AI Is About to Become Too Expensive for Small Businesses

AI used to feel like a flat-fee superpower. Now it is starting to behave like electricity: metered, variable, and tied to exactly how you use it. We dig into what that shift means for the people actually trying to run AI in the real world, from small businesses building their first workflows to sales, ops, and dev teams pushing tools like Microsoft Copilot, ChatGPT, Claude, and agents into daily work.

Josh unpacks the move from per-seat pricing to usage-based billing, where cost is driven by tokens, context retrieval, tool calls, and how long a model “thinks” behind the scenes. That change forces a new set of business questions: which workflows burn the most tokens, when premium reasoning is worth it, and how to stop paying for AI habits that feel productive but do not produce ROI. We also talk about the uncomfortable incentives that come with token-based revenue models and why customers need clearer visibility into what they are being billed for.

Then we zoom out to the market dynamics: if the best models become scarce at scale, enterprises with huge contracts may secure better pricing and reserved capacity while smaller teams get caps or slower access. Finally, we make it practical with an AI FinOps mindset: map workflows, set internal model tiers, put guardrails around agents, train teams to prompt efficiently, and tie spend to measurable outcomes.

If you are trying to budget for AI, prove business value, or keep your AI bill from quietly exploding, this one will give you a clear framework. Subscribe, share this with a teammate who owns the budget, and leave a review with your biggest question about AI costs.

Josh's LinkedIn

Transcript

Why AI Suddenly Feels Expensive

SPEAKER_00 0:00

I want to talk about the cost of AI from the perspective of the people who are actually trying to use it. Small businesses, consultants, sales teams, operations teams, developers, and companies that are finally starting to build real workflows around these tools. Because something is really changing. I don't know if you've felt that shift, but for the last couple of years, AI almost felt magically cheap. You could pay 20 bucks a month and get a chatbot and get access to a model that only a few years ago would have been impossible to access. You could use Copilot, ChatGPT, Claude, Perplexity, Cursor, and other tools like they were utilities. You did not think about the cost of every prompt. You did not think about how many tokens were being burned. You did not think about whether the model was doing one second of reasoning or 30 seconds of reasoning behind the scenes. You're just used to it. And that's the point. The first era of AI was about adoption. The companies needed people to try it. They needed businesses to get comfortable with it, and they needed employees to build habits around it. They needed developers to build products on top of it. They needed executives to start believing that AI was not just a gimmick, but a new layer of work. So the price felt simple. Pay per seat, pay per month, get

From Subscriptions To Utility Billing

SPEAKER_00 1:22

access. But now we're moving into a different phase. The free trial era is pretty much ending, and the cheap experimentation era is also ending. The don't worry about the usage era is ending. Microsoft's co-pilot co-work is a perfect example. During the Frontier preview, companies could test it. They can experiment with it. They could figure out what was possible. But once you moved it to general availability, the economics began to change. Now it requires a 365 co-pilot license, and co-work itself is billed based on usage. Microsoft breaks down the cost around things like the model being used, the context being retrieved, the tools being called, and the runtime of each task. Because that matters. This is not just one prompt equals one price anymore. A light task is not the same as a heavy task. A simple answer is not the same as an answer grounded in company files. A basic AI tool is not the same as a tool that uses premium reasoning. A short response is not the same as a long response. A simple workflow is not the same as an agent that reads context or calls tools, reasons through a problem, and produces multiple outputs. That is the shift.

Cost Becomes Behavioral And Operational

SPEAKER_00 2:39

AI is moving away from software pricing to utility pricing. And that changes everything. When software's priced per seat, the buyer mostly asks how many people need to access it. When AI is priced by usage, the buyer has to ask a different question. What are these people actually doing with it? That sounds like a small change, but it's a massive business shift. If such a thing could happen in such a short period of time, because now the cost of AI is not just tied to the license, it's tied to behavior. It is tied to how people prompt. It is tied to how much context they included, whether they use a basic model or a premium model. It's tied to whether the answer is short or long. It's tied to whether the system is doing deep reasoning. It's tied to whether the agent is looking across your tenant, your SharePoint files, your emails, your meetings, your CRM, and your internal knowledge base. In other words, AI cost is becoming operational. And most companies are not ready for that. They're used to managing software subscriptions. They're not used to managing model consumption. They know how to ask, how many licenses do we have? They do not yet know how to ask which workflows are burning the most tokens. They know how to ask about who has access, but they don't yet know how to ask, is the tool being used efficiently, or even better yet, how efficiently is the tool being used? That is where the real conversation begins.

