The market view for AI is that one of the big winners will be model makers. After all these are the people who invented it. The markets – at least the private markets – have assigned huge valuations to the companies building the models, mostly OpenAI and Anthropic.
The history of great inventions suggests that investors should be careful with this conclusion.
Some of the most transformative inventions ever created have generated enormous benefits for society, yet their inventors’ wealth was surprisingly small. World Wide Web is released royalty-free. TCP/IP has become an open standard. GPS is now available for free. Linux has become one of the most important programs in history while remaining open source.
Even patented technologies have often failed to produce wealth commensurate with their impact. Bell Labs invented the transistor in 1947, but AT&T licensed the technology extensively under regulatory pressure. This decision helped create the modern semiconductor industry, giving rise to companies such as Fairchild Semiconductor Corporation and eventually Intel. The economic value a transistor creates may be in the hundreds of trillions of dollars. Only a small portion of the accumulated to its inventors.
The lesson here is simple: inventing general-purpose technology does not mean capturing the value it creates.
This brings us to large language models.
Today, investors value AI companies as if having the best model would lead to perpetual monopoly profits. This may happen for a while, but it’s worth considering another possibility: There are no patents in AI, secrets are hard to keep, and distillation seems to produce models that are almost as good.
If frontier models continue to become commoditized through open source development, rapid distillation technologies, and relentless competition, paying a premium for the best model ever may become increasingly difficult. Models that lag behind the leaders by a month or two can be “good enough” for most commercial applications.
Moreover, there is a push for companies to control their own data, which may push them toward open source and away from big models.
In that environment, pricing power becomes elusive.
It is often said that the real money will be made by application companies built on foundation models. This sounds reasonable until you consider what the AI itself does. One of its primary capabilities is to reduce the cost of creating software. If building applications becomes significantly easier and cheaper, competition will increase and it will become more difficult to maintain excess profits there as well.
The same logic that makes the model layer a commodity can eventually turn much of the application layer into a commodity.
This raises a more interesting question for investors.
Technology companies probably won’t be the biggest beneficiaries of AI after all.
Instead, the winners may be companies with deep competitive moats operating in the physical economy. Manufacturers, industrial companies, logistics operators, utilities, mining companies and other capital-intensive businesses often earn modest margins despite high barriers to entry. A company that operates on a 2% margin and consistently expands to 3% through AI-powered productivity was able to increase its profits by 50% without selling a single additional product.
It is difficult for competitors to erase these gains when competitive advantages come from size, infrastructure, organization, geography, or capital intensity rather than software.
This would represent a very different investment story from the one that dominates the markets today.
Investors may be fighting the last war and paying too much attention to the information economy over the past 30 years.
The Internet has produced extraordinary returns for software platforms because distribution itself has become the moat. Artificial intelligence may be different. As intelligence becomes plentiful and inexpensive, it begins to resemble electricity more than enterprise software.
Electricity did not create the greatest fortunes for the scientists who discovered its principles. Michael Faraday transformed physics without becoming fabulously wealthy. Even Thomas Edison and George Westinghouse, who successfully commercialized electrical systems, captured only a small portion of the economic value that electricity ultimately created.
Most of the benefits flowed to the broader economy through increased productivity, lower costs, and entirely new industries.
Artificial intelligence could follow the same path.
This does not mean that artificial intelligence will not be one of the most important technological developments in history. I think it will become just that.
But the most important question for investors is not how valuable AI will become, but rather who will actually hold the value.
