Here is a simple truth about the current AI boom: the people building the tools cannot reliably tell you what those tools will cost in two years. Not because they are being cagey, but because the underlying economics keep moving. That tension, between enormous upfront investment and a pricing model still searching for solid ground, sits at the heart of how the world’s biggest technology companies are trying to turn artificial intelligence into a sustainable business.
nn
A Bargain With a Bill Attached
nn
When you ask a free version of ChatGPT to draft a speech or map out an itinerary, you are using technology that cost its creators a staggering amount of money to build. Microsoft, Google, and Anthropic have collectively poured hundreds of billions of dollars into developing the large language models that power tools like ChatGPT, Gemini, and Claude. The free tier is, by any honest measure, a loss-leader on a colossal scale.
nn
So those firms do what any rational business would: they offer premium, paid-for tiers packed with extra features, from advanced coding assistance to more sophisticated data processing. The logic is straightforward. The execution, however, is anything but.
nh2>The Token Problem
nn
To understand why AI pricing is such a headache, you need to understand tokens. Tokens are the fundamental building blocks of large language models, the small chunks of text, sometimes a word, sometimes just a syllable, that an AI model reads and generates. Every question you ask, every answer you receive, every line of code a model writes, all of it is counted and processed in tokens. And the cost of processing those tokens has been dropping rapidly as the technology matures and competition intensifies.
nn
That speed of change is what makes long-term pricing commitments so awkward. As the BBC reports, Simon Gooch of Saviynt, an identity management company currently weaving agentic AI into its products, put it plainly: trying to lock a client into a cost model for the next one, two, or three years simply does not make sense right now, because the market itself does not know where it is heading.
nn
Agentic AI Adds Another Layer of Complexity
nn
The challenge gets thornier when you move beyond simple chatbots into the world of AI agents. These are more sophisticated systems, typically built on top of a large language model, but trained to carry out specific, multi-step tasks autonomously. Think of an agent that can log into your company’s systems, pull relevant data, generate a report, and email it to your team, all without a human clicking a single button in between.
nn
Third-party companies are building and selling these agent-based services at pace. But pricing them is genuinely complex. A basic chatbot interaction uses a relatively predictable number of tokens. An agent completing a multi-stage workflow can consume tokens at a rate that varies wildly depending on the task, the number of steps involved, and how many times the model needs to check its own reasoning before arriving at an answer. Charging clients a flat monthly fee feels arbitrary. Charging purely per token use makes budgeting almost impossible for the buyer.
nn
Why Long-Term Contracts Are a Gamble for Everyone
nn
Consider what a vendor faces when a corporate client asks for a three-year AI services contract. The vendor needs to estimate what token costs will look like in 2028. Given that the cost of running these models has been falling sharply year on year, a price that looks profitable today could become either wildly overpriced, frustrating customers, or dangerously thin-margined, hurting the vendor, within eighteen months. Neither outcome is good for a long-term business relationship.
nn
For buyers, the uncertainty is equally uncomfortable. Committing significant budget to a platform today carries the risk that a competitor launches a cheaper, more capable alternative six months later. The pace of development in this space has made even reasonably informed forecasting feel like guesswork.
nn
How Companies Are Coping
nn
Rather than pretending they have the answer, some vendors are leaning into shorter contract cycles, pilot projects, and consumption-based billing where clients pay for what they actually use rather than a fixed allocation. This approach gives buyers more flexibility and shields vendors from being locked into pricing promises that the market might quickly make obsolete.
nn
Others are bundling AI capabilities into broader software packages, making it harder to isolate and scrutinise the cost of the AI component specifically. It is a familiar tactic from the software industry’s playbook, and it works precisely because it obscures the per-unit economics behind a more digestible subscription price.
nn
The major platforms themselves, Microsoft with its Copilot suite, Google with its Workspace integrations, and Anthropic with its API offerings, are constantly adjusting their tier structures as their own infrastructure costs evolve. What a developer paid per million tokens in early 2024 looks very different from what that same volume costs today.
nn
The Road Ahead
nn
None of this means AI as a business is broken. It means the industry is still in the phase where the product is racing ahead of the commercial framework designed to support it. Historically, this is not unusual for genuinely disruptive technology. Cloud computing went through a similar period of pricing chaos before per-second billing and reserved instances brought some stability. Mobile data pricing took years to settle after carriers initially had no idea how to meter consumption without enraging customers.
nn
Artificial intelligence will likely find its equilibrium too. But for now, anyone trying to build a long-term budget around AI-powered services is navigating in real-time, making educated guesses in a market that rewards adaptability over commitment.
nn
The free tier might still be your best deal. The question is how long that particular bargain lasts, and what happens to the businesses built on top of it when the bill finally comes due. What do you think: should AI companies commit to fixed, transparent pricing, or is flexible consumption-based billing actually better for the businesses that rely on these tools?
