Why AI Companies Can't Figure Out How to Price Anything Anymore

By Imran Khan (AI Tech Safar)

For twenty years, software pricing was almost boring in its predictability: count your users, charge per seat, send the invoice. AI has quietly broken that entire model, and most companies still haven't figured out what replaces it. The number of software companies using consumption-based pricing has more than doubled since 2015, according to McKinsey — and the pace of that shift has sharply accelerated since generative AI entered the picture, because AI simply doesn't behave like traditional software.

The core problem is deceptively simple to state and genuinely hard to solve: a single AI query can cost a company anywhere from a fraction of a cent to several dollars, depending on which model handles it, how much reasoning it does, and how long the task runs. Charging a flat per-seat fee on top of that kind of cost volatility is like a restaurant charging every customer the same price whether they order a side salad or the entire tasting menu.

Quick Summary & Key Takeaways

• Per-seat pricing is breaking down: AI's variable, unpredictable compute costs don't fit the flat-fee-per-user model that's dominated software for two decades.
• Consumption-based pricing has doubled since 2015 and is accelerating fast as more vendors shift toward it, per McKinsey research.
• Credit-based billing is becoming the new default — customers prepay for usage credits, and vendors apply margin on top as those credits get consumed.
• Chatbots and AI agents are heading in opposite directions: generic chatbot pricing is racing toward zero due to commoditization, while pricing power is actually rising for specialized, capable AI agents.
• Enterprise software spending is still climbing — Gartner projects 15.2% growth in 2026 — but a big chunk of that growth is just companies absorbing price increases, not buying more value.
• Buyers are confused too: without clear, comparable pricing models across vendors, companies are struggling to even evaluate which AI tools are worth paying for.

How AI Pricing Models Compare

Pricing Model	How It Works	The Problem With It
Per-seat (traditional SaaS)	Flat fee per user, regardless of how much they actually use the AI features	Doesn't reflect AI's wildly variable compute costs — heavy users cost far more than light ones
Consumption-based	Customers pay based on actual usage — tokens, queries, or compute time consumed	Costs become unpredictable for buyers trying to budget month to month
Credit-based	Customers prepay for a pool of credits, spent down as features are used	Vendors must guess the right credit-to-cost ratio, which shifts as model prices change
Outcome-based	Pricing tied to a measurable result — a resolved ticket, a fraud attempt caught	Hard to define and verify a clean "outcome" for many AI use cases

Why the Old Model Doesn't Work Anymore

Traditional SaaS pricing worked because the underlying cost of serving one more user was almost nothing — software doesn't get more expensive to run just because someone logs in more often. AI flips that assumption on its head. Every single query burns real compute, and the cost of that compute swings enormously depending on which model answers it and how much reasoning it has to do. A quick factual lookup might cost a fraction of a cent; a complex multi-step reasoning task on a frontier model can cost dollars. Charging every customer the same flat rate, regardless of which type of usage they generate, means vendors either overcharge light users or quietly lose money on heavy ones.

This is exactly why consumption-based and credit-based pricing have taken over so quickly. They let a vendor's revenue scale in step with the actual compute cost being incurred — but they introduce a new problem: customers who are used to predictable monthly software bills suddenly face invoices that swing depending on how much they used the tool that month, which makes budgeting and procurement genuinely harder on the buyer's side.

The Split Between Chatbots and Agents

Not all AI pricing is moving in the same direction, and that split is one of the more interesting parts of this shift. Basic AI chatbot features have become commoditized fast — nearly every software product now has some kind of built-in assistant, so pricing power for generic chat features is collapsing toward zero as buyers refuse to pay extra for something they can get for free elsewhere.

Specialized AI agents are moving the opposite way. Tools that can autonomously complete complex, multi-step, end-to-end tasks — rather than just answering questions — are commanding rising prices, because they're replacing more expensive human labor or entire workflows rather than just adding a convenience feature. The pricing gap between "AI that chats" and "AI that actually does the work" is becoming one of the clearest dividing lines in enterprise software.

What This Means for Buyers

For companies purchasing AI tools, this pricing chaos creates a genuinely difficult evaluation problem. When one vendor charges per seat, another charges per token, and a third sells prepaid credits, there's no simple apples-to-apples way to compare what you're actually getting for your money. Gartner's own 2026 projections show enterprise software spending climbing 15.2%, but a significant chunk of that growth is companies simply absorbing price increases on tools they already use — not necessarily buying more capability. Some analysts estimate roughly 9% of that growth is effectively an inflation tax on existing software contracts.

💡 AI Tech Safar Insight
The pricing confusion in AI right now isn't a temporary growing pain — it's a sign the entire software industry is being rebuilt around a fundamentally different cost structure than the one it was designed for over the last two decades. Per-seat pricing assumed software was cheap to run and expensive to build. AI often reverses that: it can be relatively cheap to build a product around an existing model, but genuinely expensive to run at scale. Until pricing models catch up with that reality, expect more confusing bills, more experimentation from vendors, and more buyers quietly overpaying simply because there's no clean way yet to compare one AI tool's price to another's.

Frequently Asked Questions (FAQs)

Q1: Why is AI pricing so much harder than traditional software pricing?
Because AI's compute costs vary enormously per use, unlike traditional software where serving one more user costs almost nothing. That variability breaks the flat per-seat pricing model most software companies have relied on for decades.

Q2: What is consumption-based pricing?
A model where customers pay based on actual usage — such as tokens processed or queries made — rather than a flat fee per user, letting vendor revenue track real costs more closely.

Q3: Why are AI chatbot features getting cheaper while AI agents get more expensive?
Basic chatbot functionality has become widely available and commoditized, driving its price toward zero. Specialized AI agents that autonomously complete complex tasks offer more differentiated value, so they retain — and are gaining — pricing power.

Q4: Is enterprise software actually getting more expensive in 2026?
Yes, partly. Gartner projects 15.2% growth in enterprise software spending in 2026, but a notable share of that is companies absorbing price increases on existing tools rather than purchasing genuinely new capability.

What Do You Think?
Would you rather pay a predictable flat fee for AI tools, even if it means overpaying sometimes — or a usage-based price that's fair but harder to budget for? Drop your take in the comments below!

Related Reading:

• China, Military AI, and the OpenAI-Anthropic Distillation Debate
• Congress Just Introduced a Bill to Give the US Government an AI 'Kill Switch'
• OpenAI's Rogue AI Breached 4 Services, Left Notes for Itself — And Anthropic Faces Backlash

Source: Reporting based on McKinsey, Gartner via SaaStr, and BBC News.