There's something remarkable about the economics of artificial intelligence.
For approximately twenty dollars a month, someone can have access to technology capable of writing software, analyzing documents, researching difficult questions, preparing reports, investigating technical problems, and helping with activities that once required substantial amounts of specialized labor.
Not perfectly, of course. AI still makes mistakes, sometimes serious ones.
But even allowing for those limitations, the amount of useful work available through an ordinary subscription is extraordinary.
Which raises an interesting question.
How can companies afford to provide something this powerful for so little money?
It's a reasonable question. But I think there's a more important question hiding behind it.
Are we confusing what AI costs to provide with what AI is worth to the person using it?
Three Numbers That Aren't the Same
When people discuss AI pricing, they frequently mix together three different numbers.
The first is the provider's cost.
What does it actually cost an AI company to process requests, operate its infrastructure, maintain its systems, and deliver the service?
The second is the subscription price.
What does the company charge its customers for access to that service?
The third is the value to the customer.
How much time, effort, expertise, or other expense does that customer avoid by using AI? And what new things become possible because the technology is available?
Those three numbers are related, but they aren't interchangeable.
A service that costs twenty dollars a month might be worth very little to someone who rarely uses it.
The same service might be worth hundreds or thousands of dollars to someone who uses it every day for substantial professional work.
And neither situation tells us what it actually costs the provider to deliver that service.
That's where the discussion gets interesting.
Not Everyone Uses AI the Same Way
Consider how people use a typical AI subscription.
Some people ask a few questions each week.
They might request help writing an email, researching a purchase, planning a trip, or explaining something they don't understand.
Others use AI regularly throughout the workday.
A programmer might spend hours discussing architecture, reviewing code, tracking down bugs, generating documentation, and exploring implementation alternatives.
A researcher might analyze large collections of information. A business owner might use it to prepare documents, review proposals, or investigate operational problems.
And a particularly heavy user might have several AI conversations open simultaneously, working on different parts of one project or entirely unrelated activities.
All of these customers may be paying the same monthly subscription price.
But their usage patterns can be dramatically different.
This isn't unusual in subscription businesses.
Many services depend on customers using different amounts of what they purchase. Some customers consume relatively little. Others make extensive use of their subscriptions.
The important distinction is that AI isn't simply delivering a prerecorded movie or an existing document.
Each substantial AI request requires computation.
That means the cost of serving customers can change with the amount and type of work they ask the system to perform.
Which naturally raises another question.
Can a flat subscription price accommodate such different levels of activity?
The Subscription Is Only Part of the Business
One reason we need to be careful about drawing conclusions is that consumer subscriptions aren't the only way AI providers earn revenue.
There are business and enterprise offerings, developer services, and API access that may be billed according to actual usage.
With an API, for example, a customer might pay according to the amount of information processed and generated, with rates varying by model and service.
That is a fundamentally different pricing arrangement from a fixed monthly subscription.
The consumer subscription offers convenience and predictable pricing, subject to whatever usage limits the provider establishes.
The API offers a more direct relationship between usage and the customer's bill.
Business customers may also pay for capabilities, administration, security, integrations, or contractual terms that individual subscribers don't require.
These different arrangements give providers multiple ways to package and sell their technology.
But we shouldn't assume that revenue from one group necessarily subsidizes another.
Without access to a provider's internal financial and usage information, we simply don't know.
We can describe the business model. We can examine publicly available prices. We can identify the factors that might make the model sustainable.
What we can't responsibly do is declare that a company is making or losing a particular amount of money on an individual subscription.
The Cost of AI Is Not Standing Still
Another important factor is that AI technology continues to evolve.
The amount of computing power required for a particular task isn't necessarily fixed.
Providers can improve the efficiency of their systems. They can use different models for different kinds of work. They can take advantage of faster hardware, better software, caching, and other techniques that reduce the computation required to deliver useful results.
At the same time, more capable models and more demanding tasks can increase costs.
An AI helping someone polish a short paragraph isn't necessarily consuming the same resources as one analyzing a large software project or performing a complicated series of reasoning steps.
We also need to distinguish the cost of processing a request from the much larger expenses involved in developing and operating an AI business.
Research, training, infrastructure, personnel, security, and ongoing product development all cost money.
