Revelations of the Coming SaaSpocalypse

Why the SaaS consolidation is coming

Almost by definition, a B2B company exists because they sell something for less than it costs the buyer to build themselves. This is almost too obvious to state, but the premise is important. A business buys pencils from another business because it’s cheaper than creating their own pencils. If they wanted to try, they’d need to buy the raw materials, the machines that turn the materials into pencils, and the specialized labor that knows what to do. But that would be immensely expensive for something as trivial as a pencil. Instead, they’d just buy pencils for a small amount from a pencil maker. In economics, this is called comparative advantage, in business, it’s build versus buy.

B2B SaaS companies are no different in this sense, but their cost structure is. Unlike a pencil maker, they don’t require any raw materials or special machinery to make their product - the only meaningful input to making software is labor. Sure, it’s specialized software engineering labor, but in the end it’s just labor. For this reason, B2B SaaS is particularly vulnerable to what will be undoubtably be a tremendous labor saving technology: AI. This leads us to 3 premises that this essay is based on:

1. AI will continue to improve over the next 10 years

2. These improvements will allow software engineers will become 10 to 100 times more productive

3. Labor will continue to make up around 95% of software development costs

If these assumptions hold true, then we can derive that software development costs will conservatively become 7 to 17 times cheaper over the next 10 years. This means the build vs. buy decision looks a lot different for would-be purchasers of SaaS tools, and I’m having a very hard time seeing how the majority of B2B SaaS companies don’t go completely extinct.

Consider a simple CRM vendor. As a customer, I may pay hundreds of thousands per year for the right to use this software. Today, this economic model makes sense, it would cost many times my annual subscription to build a CRM from scratch. But if software development costs were 17 times cheaper, then what would cost $1,000,000 could be done with $59K. I am obviously using sample numbers, but the dynamic is the same - the economic decision of the buyer will eventually flip to favor build over buy. With AI enhanced productivity, one internal senior engineer or PM could build and maintain many times their cost in SaaS subscription savings.

The obvious foil to this prediction is that I am just making up numbers. While this is true, it’s the directionality of the trend that is impossible to argue against. Below is a chart I made that should help to visualize the dynamic.

Once the software development cost comes down to where total cost of ownership (TCO) is less to build than to buy, the seller’s business is finished. However, we must also remember that AI will make SaaS companies more productive too, allowing them to drop their prices leading to lower TCO for the buyer. This is represented by the slope of the “Buy” line. I think it is reasonable to assume that the TCO for building will drop much faster than SaaS companies will drop their prices. Remember that the lack of a marginal cost allows SaaS prices to be based on an abstract “value-based pricing” model, which is why SaaS firms have notoriously high margins. While this does give them wiggle room to go down in price (goodbye cold brew and fitness stipends!), the compression will still fail to match the drop in build TCO.

Moats

Admittedly, my analysis thus far has simplified the value of the average SaaS business to be purely about cost. While this is true for the majority of B2B SaaS businesses, it certainly isn’t for all of them. There are many SaaS tools that offer something that simply cannot be built, no matter how cheap software development gets. These are the B2B SaaS businesses that will survive the Saaspocalypse.

Let’s look at the traditional moats that exist B2B SaaS today, and why they may or may not protect businesses in the future.

Scale Economies

Scale economies are cost efficiencies gained through size and scale. As we just learned above, this may extend your lifespan, but its value as a moat diminishes significantly. This will only be a relevant moat for the largest of B2B SaaS companies; Microsoft, Google, etc.

Switching Costs

B2B SaaS tools are very sticky. Once you’re on one, it is very expensive and time consuming to switch to another. Many times companies are stuck in a situation where they have a better option, but the switching costs are too high to justify the move. In this scenario, the moat for the incumbent software tool is its high switching costs. However, as with scale economies, this is just a cost argument. AI productivity improvements for software development includes migration costs, making this moat no longer viable.

Network Effects

Having a strong network effect means that additional users of the software improve the value to existing users, creating a virtuous flywheel. The best example of a network effect is a telephone - the more people that have telephones, the more useful the telephone is to you. This is a rare moat for a B2B SaaS company to have, but there are a few examples. LinkedIn has a virtually impenetrable network effect moat that will be unaffected by AI. It doesn’t matter how cheaply you can replicate LinkedIn’s code, if you don’t have the end users, you have a useless tool.

