AI Is Getting Cheaper, Business Models Need To Catch Up
Tim Lee is the Founder and CEO of Bookipi, an international SaaS company trusted by millions of small businesses worldwide.
gettyA sharp correction in Chinese technology stocks recently caught my attention, but not because I think it predicts what will happen to AI stocks elsewhere.
Reuters reported that China’s STAR50 technology index fell 40% in less than four months, erasing about US$480 billion in market value. The interesting part for business leaders is not the market movement itself but the tension underneath it: huge amounts of capital invested in AI while the technology itself becomes increasingly competitive.
That tension is starting to change how I think about building with AI.
In my work deploying AI into software products and internal workflows, I have become less interested in which model is considered the best and more interested in the economics of completing a task reliably.
That sounds like a small distinction. I think it will become an important one.
Stanford’s “2026 AI Index Report“ shows how quickly the competitive landscape is tightening. The performance gap between leading models has narrowed considerably, including between leading U.S. and Chinese models. Stanford notes that this is shifting competition toward factors such as cost, reliability and domain-specific performance.
This reflects what I see when we build AI workflows.
For a business task, the question is rarely: Which model scored highest on a benchmark?
The questions are more practical. Did it complete the task correctly? How often did a person need to intervene? How long did it take? What did a successful completion cost?
For example, an expensive model may perform slightly better on a benchmark but be unnecessary for a simple classification or extraction task. A smaller model may complete that task reliably at a lower cost. The more complex model can then be reserved for jobs where its additional capability creates measurable value.
This makes model selection an operating decision rather than a technology preference.
One lesson I have learned from deploying AI is that increasing usage does not automatically mean the product is improving.
Every AI action can carry a variable cost. If usage doubles but revenue, retention or productivity does not improve with it, the business may simply be spending more money.
This is different from much of traditional SaaS, where serving another interaction can have a very small marginal cost.
The 2026 Stanford AI Index captures the scale of this challenge. AI company revenue is growing quickly, but compute costs and infrastructure investment are also reaching record levels. Global corporate AI investment more than doubled in 2025.
For operators, that makes unit economics important much earlier in the product development cycle.
When I look at an AI workflow now, I want to understand the cost per successful outcome rather than simply the number of prompts, conversations or agent actions.
If an agent processes 10,000 tasks, I want to know how many were completed correctly, how many required human assistance and what the total cost was. Then I want to compare that with the value created.
That is a more useful measure of adoption.
There is another side to these economics.
If the cost of intelligence continues to decline while capability improves, many AI use cases that do not make economic sense today may become viable.
This can benefit the companies applying AI rather than only those developing the underlying models.
A customer support workflow that is too expensive to automate today may work at half the inference cost. An agent that is economical for enterprise customers may eventually work for a small business. A process that currently requires one employee to supervise 10 AI tasks may improve until that employee can supervise hundreds.
This is where I see significant opportunity.
Businesses do not necessarily need to own the foundational technology. They need to understand where cheaper intelligence can improve a workflow they already know well.
Customer relationships, proprietary context, distribution and deep understanding of a workflow can remain valuable even as the models underneath a product change.
For leaders evaluating AI investments, I would focus on a few operating questions.
What business outcome is the AI responsible for? What does a successful outcome cost? How often does a human need to intervene? Does the customer keep using the capability after trying it? And do the economics improve as usage grows?
Those answers matter more to me than how many AI features have been launched.
The recent correction in China is a useful reminder of the difference between technological value and economic value. AI can become extraordinarily useful while individual companies struggle to capture enough of that value to justify their costs or valuations.
For businesses adopting AI, that may actually be good news.
More competition between models can mean better capability, lower prices and more choice. The opportunity for business leaders is to turn those improvements into workflows that customers repeatedly value at economics that make sense.
That is where I would expect durable AI businesses to be built.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?

