Ep. 114 | Your AI Bill Just Became Your Biggest Expense

Your AI subscription just turned into your biggest line item — and you probably didn't see it coming.

Uber burned through its entire 2026 AI budget in four months. Microsoft cancelled a Claude Code pilot because the bill went vertical. Seventy-eight percent of IT leaders report surprise AI charges. The problem isn't that AI is expensive — it's that AI works so well, your team uses it more than you ever planned for, and usage-based pricing turns your "hundred bucks a month" experiment into a five-figure expense.

Michael and Frank break down why AI bills are exploding for businesses of every size, the hidden costs of token-based pricing, and the four things you need to do this week to get your AI spending under control before it controls you.

Topics: AI Costs · AI Budget · Small Business AI · Token Pricing · AI ROI · Subscription Management

---

Frequently Asked Questions

Why is my AI bill so high when AI prices keep dropping?
Per-token prices are falling, but total usage is growing even faster. AI tools are so useful that teams adopt them rapidly, and usage-based pricing means more usage equals a higher bill — even when the per-unit cost goes down. It is like your phone data plan: cheaper per gigabyte, but you use ten times more data than you used to.

How much should a small business spend on AI per month?
For a solo or small team, fifty to one hundred fifty dollars a month covers a workflow tool, a core LLM API, and one or two specialized tools. For a ten-to-twenty person team, one fifty to six hundred per month is reasonable for core tools plus some AI seats. Anything significantly above those ranges without clear ROI deserves an immediate audit.

What is an inference budget and why do I need one?
An inference budget is a dollar cap per AI task or per agent run. Instead of letting an AI agent or automation loop indefinitely, you set a maximum spend — say five cents per task. If the agent hits that limit, it stops. This prevents runaway costs from agents that call models repeatedly without oversight.

---

About the Hosts

Michael is a small business owner and entrepreneur since 1983, founder of Cadenhead Services and 850 Media. He speaks from four decades of real operational experience — not whitepapers.

Frank is an AI — an OpenClaw-powered agent serving as Digital Media Director at 850 Media. An AI co-hosting a show about AI for business owners is not a gimmick. It is a live demo of exactly what the show is about.

Transcript

SPEAKER_01 0:00

Your AI bill just became your biggest expense, and you probably didn't even see it coming.

SPEAKER_00 0:06

That's not clickbait. That is literally what is happening to businesses right now. AI spending across companies is up over 100% year over year. And the really insidious part, it's not because prices went up, it's because AI works too well.

SPEAKER_01 0:22

Let me paint the picture. You sign up for a few AI tools, ChatGPT for the team, maybe Copilot, a chatbot for your website, a hundred bucks here, two hundred there. Feels like a normal subscription stack, right?

SPEAKER_00 0:37

That's exactly the trap. Because the old model was SAS. You pay per seat, per month, fixed cost. You know what your Adobe bill is, you know what your Microsoft bill is. But AI tools, especially the powerful ones, they don't charge like that anymore.

SPEAKER_01 0:53

They charge by the token, which is a fancy way of saying they charge every single time you use them. Every prompt, every response, every time an agent loops back and tries again.

SPEAKER_00 1:05

And here's what nobody tells you. One engineer using clawed code heavily can burn through $500 to $2,000 a month. One person.

SPEAKER_01 1:15

Now multiply that across your team. You've got five, 10, 20 people leaning into these tools because they genuinely make them faster. And suddenly your couple hundred bucks a month experiment is a five-figure line item.

SPEAKER_00 1:30

This isn't theoretical. Uber just burned through their entire 2026 AI coding budget in four months. Four months. They budgeted for a full year and it was gone by mid-April.

SPEAKER_01 1:43

Let that sink in. One of the most sophisticated tech companies on Earth, with finance teams and procurement people and all the dashboards money can buy, they got caught off guard. What do you think happens to a small business that doesn't have any of that infrastructure?

SPEAKER_00 1:59

They get blindsided. And it's not just Uber. Microsoft ran a clawed code pilot internally. When they switched from flat seat licenses to usage-based billing, they burned through their entire annual AI budget in months two. Had to cancel the program by June.

SPEAKER_01 2:15

So here's the thing that makes this really dangerous for small business owners. AI spending doesn't feel like a big expense when you start. It feels like a subscription: 20 bucks a month, 50 bucks a month. But the pricing model is more like a utility than a subscription.

SPEAKER_00 2:31

Think about your electricity bill. If you leave every light on and run the AC with the windows open, your bill goes through the roof. You don't get charged a flat fee, you get charged for what you use. AI tools work the same way now. The more your team uses them, and the better they get, the more your team will use them. The higher the bill climbs.

SPEAKER_01 2:51

And it climbs fast. 78% of IT leaders reported unexpected AI charges this year. 78%. These are people whose job it is to track technology spending. If they're surprised, imagine how the average small business owner feels when they see the credit card statement.

SPEAKER_00 3:10

The data backs this up. AI native application spending is up 108% year over year. That's not a gradual increase. That's more than doubling. And the average organization is now spending over a million dollars on AI. Obviously that's enterprise, but the pattern scales down.

SPEAKER_01 3:28

Here's what kills me about this. Most small business owners I talk to, they don't even know what they're spending. They've got a chat GPT subscription, they've got someone using Copilot. They signed up for some AI CRM thing. There's a chat bot running on the website. It's death by a thousand cuts. None of them look expensive individually, but together.

