The AI Free Ride is Over: Why Your Product’s Unit Economics Are Broken
2026 may be the year when everyone starts asking AI: “Where is my money?”
The last few years were a crazy, neck-breaking run for AI. Everyone wanted to create AI products or add AI features to existing ones. Often without looking closely at costs, monetisation, or whether the feature made actual business sense.
Investors are starting to ask about revenue, margins, and returns. This year, verification will start.
Why AI monetisation is harder than it looks
Monetising a digital product is not just putting a “price sticker” on it and waiting for things to happen. Even in traditional software, monetisation has always been a complex task: navigating limitations and finding common ground between product value, finance, technology, and strategic goals.
The specificity of AI, especially LLM-based solutions, adds its own categories of problems — or changes existing ones.
AI changes unit economics
Unit economics behave differently for LLM-based AI-native features than for traditional software features.
In many traditional software products, marginal costs scale sublinearly and may become almost negligible. The more users or customers you have, the better your balance sheet should look. At some point, a new customer adds little or almost nothing to the cost base, because the unit cost is close to zero.
AI costs are different.
The cost of LLM-based functionalities scales linearly or even super-linearly: unit cost stays at the same non-zero level, or may even grow. The more users you have, and the more they use the feature, the bigger the bill. The potential for super-linear scaling is even more dangerous. When customers increase their usage of AI features, the number of tokens used may grow even exponentially. This results in weaker gross margin: 52% for AI products vs 75-85% for traditional software; while Gross margins for AI products are improving, they are still far from traditional software.
It’s hard to monetise Nice to haves
Unit economics is one thing. Willingness to pay is even more important.
The value generated by an AI feature matters more than the AI sticker. At some point, having “AI” in the product name may even become a burden, not an advantage. Customers need to be convinced to pay for AI features, not just try them because they are available.
Right now, many customers may be using AI features because they are convenient and free. The moment you start charging enough to balance the costs, they may refuse to pay as much as you need. Even worse, they may blame you for making the product worse, because something that used to be included is now paid, limited, or metered.
The risk that AI features become costly “nice-to-haves” is real.
Without actual, measurable, visible value for the customer, AI features may become just another add-on — but a costly one. If customers do not see a clear connection between AI functionality and meaningful value, they will not pay for it.
Btw, part of the problem is where AI features and products are focused.
It seems that a lot of AI value today is framed as cost cutting: reducing time spent on a ticket, reducing manual work, reducing the need for workforce, or handling mundane, time-intensive tasks. There is value in that. But it has its own limitation: you cannot reduce costs by more than 100%. The bigger AI come from increasing existing value, not only reducing existing costs. AI should be a value multiplier, not only a cost-cutting tool.
Specific examples worth mentioning here are various “shopping assistants” and similar helpful advisors. Building a free of charge, add-on like shopping assistant, based on third party LLM providers, is a huge risk for balance. The feature won't have a significant impact on revenue, because it is not directly related to conversion, order value etc, but will have meaningful costs. Moreover, poorly implemented assistants may even hurt revenue by driving customers out or prolonging their decision process.
For this discussion, the important point is this: AI functionality designed only to cut costs or just to have “AI sticker” has limited monetisation potential. It may be valuable, but its ceiling is naturally constrained.
Weak moats, weak pricing power
The market is packed and tough. AI startups rise and fall in hundreds. It is difficult to build a moat that protects an AI product from competitors taking over users or building a very similar feature.
Even frontier models from top AI labs are, in many use cases, becoming increasingly interchangeable. One model may be slightly better at one type of task, another model at something else. But for many product use cases, they can do similar things well enough.
This makes differentiation harder.
The market is also being squeezed by big players. Simple LLM wrappers have a hard time, even if they become recognisable, because OpenAI, Anthropic, Google, Microsoft, Meta, or other large players can build similar functionality directly into their own products or platforms.
This does not make AI monetisation easier.
Providers need to become profitable
When we talk about OpenAI, Anthropic, and other big players, they are part of the problem as well.
