Epistula #4: The New AI Landscape in VC
How AI is impacting our investment approach
In the world of venture capital, keeping a pulse on emerging trends and innovations is paramount. One topic that is top of mind for us is the new generation of artificial intelligence (AI), specifically generative approaches to AI and the usage of large language models (LLMs). We want to use this opportunity to share our thoughts on these technologies.
We have been hugely excited about what has been happening recently in AI. It feels like you are interacting with a human. And more generally, the excitement around these technologies, the companies developing them, and those creating them have been at a fever pitch.
The AI Wave and Our Key Reflections
We can break down our thoughts on this AI revolution into four key reflections:
This is a stepwise advance in the capabilities of these systems. The history of AI is one of punctuated equilibrium. We have been through many “AI winters.” And now we are in an “AI summer.”
It is not just hype. We have multiple portfolio companies that are using these technologies with real effect. One example is Chipper Cash, which has replaced $Ms yearly in software costs with free and open-source AI solutions.
Today’s AI is not actual intelligence – at least as humans define it. That is known as artificial general intelligence. To paraphrase Ted Chiang, what we are seeing now is applied statistics – excellent applied statistics, but applied statistics all the same. We think it is unlikely that generative AI will lead to computers' ability to engage in abductive reasoning, and therefore, the hysteria related to what we are seeing now seems misplaced.
The behavior of a lot of venture capitalists toward the LLM space baffles us. We have seen VCs beg founders to take their money, without any clear understanding of how real the innovations are, whether a specific technology will be impactful, and, most importantly, how they will monetize. And the justification – that AI is the next platform shift, akin to the birth of smartphones – seems to fall apart after even basic stress testing.
The human ability to engage in speculative excess seems to know no bounds (see crypto).
The Deflationary Impact of AI
We think AI is part of the long deflationary process of technology. I recall my first computer – a Compaq that, inflation-adjusted, cost $7,500 and needed an additional suite of expensive, shrink-wrapped software. Today, you can get a world-class computer for under $1,000 and fill it with free software. Generative AI is simply another turning of the crank of innovation in the “software eating the world” story that has played out over five decades.
For a VC thinking about the impact of AI, deflation should be top of mind. Consumers, small businesses, and enterprises all expect their software to do more every year without a material price increase. Few businesses can raise prices yearly, year after year, and not face angry customers. AI-enabled features will become de rigueur, commoditizing the value they create. The omnipresence of AI will also mean that the software we use will be much more powerful, further increasing consumer expectations of the products and services they consume.
If we are correct – that deflation and, by extension, margin compression are the primary anxieties a VC should have during this new AI summer – what is there to do about it? What do we tell portfolio companies to do? And how does it change our thinking about the investing landscape? Here are our takes:
The value of data moats is much greater in this new, AI-driven world. Being able to build a proprietary, extensible, and valuable data corpus remains the primary point of operating leverage and value creation in this new world. And the ability to create value from it seems stronger today than it was a year ago. The companies that can figure out how to acquire valuable, proprietary data at scale and then turn that into products, services, experiences, and cost advantages will have compounding competitive advantages that seem hard to beat.
No particular AI system will constitute a new moat. Most of the cutting-edge technology will ultimately be commodified and likely available to all players as an open API (like OpenAI’s) or as public domain via open source (the approach taken by MosaicML before their acquisition by Databricks for $1.3B). Also, these technologies do not meet the definition of being defensible and moat-creating – if I work at Company Y, I can hire someone from Company X to build me a feature similar to one that Company X has. If I can hire their (or similar) talent to build me a similar capability, that capability is fundamentally not defensible and does not advance moat creation. We see this today, with many people “building” and using LLMs but really just putting a veneer on already-commoditized infrastructure and public-source models.
In the long run, the current AI summer will not benefit start-ups. The companies that already have reach are most likely to acquire the data assets needed to make generative AI work for them. Microsoft, for example, is a clear beneficiary here. They have embedded AI into their products, with seemingly significant effect but without material additional costs, a classic deflationary move.
As we evaluate new opportunities for our portfolio, the importance of data strategies has come into starker relief. Having a flywheel where data comes into a company for a very low cost or free and then is transformed into something extremely valuable is at the forefront of many conversations. We think that having these flywheels is one of the best ways to fight margin erosion. And, at scale, these flywheels turn companies into enviable assets that trade at super premium prices. It is not that data strategies have not been top of mind since our inception. But their importance is magnified.
Portfolio Company Applications
In the short term, for the existing portfolio, we are starting to see investments made in the use of generative AI play out in significant ways. Let’s spotlight two that particularly stand out – Chipper Cash and SimplyWise:
Chipper Cash
With 6M customers, Chipper’s ability to deliver efficient, effective front-line customer service using AI-driven chatbots has been a real win. Rather than simply buying compliance solutions from vendors, the Chipper team created custom technology using open-source solutions. At Chipper’s scale, the switch will save the company over $Ms per year, while building up a first-party data set of verified customers, which, in turn, the company is monetizing by launching Africa’s first identity-as-a-service offering.
SimplyWise
A Fund 2 company specializing in organizing financial and healthcare information, SimplyWise leverages cutting-edge generative AI. They replaced their costly third-party optical character recognition (OCR) provider with a homegrown, highly customized one – boosting gross margins by thirty points while enhancing speed and accuracy. Additionally, SimplyWise has cultivated a data flywheel, enabling them to serve their user base more effectively as their data repository grows.
Moving Forward
As we have talked about many times, we want to find great businesses. A great one can generate high rates of return on invested capital (higher than its weighted average cost of capital) for many years, net of the dilutive effect of raising additional equity capital.
A great company must have increasing returns to scale and an ever-deepening moat in a winner-take-all-or-most industry that is very large. While we remain anchored in this tried-and-true framework, we cannot ignore the subtle shifts unfolding in our constantly evolving landscape – one that presents unique challenges and opportunities for our portfolio companies every day.
