AI investments have been so hot over the last few years that they have eclipsed almost everything else. But AI companies have struggled to monetize their investments in model development, and investors are starting to get nervous about whether this latest and coolest technology can actually be the basis for a profitable business.

But perhaps there is good news, and it comes from the world of COSS. 

What AI Businesses Can Learn from COSS

Lots of people seem to be in denial about the commercial value of open source. Academics scratch their heads about why people contribute to open source, because they see it as economically irrational, citing the tragedy of the commons. Investors say it’s crazy to think you can make money giving software away. But they are missing what is right in front of them. Commercial open source software (COSS) has an excellent track record for making money out of developing open source software. 

Meanwhile, AI–meaning the development of machine learning models–is quickly becoming commoditized. The market did not expect this, and investors have been dismayed by it. They are starting to ask whether it is truly possible to monetize AI. 

What a lot of analysis about the commoditization of AI misses is that the technology is not all that proprietary. The transformer technology on which all the LLMs are based is actually a bit old, and is free for anyone to use. What changed a few years ago is that a handful of wealthy companies were willing to throw billions in resources at executing on the model development technology. In other words, the only market advantage was a boatload of capital. This caused a very quick commoditization — anyone with enough money could develop LLMs. And while few companies had enough capital–much less the data to train it on–they all started releasing competing LLMs. It only takes a few competitors to create price competition, and so the price of using a basic model quickly fell to zero, and the price of a business account was modest for all but the highest volume users. So, investors started asking whether it was possible to build a business around AI if it was essentially a free product. 

But we have been through this conundrum before. COSS companies figured this out a long time ago. When technology is commoditized, companies make money by packaging it into useful products and delivering it at scale. That this is possible is something that traditional analysis often misses. In COSS, the thing being given away–source code–is not really a product. Similarly, ML models are not really products. They need a bunch of software, services and other artifacts to make them usable. So, in the long run, companies who are good at packaging up AI may be the winners in this sector, rather than the model developers. Companies learning to package models for use can look to COSS for lessons on how to structure their businesses.

This leads to a corollary. Intellectual property specialists have sounded an alarm that licenses for AI might not be enforceable. At this time, models that are developed without human intervention don’t enjoy any copyright or patent protection in the US, and in many other nations. But at the end of the day, that might not matter. Open source is the ultimate commoditization, and open source licenses are rarely enforced in formal legal proceedings. COSS companies win in the marketplace, not the courtroom.  If you can make money deploying open source, you can make money deploying AI.

What AI Businesses Cannot Learn from COSS

Although there are useful parallels between COSS businesses and AI-based businesses, there are also a few stark differences.

First, AI models can be extremely expensive to develop, but COSS is relatively inexpensive to develop–each in comparison to conventional proprietary enterprise software. So, the teachings of COSS can apply to AI add-ons like agents, UI, and AI-driven apps, but don’t help much for training large language models at scale. Those “add-on” businesses, however, also run the risk of being a “feature not a product” that is beholden to the business policies and strategy of its LLM vendor. In contrast, most COSS products run on a large stack of software, but most of the stack is open source, which frees the COSS business from vendor lock-in.

Also, the pace of AI-related businesses is beyond fast. COSS companies generally have a one-to-three year ramp-up period, because they usually build community before going to market with products. So, some of the branding benefits of COSS might not translate well to the AI space. The community generation just does not have time to percolate. New AI tools are coming along every day, and AI companies risk getting lost in the shuffle.

AI-related businesses have recently enjoyed a huge capital infusion uplift–so much so, that it seems all the companies completing investment rounds in the last few years have been AI businesses. COSS has been harder to finance during that time. COSS tends to play second fiddle to the flavor of the month: first crypto, then AI. That can be frustrating for those of us investing in COSS. But this is a double-edged sword. Businesses that start out with large infusions of venture capital are under extreme pressure to perform, and getting a business to revenue generation–much less positive cash flow–is always difficult. COSS businesses tend to have more leeway, because they are expected to have a somewhat longer go-to-market timeline.

Which is better: COSS or AI?

That’s a false dichotomy. Going forward from 2026, enterprise software will almost always have both COSS and AI elements. Entrepreneurs and investors need to understand how the software business of the future can learn from COSS–and how they will have to break new ground.