Cornered Resource

Having a cornered resource for a SaaS business means having access to a proprietary data set that benefits your product offering. There are quite a few examples of businesses with this moat, including ZoomInfo, Bloomberg, and Palantir. Like network effects, this moat will allow a SaaS business to survive the impending cost reduction since their offering doesn’t just rest on price.

What comes next

To recap, our simplified thesis is the cost structure inversion brought on by AI will eliminate any SaaS business that doesn’t have a network effect or a cornered resource. I’d now like to explore what I think this change looks like from the customers perspective. I think this will look like one of the following:

1. The customer does the building themselves.

2. The customers pay a much smaller set of SaaS vendors (perhaps as low as one or two) that give them their entire IT stack, highly customizable, for a subscription cost less than #1.

Which of these a company chooses will depend on things like size and industry. For large companies, it will make sense for them to go with #1. For smaller companies that have lower software dependance, it may make sense to go with #2.

The future may also look like some combination of both: AI companies that sell “IT as a Service”. Instead of 100 software vendors you need one, let’s call them OpenIT. OpenIT will build all your existing IT stack perfectly customized to your company and allows you to flexibly make changes in real time (is this SaaSaaS? I hope so). The customer would likely still need some IT staff to: 1) ensure OpenIT builds the tool correctly, 2) helps steer the AI to adapt the software to changing business needs, 3) maintain and oversee ad hoc errors and requests. But even so, the IT team will need to be a fraction of its current size. In practice this looks like the final substantiation of Low-Code/No-Code.

Either way, the benefit of this to the customer will be immense:

• They’ll get exactly what they want, tailored perfectly to their business. No more crazy workarounds or “this feature is coming in Q4”. Any enhancement could be added quickly and cheaply. The full value of software would be unlocked for every business.

• Since all the tools are built internally or with one or two vendors, integration workloads would be reduced significantly. The AI could build each tool in a custom manner that allows it to seamlessly connect to other areas of the stack.

• Significant Cost Reduction. This will come from many sources, but the biggest would be great reduction in SaaS spend and software maintenance overhead (internal roles or professional services)

• Future AI compatible. Since the stack has been fully built by AI, it stands to reason that it will now be ready for all future AI advancements. For example, what will be easier to release an AI agent into, an interconnected web of 100 different SaaS tools? Or a single IT stack built custom by an AI?

Counter Arguments

Where does this argument fail? Below are a few counterpoints that I can think of.

If software development costs get close to zero, won’t that mean SaaS businesses will be able to build as much new software as they want too? The combined pace of development could make their offerings more appealing to customers.

Yes, they definitely can, but all the other SaaS vendors will be doing the same thing, causing a SaaS price death spiral since the supply for a given software solution will skyrocket. The only ones who will survive the price death spiral (who don’t have a non-cost moat) will be the hyperscalers. This is why I said scale economies is still a moat for about 5 companies. Everyone can try to become OpenIT, but everyone can’t be OpenIT. This still leads to SaaSpocalypse.

In Premise #3 you said that software costs are 5% non-labor. This may not be true in the future, and it may not even be true now.

I’ll admit I based this number on my personal experience with vibes mixed in. I think you can definitely make a case for this today - are cloud infrastructure and dev tools more than 1/20th labor cost? Maybe in some cases but it certainly hasn’t been in my experience.

AI capabilities are far away from 100xing software developers. This prediction may be true, but its decades away.

Fair enough. But it is very hard to argue that this isn’t something that will eventually happen. It’s just a matter of when. 5 years? 15? I think it’s much closer to the former than the latter.

Conclusion

In the end, given that AI improvements continue, I have a hard time of thinking of scenarios where the Saaspocalypse doesn’t happen. If you’re a SaaS business, the thing to ask yourself is: “if my customer could build my tool, line for line, would they still pay to use my software?” If the answer is no, then you need to start digging your moat, and finding a cornered resource or a network effect would be a good place to start.