The public price of an AI service doesn't reveal how those expenses are allocated.
And while improvements in computing efficiency may help reduce costs, increased demand and more ambitious capabilities can work in the opposite direction.
So it would be a mistake to assume that today's subscription prices are guaranteed to remain unchanged.
What About the Heavy User?
Suppose someone pays approximately twenty dollars a month and uses AI extensively every day.
Is that customer costing the provider more than the subscription brings in?
Possibly.
But we can't determine that simply by looking at the number of hours someone spends using the application.
Having ten AI conversations open doesn't mean ten models are continuously consuming computing resources.
A conversation may sit idle while the user reviews a result, edits code, tests an application, or decides what to do next.
Other requests may require substantial processing.
The actual cost depends on what the system is doing, which capabilities are being used, and how efficiently the work is being performed.
Providers can also establish usage limits and adjust how their services are delivered.
There are too many unknowns to calculate the economics of an individual account from the outside.
But that doesn't mean we can't evaluate the economics from the customer's perspective.
In fact, that's where I think the more interesting story begins.
What Would the Alternative Cost?
Consider someone who uses AI to help develop software.
Traditionally, expanding what one developer could accomplish meant investing more personal time, purchasing specialized tools, or hiring additional expertise.
If a project needed help with database architecture, user interfaces, documentation, testing, or unfamiliar technology, those needs represented additional work.
Sometimes the developer had the necessary knowledge and could handle the tasks personally.
Sometimes the work required research, experimentation, or outside assistance.
And sometimes a project simply wasn't practical because the available time and money weren't sufficient.
AI changes that equation.
It can help one person investigate more alternatives, work across unfamiliar technologies, accelerate implementation, and perform tasks that otherwise might have been postponed or abandoned.
That doesn't mean an AI subscription is equivalent to hiring an experienced programmer.
It isn't.
Someone still has to direct the work, understand the requirements, evaluate the decisions, test the results, and accept responsibility for what gets delivered.
AI can create additional work when it makes mistakes, too.
But even after accounting for those responsibilities, the economic value can be substantial.
Imagine a developer who would happily pay two hundred dollars a month for the capabilities provided by a service currently costing about twenty dollars.
That doesn't establish what the service costs the provider.
It establishes something entirely different.
The customer believes the value received substantially exceeds the price paid.
And that's a legitimate economic observation.
Cheap Doesn't Mean Free
There's another trap worth avoiding.
When AI makes something inexpensive to produce, we can begin treating the result as though it has no cost at all.
Generate another implementation. Rewrite the documentation. Explore a different architecture. Build another prototype.
Why not? The subscription is already paid for.
But our time still has value.
Every generated implementation needs appropriate review. Every architectural alternative needs evaluation. Every promising result needs to be checked against the actual requirements.
If AI helps us produce ten possible solutions, that doesn't automatically mean we've saved time.
We may have created ten things that require consideration.
The economic benefit comes from using AI to accomplish worthwhile work more effectively, not simply from generating more output.
That distinction matters whether the subscription costs twenty dollars or two hundred.
A Better Way to Think About AI Pricing
I don't know what the long-term economics of consumer AI subscriptions will look like.
Perhaps prices will increase as more people begin using these systems for serious professional work.
Perhaps improvements in efficiency and competition will keep prices relatively low.
Perhaps we'll see more specialized plans that distinguish occasional use from demanding, sustained workloads.
Quite possibly, we'll see some combination of all three.
But I think the questions worth asking are fairly straightforward:
- What does the subscription actually allow me to accomplish?
- How much of my own time or outside expense does it save?
- What additional work does it make practical?
- How much effort do I still need to invest in directing and verifying the results?
- Would the service still represent good value if its price were significantly higher?
Those questions are more useful than simply asking whether AI is too cheap.
Because the answer depends on which side of the transaction we're examining.
The provider needs a business model that can sustain the service.
The customer needs a service that delivers more value than it costs.
Those aren't necessarily the same calculation.
And perhaps the most remarkable thing about AI today isn't that someone can buy access to it for twenty dollars a month.
It's what a knowledgeable person can accomplish with that access.
Related Field Note: Ten AI Sessions for $21 a Month