The software industry is a very dynamic environment. According to some estimates, there are currently over 27,000 active SaaS companies, and the number keeps growing.
With all this activity, it’s easy to forget that the current business model of software as a service is already the third fundamentally different way in which software is sold.
It’s more than likely that we are on the cusp of entering a new era, the fourth major wave of software. Of course, this new innovation is driven by AI and its many new capabilities.
The Evolution of Software Business Models
The Mainframe Era (1960s-1980s)
Back in the days of mainframe and mini-computers, software was rarely seen as a standalone product. The hardware makers — led by the dominant behemoth IBM — typically sold complete systems with everything included. You bought (or leased) a huge mainframe computer, database software, business applications, consulting services, and maintenance all from one vendor. Over time, a cottage industry of third-party software and services companies emerged, offering specialized solutions for particular vertical industries. But the main vendor you chose dominated everything. You were an IBM or DEC shop, and everything else followed from there.
The PC Era (Late 1970s-1990s)
The personal computer, emerging in the late 70s, changed the game fundamentally. After an initial era of integrated systems, the wave of IBM-compatible PCs brought a new model to life. PC hardware from multiple vendors was able to run the same standardized operating systems. Initially, there was a choice of multiple OSes, but after a while, Microsoft DOS and later Windows emerged as the dominant standard. PC users were then able to pick from a variety of application software products, which led to an explosion in the number of independent software companies.
In this era, software was sold as a pre-paid, locally installed product — you paid an initial price and were then able to use the software for as long as you wanted. However, you had to buy and maintain the necessary client and server hardware yourself, which added cost and hassle. If you wanted to upgrade to the latest version of a software product, you often had to pay full price again. In economic terms, software became a very traditionally structured product — pay money, get the product. The advantage for software vendors was that they got paid in full up-front, but they had to earn the right to get more money from existing customers with every new version.
The SaaS Era (Late 1990s-Present)
Starting in the late 1990s, the Internet enabled the next model for the software industry. Universal connectivity made it possible to not buy, but rent software products. Software-as-a-service (SaaS) emerged, in which you use software that is hosted on the vendor’s own server infrastructure and pay for it in typically monthly fees. This is very convenient for customers — they don’t have to worry about server hardware, and they don’t have to pay a large amount of money upfront.
Software vendors like the model as well because they’re getting recurring revenue that makes it much easier to plan ahead compared to the very chunky model of pre-paid software licenses. And another problem went away: Software piracy. Installed software can be easily copied and illegally distributed, but SaaS doesn’t have this issue. The only problem with SaaS is that software companies don’t get paid quickly but are only recouping development and customer acquisition costs over time. This made running and scaling a software business quite a bit more capital intensive, but the VC industry was happy to help.
Software in the AI Era: What Could It Look Like?
Few people doubt that AI will transform the way we think about software. But it’s anything but clear what the new dominant model will look like. There are currently four possible models being discussed in the industry.
1. Enhanced SaaS Products
The most conservative theory says that AI will simply make SaaS products more capable and therefore more valuable to customers. In other words, we might not see a revolution in how software is bought, but customers will get higher ROI out of it and therefore will be willing to pay more. An example would be the efficient processing of unstructured documents in many verticals, for example in healthcare, supply chains or finance.
2. AI-Driven Service Companies
Another theory postulates that AI will add so many new opportunities for automation that the very concept of software will become less relevant. Instead, companies will emerge that use AI to provide a service or solution holistically at much greater efficiency and/or quality — “selling the work” instead of selling a product. For example, selling AI-based software to law firms might not be very interesting. Instead, entrepreneurs could build a next-generation law firm that is able to provide much better and cheaper services with AI.
3. Democratization of Software Creation
Yet another theory focuses on the potentially dramatic productivity impact of AI on software creation. In this world, making software is becoming so democratized that subject matter experts can easily create and sell their own software products based on their deep knowledge of a particular customer problem — and thanks to AI this could be efficient even for the smallest of niche markets. We would go from hundreds of software vendors (mainframe era) to thousands (PC era) to tens of thousands (SaaS era) to potentially millions making niche software with AI. The result might look a lot more like highly fragmented service markets than the software industry we know today.
4. Obsolescence of Traditional Software
Finally, some of the more extreme theories assume that AI models will become so good and universally intelligent that something that we would recognize as software will become entirely obsolete. At the end of the day, software is just a tool to support a particular mental task. If a machine can get so flexible and intelligent that it can solve most problems right away, you don’t really need specialized software anymore. If you’re a salesperson, why would you need a CRM system if your AI assistant can just follow, remember, and analyze all your communication and tell you who you should reach out to next and what you should tell them? And for that matter, why even talk to a person if your AI can communicate with the prospect’s AI and figure out if they might need your product? Obviously, thinking this through further can lead to some quite mind-boggling scenarios.
Conclusion: Interesting Times Ahead
If history is any guide, it is extremely difficult to predict exactly where we will end up on this issue. We may see some elements of all of these scenarios, and there may be others we can’t even imagine yet.
To put a possible timeline on this based on previous waves: Salesforce, arguably the first major SaaS company, was founded in 1999. It took about ten years for the modern SaaS model to be fully baked, leading to a Cambrian explosion of companies in the sector. The familiar mechanics of SaaS businesses that we now take for granted had to evolve over that time.
Much depends on the speed of technological development. As impressive as current AI models are, we are still very far away from a universally intelligent AI assistant. The underlying capabilities of the AI might not even be the biggest obstacle for the implementation of this vision. Integrating all the many different data sources we all work with on a daily basis is a very complicated thing, but it’s the precondition for any major steps towards AI assistants. There is a lot of technical and organizational friction to overcome, and that will take time.
Even more importantly, the true economics of AI are largely unknown at this point. AI companies that build foundation models need large amounts of capital — in the billions — to buy or rent GPU capacity and get training data. Can these gigantic investments ever be justified? Even vertical AI applications that build on these models currently don’t seem to have the favorable gross margin situation that SaaS companies enjoy. Running inference on AI models is expensive, and it’s not clear yet if the resulting ROI will justify these costs anytime soon.
But these are probably just short-term concerns. It’s more than likely that AI will fundamentally redefine how we interact with computers, and the business models of the future could surprise us all.
