I’ve been thinking about a Wall Street Journal story this week reporting that OpenAI is considering lowering the price of its AI services as it prepares for increased competition from Anthropic.
The article itself is about pricing.
What I can’t stop thinking about is business models.
For the better part of three years, the conversation around artificial intelligence has been dominated by a single question: Who has the best model? Every benchmark, every product announcement and every leaderboard has been treated as another scorecard in a race that supposedly determines who will own the future.
The Wall Street Journal story suggests we may be entering a different phase.
If the companies building the most advanced intelligence in the world are already preparing to compete on price, perhaps intelligence itself is beginning the same journey every foundational technology eventually takes. It starts scarce and expensive, becomes increasingly capable, attracts competition and, over time, becomes broadly available.
That is not necessarily bad news for the companies building it. It may simply mean the economics move somewhere else.
Electricity followed that path. Railroads did. Telecommunications did. Computing power and cloud infrastructure certainly did. None of those technologies became less important as they became less expensive. Quite the opposite. Their economic impact exploded because lower costs allowed millions of people and businesses to use them in ways that had previously been impossible.
The cloud is a good example.
The early discussion around cloud computing focused on the infrastructure providers. Today, some of the most valuable software companies in the world exist because developers stopped worrying about buying servers and started building applications.
Cheap infrastructure created expensive businesses.
Artificial intelligence may be approaching a similar transition.
If inference costs continue to fall, the greatest opportunity may not belong to the companies producing intelligence but to the companies figuring out what to do with abundant intelligence.
That future is easy to imagine.
Every enterprise will need governance for autonomous agents. Every organization will need persistent memory that allows AI systems to accumulate institutional knowledge rather than start fresh with every interaction. Industries will demand specialized intelligence tuned for healthcare, legal services, manufacturing, financial services and semiconductor design. Someone will have to orchestrate millions of interacting agents, monitor their behavior, secure them and integrate them into existing workflows.
Those are attractive businesses.
History suggests they are often more attractive than the infrastructure beneath them.
But there is another side to this discussion.
The assumption that intelligence naturally becomes infrastructure also assumes the companies creating it are willing to accept infrastructure economics.
That is a very different proposition.
Utilities are outstanding businesses. They generate dependable cash flow, invest for the long term and earn healthy returns while supporting modern civilization. Investors do not generally assign trillion-dollar valuations to utilities because the expectation is stability rather than extraordinary growth.
The frontier AI companies being built today are attracting a very different kind of capital.
If OpenAI, Anthropic and others eventually enter the public markets with valuations measured in the hundreds of billions or even trillions of dollars, shareholders will expect them to continue expanding rather than settle into the role of digital utilities.
The logical response is to move up the stack.
👉The model becomes only the engine.
👉The enterprise platform becomes the product.
👉The persistent memory layer becomes the moat.
👉The agent ecosystem becomes the marketplace.
👉The workflow becomes the customer relationship.
Technology history is full of companies that started by providing infrastructure and gradually expanded into the layers above it because that is where differentiation and margins reside. Search engines became advertising platforms. Operating systems became productivity suites. Retail companies became cloud providers. Every successful platform eventually looks for opportunities to own more of the customer’s experience.
Artificial intelligence may follow the same pattern.
That creates an interesting tension.
The broader economy benefits if intelligence becomes inexpensive. Lower inference costs, cheaper energy and more efficient AI factories reduce barriers to entry and encourage experimentation. The next transformative AI company may never need to build a frontier model because intelligence has become as accessible as cloud infrastructure is today.
The companies creating that infrastructure, however, may have every incentive to ensure they capture more of the value than a traditional utility ever could.
Those two objectives are not necessarily incompatible, but they are different.
There is also a third possibility that receives far less attention.
As artificial intelligence becomes embedded in healthcare, finance, education, transportation and government, policymakers may eventually conclude that access to frontier intelligence is too important to be governed entirely by market dynamics.
History suggests societies often treat foundational infrastructure differently once it becomes essential to economic growth. Electricity, water, telecommunications and broadband all evolved under regulatory frameworks that attempted to balance private investment with broad public access. Operators earned attractive returns while recognizing that the infrastructure itself served a larger societal purpose.
It is not impossible to imagine a similar conversation around artificial intelligence.
That future would still produce profitable companies. It might even provide the stability necessary to support long-term investment in increasingly expensive frontier models.
Whether it would satisfy investors expecting unlimited upside is another matter.
Perhaps that is the real tension hidden inside today’s discussion about token pricing.
One future sees intelligence becoming an abundant infrastructure that enables thousands of new industries. Another sees today’s frontier model companies successfully moving up the stack and capturing much of that value themselves. A third imagines governments viewing intelligence as critical infrastructure deserving both protection and oversight.
The reality may include elements of all three.
The first generation of AI companies is unlikely to voluntarily accept utility economics if platform economics remain available. At the same time, competition has a way of pushing infrastructure toward commoditization regardless of the intentions of those who built it. Society, meanwhile, tends to favor broad access to technologies that become essential to economic life.
Those forces will shape the next decade of AI as much as benchmark scores or product launches.
The irony is that the long-term success of artificial intelligence almost certainly depends on intelligence becoming cheaper rather than more expensive. Every reduction in the cost of inference expands the number of people and organizations able to build with it. Every improvement in energy efficiency makes the economics more compelling. Every advance in AI infrastructure creates room for another entrepreneur to imagine something no one has thought of yet.
The Wall Street Journal story may ultimately prove to be less about OpenAI and Anthropic than about the natural evolution of every transformative technology.
The companies creating intelligence are understandably focused on protecting their margins.
History suggests the larger opportunity emerges when intelligence becomes abundant enough that everyone else stops worrying about buying it and starts worrying about what they can build with it.
And if that happens, the greatest fortunes of the AI era may belong not to the companies that manufactured intelligence, but to the ones that recognized abundant intelligence as the raw material for an entirely new economy.