Incentives To Sell More Tokens

SPEAKER_00 4:04

Now, let me put my tinfoil hat on for a second. There are two theories that I keep coming back to. The first theory is that the frontier AI companies have an incentive to push more token usage than customers actually need. I'm not saying that they're doing this secretly or they're doing something illegal, but I am saying that they have an incentive and it's obvious. If the business model is based on usage, then more usage means more revenue. Longer answers mean more tokens. More reasoning, more tokens. More context, more tokens. More calls, more tokens. It's all billable activity. And because a lot of this happens behind the scenes, the customer often doesn't fully understand what they're paying for. That is where things get a little bit uncomfortable. And I'm starting to have these uncomfortable conversations. If AI is going to become the core business utility, businesses need consistency. That brings me to my second tinfoil hat

Will Big Companies Get Priority

SPEAKER_00 5:04

theory. And so I'll put my tinfoil hat back on and let's see what happens when the largest companies start using the best models at scale. Right now, small businesses can access amazing tools. That is one of the most exciting things about AI. A small company can punch above its weight and use technology that a Fortune 500 company uses. A local business owner can generate proposals or analyze documents, summarize meetings, automate follow-ups, create marketing content, and build internal workflows without needing a massive engineering team. That feels democratizing, and that's why we all love AI. How long will it stay that way? The biggest companies have the money, they have the data, they have the infrastructure and the contracts. They can sign huge enterprise agreements, they can reserve capacity, they can pay for the premium models, they can negotiate custom terms, they can embed AI deeply into every department, and that scales. If demand for the best models grow faster than supply, then who gets priority? Small businesses paying a few hundred dollars a month? Or the enterprise paying millions. This usually works. Scarce resources go to whoever can pay the most. The best pricing goes to the biggest contracts. The best capacity goes to strategic accounts. So even if AI started as a democratizing sort of force, there's a real chance that the next phase becomes more unequal. Small businesses may still get access, but not necessarily to the best AI. Not the fastest AI, and not the most deeply integrated AI. They may get the basic tier while large enterprises get the premium tier. They may get usage caps while large companies get reserved capacity. They may get generic models while large companies get customized systems trained around their workflows. That is not just science fiction. It is just enterprise software history repeating itself. The internet started open and then platforms formed. Cloud computing started flexible. Then hyperscalers dominated. AI started cheap and accessible. Now it is becoming metered, tiered, bundled, and governed by the companies that own the infrastructure. That does not mean that AI is bad. I make my living with AI. Love AI. But it does mean that the economics are

The Real Costs Behind AI Infrastructure

SPEAKER_00 7:27

maturing. And when economics mature, somebody has to pay the real cost because there's no such thing as a free lunch. Microsoft, Amazon, Google, Meta, OpenAI, Anthropic, and others are not building this future with pocket change. They are spending massive amounts of money on data centers, GPUs, CPUs, energy research, talent, and infrastructure. These companies are betting that AI will become the core layer of their business and consumer life. And here's the thing: adoption is real, but monetization is still uneven. A lot of people are trying AI. A lot of employees are experimenting with it. A lot of companies are piloting it, but turning that experimentation into measurable business value is much harder. And that's what I see with clients all the time. People are impressed by AI. They see the potential, but potential is not the same as ROI. A chatbot demo is easy, a real workflow is hard. Governed process that saves time every week, avoids security risk, fits the company's data policies, and produces a measurable business outcome is really hard. That gap is where a lot of the money is being lost.

ROI Gaps And The Rise Of AI FinOps

SPEAKER_00 8:40

The future of AI cost control is going to look more like financial operations. Companies will need AI usage policies. They will need internal model tiers. They'll need to decide which tasks deserve premium reasoning and which tasks should use cheaper models. They will need to monitor usage by department, user, workflow, and business outcome. If an AI agent spends $2 to complete a task that would have taken a manager two hours, then that's a win. But if an employee uses a premium model 10 times a day to rewrite emails that did not need rewriting in the first place, that's probably a waste. If an AI workflow helps close deals faster, reduce admin work, or speed up invoicing, that's valuable. But if AI is being used because it feels futuristic, that cost is eventually going to become pretty obvious. That's where the market is heading. And we're going from can AI do it? Sure it can, to should AI do it? And what does it cost? And honestly, it's the question that businesses should have been asking from the beginning. I feel kind of duped by the big AI companies, and um, I don't know how I feel about it. I'll have to digest it a little bit. And the uncomfortable truth here is that AI vendors train the market to think in terms of magic. Now they're forcing the market to think in terms of meters. And that's the shift. First, AI was a demo, then it became a habit, and now it's becoming a bill. And that bill is gonna go up and up and up, and it's going to expose companies to what is actually creating value and what AI is really doing for the company.

Practical Guardrails To Control Spend

SPEAKER_00 10:20

So here are some things that you can put into action today. Map the workflows, identify the expensive tasks, decide whether AI creates measurable value, use the cheaper models if they're good enough, and save the frontier models for the work that actually requires frontier intelligence. I don't have frontier intelligence, okay? Don't come for me. Put guardrails around agents, track usage, train people on how to prompt efficiently, build templates, standardize outputs, and most of all, monitor the bill. The companies that will win with AI will not simply be the companies that use the most AI. They will be the companies that used the right AI in the right workflow and at the right cost. Because the free AI era has pretty much ended. The metered AI era is

Free Resources And Closing Thoughts

SPEAKER_00 11:07

here. Check out cybernomics.io. Tons of free resources, up to date AI news, research papers, reports. I give it all away for free because I am on a mission.