
On some level this seems logical, and smart.
You’re a VC. A founder comes to you with an idea. You could simply invest. But where’s your value add? Seems smarter to say, “hey, wait a sec, if this founder is pitching this idea, odds are that other founders have put similar ideas up on whiteboards all over the place. Investing in just one startup is high risk. But if we go meta, and invest in the picks and shovels that all those other guys are going to need, we can distribute the risk across multiple startups.”
In effect, the risk is being shifted onto the investors who are venturesome enough to invest in those other companies. The ones that aren’t making picks and shovels. The ones that are actually putting picks and shovels to work trying to extract revenue from…somewhere.
This is true in many sectors of the tech industry. How does the picture change if we focus on AI? I’d argue that this materially changes the “pick and shovel” conversation. It breaks down into how expensive the picks and shovels are, and exactly how those picks and shovels generate revenue.
How much does it cost to make these picks and shovels?
In old-school tech, the tools—the picks and shovels—can be comparatively easy and cheap to make. A small team of programmers, or even an individual programmer, can sit down with ordinary tools such as an IDE and begin building them right away. And the thing is that they are probably going to be doing that anyway as a normal part of product development.
When my team at Magic Leap set out to make an AR application called Baby Goats, we found that we had to create a debugger and a suite of basic utilities simply in order to buckle down to work. In and of itself, that toolset turned out to be so useful that Rony Abovitz, the CEO, asked us to release it as sample code so that any developer, inside or outside of the company, could use it.
So, that was a classic pick and shovel gambit. In order to ship it, we only had to pull in a technical writer to generate documentation. That’s a pretty common scenario in tech. So maybe it’s natural for VCs to want to see the same pattern play out in AI.
But AI is different. Not just any programmer can sit down and make new AI tools. These are very complicated systems that can’t work without access to vast data sets and hardware farms only possessed by large companies. Specialists who know how to build such systems are in extremely high demand. So, when a VC suggests that an AI startup pivot to a pick and shovel play, what they are implicitly asking for is for them to stop what they’re doing and attempt to recruit a team of incredibly expensive AI specialists.
But even if that succeeds, that team is going to find itself competing against incumbents that are already running big platforms that cost many billions to stand up. I’d argue that a better place for entrepreneurs and investors to focus is on putting those platforms to use in starting new businesses. Which brings me to…
How exactly do these picks and shovels generate revenue?
1. In literal gold mining
In the gold mining analogy, picks and shovels literally pull gold out of the ground.
2. In enterprise
What’s the equivalent of that in the AI world? Where does money come from?
The entire purpose of Seth Rosenberg’s “Product-Led AI” initiative is to back (metaphorical) gold miners. As far as I can make out from his site, he’s focused more on enterprise software. And I’m sure that’s for good reasons.
My focus here on Graphomane, however, is the intersection of tech and entertainment, so I’m going to come at it from that angle. What does the picture look like for creators who want to build AI driven projects?
3. In entertainment
Let’s say it’s a hundred years ago. You’re in Hollywood. You have an idea for a movie. You approach a potential financier, and pitch your idea.
If the financier is thinking like a modern-day VC, he’ll say something like this. “Unless you’re making the biggest budget movie of all time, your project simply isn’t big enough to move the needle for my LPs and so we shouldn’t be talking. However…if you’re pitching a movie to me, then there must be a lot of other people running around Southern California pitching other movies. All of those are going to require movie cameras. Movie cameras all have lenses. So, the smart play here is to go into the lens grinding business! Why don’t you go out and hire a team of world class lens grinders?”
The would-be moviemaker might counter in a couple of ways.
First of all, flip the script by pointing out that lens grinders already exist and that one can simply go out and buy lenses, as well as fully operational movie cameras. That stuff—the production infrastructure—is going to end up being commoditized. It’s not where growth is going to happen.
Secondly, ask the question of where money comes from in this business. What actually makes lenses valuable? It’s the light that passes through them. The light hits film and makes movies. Millions of people will buy tickets to sit in theaters and watch those movies—but only if the movies are telling stories that they love, featuring movie stars they identify with.
Selling the pitch to that VC, and raising the money, thus boils down to getting the VC to take a risk on a collection of intangibles that, to a tech investor, sound…crazy.
This is not a fit for tech industry VCs. It requires specialist investors. In the film industry, they’re called movie studios. In the book business, they’re called publishers.
So much for old media. In 2024, games (which I use as a general term for interactive experiences in general) have long since supplanted movies as the primary source of entertainment revenue.
This sounds great if you’re a game maker—or an investor—twenty years ago. But there’s a catch, which is well summed up by this epic thread that Matthew Ball recently posted. The very success of the game industry has created structural obstacles to its further expansion. Those come in a number of forms—too many to cover here, and anyway I can’t improve on Matt’s analysis.
What I am going to write about, though, in another post coming up soon, is what I’m going to call Platform Overhang. That’s what you get when there is too much investment in picks and shovels, and not enough in gold miners. I’m going to argue that it’s particularly acute in the world of games, because
tech investors are skittish about putting money into creative projects, the success of which is governed not only by technical achievement but by considerations of taste, aesthetics, and what people consider fun.
In games, the culture gap between platform engineers, and the people who develop on those platforms—even when those developers are engineers—is deceptively wide.
game platforms are becoming very sophisticated. This limits the pool of developers who can productively use them, and raises the cost of building a first playable.
AI is a salient case study that helps make this clear. But whether we’re talking about AI systems or other complex platforms, I think that the end result is that platform owners could reap more profits from what they’ve built if they showed more love to startups trying to build businesses on their systems.