Model providers are under pressure to become profitable, or at least to show a credible path toward profitability. Growing the number of users will not fully solve that problem on its own. They also need to work on ARPU: increase usage, improve packaging, introduce higher-tier plans, or increase prices.
While LLM inference costs are falling, they will probably continue to fall. But this does not automatically mean that the total cost of running AI-based products will fall in the same way. Usage will grow, prices for top models won’t become cheaper and there is a huge chance that you will need to use top models for their better quality.
So costs of running AI-based features may be affected by global providers' economies and their need to effectively increase average revenue per customer. The biggest proof of that risk, last revelations about OpenAI financials: $21 billion losses, costs growing almost at the same rate as revenue, and huge growth of “Sales & Marketing” costs (418%) indicate that current AI prices are highly subsidised by providers.
Sure, there is a different option – run all on your own model. There exists open source models, can be fully run on your own infrastructure. But… it doesn’t come for free. R&D, model training or tuning and infrastructure for inference still are there and costs. Also, it isn't a one-time cost. Training, tuning have to be repeated constantly, inference happens with each prompt/api call.
Surfaced OpenAI financials, double down on how economy is crucial for AI-based products. Huge growth of revenue but also huge jump in costs, gigantic spending on subsidised accounts are not only the problem for OpenAI, but for all building AI products.
What this means for product managers
Ok, what this means for you, as a product manager
Stop forcing AI everywhere
Slapping an AI sticker on a product is no longer an actual option. In fact, it never was, but that is a different story.
There is no value in AI in the name alone. Not enough to monetise. Not enough to defend margins. Not enough to justify costs. AI should be added where it improves the product in a way customers can understand, measure, and pay for.
Know your product economics
Product economics is in the front seat now. Until recently, the main cost of adding a feature to was building cost. After that running costs were so low that pricing was a business decision. Now, running an AI product feature costs and you may end up with a feature that doesn’t do anything meaningful for improving product revenue yet it worsens margins.
Focus on value, not AI itself
Think and work on actual, meaningful value, solving real, crucial problems for customers. Any product feature should provide some value for the user or customer. At least in theory. But even if there is tangible value, customers will not always be willing to pay for it at all, or at the level needed to exceed costs and make the feature economically valid.
Speak financial language
Researching and understanding product financial constraints is not an option with AI features. You have to ask how the product impacts margins, how to validate pricing and how to make financial predictions. Understanding simple concepts of LTV and CAC is no longer enough. PMs should work closely with Finance, Operation and tech, be able to bring customers financial insights and predictions to the table and inform financial modelling.
Look at the full picture and don’t be afraid of changes
AI products and features may, and in many cases will, require rethinking product monetisation. Sometimes they may require completely new concepts, such as outcome-based billing.
There are many options: tokens, credits, usage limits, seat-based pricing, single payments, outcome-based pricing, access-based pricing, integration-based pricing, or hybrid models.
There is no one-size-fits-all answer. But a few rules of thumb may help navigate the AI monetisation challenge:
- Focus on where the value for the customer really is and how the feature helps them achieve success. For AI features, value-based pricing should usually be the starting point.
- Understand how usage is structured and how it translates into cost. A vertical AI workflow will behave differently than a horizontal AI assistant. A support automation product will behave differently than a writing copilot.
- Adopt the right metering. This depends on usage patterns, cost drivers, customer expectations, and billing clarity.
- Make it clear to the customer what they are paying for and why. Billing clarity matters. If customers do not understand the pricing logic, they will see it as punishment, not value capture.
Summary
AI as a technology will stay with us in some form. But generating revenue from AI will not be a simple task.
Companies will have to create real monetisation models for AI features, not just add them to existing products and hope the economics work later.
Sources
- ICONIQ, January 2026, The Execution Era of AI. State of AI: Bi-Annual Snapshot, https://www.iconiq.com/growth/reports/2026-state-of-ai-bi-annual-snapshot
- Ed Zitron, June 2026, OpenAI Losses Increased Nearly 8X in 2025, With Spending Hitting $34 Billion, https://www.wheresyoured.at/exclusive-openai-financials/