The Elusive “AI Business Model” Is Just B2B SaaS
If you follow the AI narrative in the popular business and finance press, you’ll see a constant drumbeat about how AI is hemorrhaging money with no obvious profitable business model in sight. The logic usually looks like:
Foundation models are expensive to train
Inference costs eat margins
Competition and low switching costs drive prices toward zero
And who wants to pay for a chatbot, anyway?
So the tech industry is left arguing our general feeling that “AI is really important,” while the Wall Street types argue their general feeling that nobody will make serious money from it.
In my view, the problem with this framing is the assumption that there is any such thing as a “pure AI business model.”
AI isn’t a business. It’s an enabling technology for better SaaS.
AI Expands What You Can “SaaS”
Smart practitioners will notice that AI unlocks new workflows that have historically resisted software automation.
Think about legal operations, procurement, vendor reviews, compliance, claims processing. Past attempts to build SaaS in these domains have been spotty at best, because at some point the workflow hits a bottleneck that requires human interpretation and judgment.
But AI is already surprisingly good at mimicking those human layers.
And as a result:
AI turns previously human-only or human-mostly workflows into software you can actually sell.
Real AI-Native SaaS
Thanks to AI, entire categories of work that used to depend on an amalgam of PDFs, emails, spreadsheets, and tribal knowledge suddenly become addressable by software. Someone is going to make a lot of money here.
So far, the closest things to “AI-native SaaS businesses” fall into one of three categories:
LLM APIs
Domain-tuned chatbots
Various software development tools*
(*I put an asterisk by development tools because dev tools have historically been a terrible business for reasons unrelated to AI. I may come back to that in a future post.)
Meanwhile legacy SaaS companies are bolting AI onto existing systems by throwing a chatbot in here and an auto-summarizer there. Maybe they’ll innovate further with AI. Maybe they won’t.
A real AI-native SaaS company will look a lot more like Salesforce, Workday, or ServiceNow but minus the legacy baggage and with entirely new AI-enabled ways of working. I think we’re still waiting for one of these businesses to truly break out, but I have no doubt they’re out there and on the way.
But What About Better AI Models?
Training costs and GPUs are where “AI bubble” alarmists focus, and that’s entirely valid if you’re trying to be a model provider or you’re foolishly pursuing “AGI.”
Pro tip: You don’t want to be a foundation model provider, and you really don’t want to be pursuing AGI.
You don’t need to train your own models to build a profitable AI business.
In fact, if your non-AI-infrastructure business is training custom models or running its own hardware—as opposed to smartly implementing, say, retrieval-augmented generation—there’s a decent chance you’re doing it wrong. I’m not saying it’s never appropriate, but leaders should be at least somewhat skeptical when their engineers confidently assert that these things are “required” and not the result of resume-driven development.
The existing general-purpose models are already good enough to tackle a large number of SaaS business problems. And improvements in foundation models increasingly look incremental. Meanwhile, inference efficiency is improving at something like a Moore’s-Law pace. As a consequence, instead of getting squeezed, an AI-enabled SaaS company can actually expand margins as the underlying tech gets faster and cheaper.
Most Billion-Dollar AI Companies Won’t Look Like OpenAI
If you’re trying to train and monetize your own models, then yes, life is hard.
But if you’re using AI to deliver a vertical workflow outcome to traditional enterprises? That business looks just like the best parts of SaaS:
Recurring revenue
High switching costs
Deep organizational integration
Genuine system-of-record moats
Directly measurable ROI
AI is not a “product,” or a “feature,” or even a “platform.” It’s the new technological basis for SaaS.
If You’re Trying to Navigate This
If you’re a founder, engineering leader, or executive trying to make sense of where AI actually fits into your product roadmap (or how to avoid wasting time and money on the wrong technical bets) I help companies think through exactly these questions.
If you’d like to talk through your strategy, feel free to reach out.
