The 7 Approaches to Companies Making Money from AI
As the AI boom continues, startups, incumbents and investors have grappled with “who will make money.” In the cloud boom, large tech incumbents (i.e. Amazon, Microsoft, Google) were the biggest beneficiaries, as well as a handful of cloud-native tools. In the mobile boom, big tech building hardware (especially Apple), app stores (again, Apple), and consumer-facing high network effect startups (Instagram, Uber, etc.) were the most successful business models.
As AI emerges, an open question remains: which business models will dominate. Will foundational models and chips take all the returns for the industry, or will they commoditize? Are vertical apps a viable approach, or do they lack defensibility? Is the best approach just to automate legacy businesses?
From private equity and large incumbents to venture capital and small startups, the market is betting on 7 core approaches to monetizing AI. This post walks through each approach, and their distinct conditions for success and failure.
Approach 1: Build “must-have” infrastructure for AI
This category has seen the biggest winners of AI so far, such as OpenAI and NVIDIA. These are companies that aim to have every AI use case built on top of them, and ultimately take a royalty from all AI usage across the economy.
This is a tough category to break into (there are only a handful of true “must-have” components of the stack), and questions around ultimate defensibility for each layer of infrastructure (will there be one foundational model that dominates or many?) — but it seems a safe bet that the single-biggest winners of AI will be a handful of foundational infrastructure players.
Approach 2: Build a vertical AI application (point-solution)
Most AI startups are vertical apps built to bring AI to X workflow, and there are hundreds of new startups being minted each month with this approach. The last few years have seen a handful of these take off with massive adoption and revenue growth — for example Abridge (healthcare), Cursor (coding), Harvey (legal services) are growing at extreme paces.
The core challenges in this category are around distribution and defensibility. So many areas are so easy to disrupt right now, that there are hundreds of companies simultaneously building “a platform to automate X.” At the same time, for most verticals, there’s an open question about whether the horizontal infrastructure players will ultimately solve most of the use cases themselves. So competition is extremely fierce. In order for vertical AI companies to ultimately thrive, they need to be in a market with a clear moat.
Approach 3: Build an AI-native company that solves a legacy business end-to-end
Rather than build a “point solution” company, some entrepreneurs are choosing to build an end-to-end service that competes with incumbents — but with a cost structure and features associated with an AI-native business, rather than a traditional business model. This tracks the dot-com boom (some of the big winners were internet-native remakes of existing business models, like Amazon.com for books).
This has been most prominent in areas like pharma, where companies focused on AI for drug discovery have chosen to become end-to-end pharma companies (that discover, develop and commercialize drugs) rather than monetize through licensing the AI itself. But it has spread beyond that; most industries are seeing a new set of AI-native competitors taking on the incumbents.
The main challenge with this approach is that legacy businesses typically have fairly strong structural advantages (such as distribution moats or customer stickiness), so even if AI can provide a 20% advantage in cost structure, that’s insufficient in many cases to overcome the existing moats of incumbents. So this only works in industries where the benefits of AI are so massive in the near-term that they can overcome the structural advantages of incumbents quickly.
Approach 4: Large incumbents automating portions of their business
One of the superpowers of generative AI so far has been the ability to automate bespoke workflows. From call center optimization to coding automation to customer service chatbots, early adopters are already seeing AI as a way to fundamentally change the cost structure of existing business workflows while keeping the distribution advantages of incumbency.
In addition to this approach being popular with existing incumbents, it is also a popular bet within private equity: acquire incumbents in industries that are ripe for AI automation, and become the first to automate. The flurry of deals around industries like accounting services, for example, are likely a bet that some of the work can ultimately be automated by early adopters.
The math on this is simple: let’s say a product generates $100 of revenue, and costs $80 to make (fully-loaded). Let’s say you can use AI to automate away 25% of the human effort; now it costs $60 to make. The company can then double its profit (from $20 to $40 per unit) — dramatically changing the business model.
The big assumption here is that prices don’t fall for incumbents as everyone takes advantage of the AI opportunities; this depends on the structure of the industry and the size of the distribution moats. The other challenges here are:
- Navigating the change management needed to be an early-adopter is harder for large incumbents
- Large incumbents typically struggle to get top-tier AI talent to lead a technical transformation
Approach 5: Combining large incumbent with AI-focused startup for automation
As a modification of the strategy above, another approach is to combine legacy incumbents with vertical AI startups for automation. This captures the same economics, but it can solve the talent gap by getting talent via acquisition.
This is especially attractive in industries where the AI tech quickly commoditizes so there are dozens of vertical AI players (so none of them can command a massive acquisition price).
The viability of this approach hinges on integration — it is complex to integrate AI-tech startups and legacy incumbents, and the thesis here involves retaining talent. Without exceptional integration, acquisitions typically don’t play out as well in practice as they do on paper. I suspect that the best PE firms in the coming decade will master this integration playbook and look for a number of acquisitions with this thesis in mind.
Approach 6: Combining large incumbent with AI-focused startup for distribution
Another acquisition-centered thesis is to combine an incumbent’s distribution moats with a startup’s AI technology. Rather than focusing on cost-cutting, this approach involves increased sales by offering a fast path-to-market for new tech.
I expect many big enterprise tech companies will do this (for example, Salesforce or Oracle acquiring companies to bring their tech into enterprise), but there are also many vertical-specific approaches. For example, in healthcare, scaled electronic medical record companies like Epic can bundle “AI for X workflow” products and take advantage of their existing distribution advantages.
A large amount of public company M&A and PE-driven M&A deals are built around variations of this thesis. The main challenges around this approach are:
- making sure there really are co-sell distribution synergies,
- making sure the integration is successful
- moving fast enough as the AI market moves quickly, and
- ensuring that the distribution advantages are long-term as the market changes.
Approach 7: Build a complement to AI
Some of the most interesting hypotheses for making money with AI come from theories of what products will be “complementary.” Rather than building an AI business, some companies are betting on non-AI products where demand will increase as AI demand grows.
Some examples of bets around this include data (proprietary data is more valuable in a world where AI processes it successfully), talent (elite talent is more valuable as it gets more leverage from AI), power (AI workflows will consume massive amounts of electricity), cyber insurance (risk of data breaches goes up), and real estate in San Francisco (an AI boom will bring lots of talent into SF).
Of course, there are counterarguments on many of these hypotheses — talent could become less valuable, and the world could move away from San Francisco — so the risk becomes in picking the industries that are true complements.
***
Across these strategies, the trillion-dollar question is which will be the dominant approach. Consider an area like legal services. Will the biggest winner of AI be AI-tech companies that sell products to existing law firms? Or an AI-native law firm that incorporates AI from day 1 in its own workflows? Or will NVIDIA and OpenAI be the biggest beneficiaries even in specific vertical workflows like law and crowd out all vertical players?
Personally, I have bet the most on variations of #7 (looking at data and talent as complements to AI), would bet strongly on incumbents with strong distribution doing smart M&A (approach #6), and am generally skeptical of the amount of hype around vertical AI apps in approach #2 (where most of the investment is going now — but where there is usually light defensibility). Massive amounts of capital and talent are casting their own votes right now as this debate plays out across the economy.