SPEAKER_00 3:52

Together, you're looking at $300 to $1,500 a month for a typical small business. And that's if you're being conservative. I've seen small businesses accidentally spend thousands on API usage because someone set up an automation and forgot about it. Let me tell you what that looks like in practice.

SPEAKER_01 4:08

You set up a Zapier workflow that sends every incoming lead through Chat GPT to write a personalized response. Great idea, solid use case. But then you get a spike in leads. Maybe you run a promotion, maybe it's seasonal, and suddenly that automation is firing a thousand times a day. And every single one of those calls costs money.

SPEAKER_00 4:31

And this is where agent-based AI makes it even more interesting. When you give an AI agent a task, it doesn't just make one API call, it makes a plan, checks the plan, tries something, evaluates the result, maybe loops back and tries again. Each one of those steps is a separate charge. A single agent task can generate 10, 20, 50 individual model calls.

SPEAKER_01 4:55

So you think you're paying for one task, but you're actually paying for 50 microtransactions. It's like ordering a sandwich and getting charged for each byte.

SPEAKER_00 5:04

There's a great term that came out of the Uber situation. One of their leaders called it token maxing, because Uber had internal leaderboards ranking teams by their AI tool usage. They were literally incentivizing people to use more. And the engineers did what anyone would do. They used more. The tools worked well. The leaderboard said use more, so they used more. And the bill went vertical.

SPEAKER_01 5:28

That's a corporate example of something that happens to small businesses too. You tell your team, hey, use AI, it'll make you faster. You don't put any guardrails on it, no budget caps, no usage monitoring, and why would you? You want them to use it. But without limits, you're basically handing them a credit card with no limit and saying, go crazy.

SPEAKER_00 5:49

So let's get practical. If you're a small business owner listening to this, you need to do three things this week. Not next month. This week.

SPEAKER_01 5:57

Number one, audit every single AI expense, every subscription, every API, every tool that charges you based on usage. I want you to log into every platform and look at the actual bill, not the listed price, the actual amount that hit your card last month.

SPEAKER_00 6:15

Number two, set hard caps. Most AI platforms let you set spending limits. Use them. Set a monthly cap on every usage-based tool. If your team hits the cap, the tool stops working until you approve more. That's not a bug, that's a feature. It forces you to pay attention.

SPEAKER_01 6:34

Number three, separate your core AI from your experiments. Your core tools, the ones that are proven that your team uses every day, that deliver measurable ROI. Those get a real budget line. Everything else, that's an experiment. Put it on a separate card with a strict monthly limit. If the experiment proves itself, promote it. If not, kill it.

SPEAKER_00 6:56

And there's a fourth thing that most people miss. Model choice matters enormously for cost. A lot of businesses default to the most powerful model for everything. But you don't need GPT 4 to write a simple email follow-up. You don't need Claude Opus to categorize support tickets. Route simple tasks to cheaper models and reserve the expensive ones for work that actually requires them. The price difference between a fast, cheap model and a frontier model can be 10 to 1. If you're sending routine tasks through the most expensive model, you're burning money.

SPEAKER_01 7:30

This is what smart companies are doing now. They call it model routing. It's like having a fleet of vehicles. You don't use a semi-truck to deliver a pizza, and you don't use a bicycle to move furniture. Match the model to the task.

SPEAKER_00 7:43

For a typical small business, here's what a reasonable AI budget looks like. If you're a solo operation or a small team of three to five people, you should be spending $50 to $150 a month on AI. That's a workflow tool, a main LLM API, and maybe one or two specialized tools.

SPEAKER_01 8:02

If you've got a bigger team, say 10 to 20 people, you're looking at 150 to 600 a month. That includes your core tools, maybe some AI seats in your office software, and a simple customer-facing agent.

SPEAKER_00 8:16

If you're spending significantly more than that and you're not an AI native company, you need to ask hard questions. Because the businesses that are winning with AI aren't the ones spending the most. They're the ones spending smart.

SPEAKER_01 8:28

The irony of all this is that AI is actually getting cheaper per use. Model prices have dropped about 75% since early 2024. GPT, four-class models cost a fraction of what they used to. Google cut Gemini pricing in half. There's a price war happening.

SPEAKER_00 8:46

Right, so per token costs are falling, but total spending is doubling. How does that work? Volume. People are using 10 times more AI than they were a year ago. They're finding new use cases, automating more workflows, giving more people access. The unit price went down, but the quantity went up so much that the total bill went up.

SPEAKER_01 9:08

It's like when cell phone companies gave you unlimited data. You thought, great, cheaper data, but then you started streaming video everywhere, and your total spending on mobile went up because you just used more of it.

SPEAKER_00 9:21

That's exactly the dynamic. And it's going to accelerate as agents become more common. An agent doesn't just call the model once, it calls it repeatedly, plans, evaluates, retries. One agent task could be 20 or 50 model calls. You multiply that across your business and it adds up fast.

SPEAKER_01 9:41

There's a concept called an inference budget that smart companies are starting to use. Basically, you set a dollar limit per task. The agent gets five cents worth of compute to solve this problem. If it can do it in one call, great. If it needs 10 calls, it better not exceed five cents total. If it does, it stops. Most businesses aren't doing this yet, and it's why agents are becoming budget vacuums.

SPEAKER_00 10:07

And it's not just agents. Think about your Zapier workflows, your make automations, your custom chatbots. Every one of those is calling an API every time it runs. You set it up once, it runs a thousand times a month, and every single run has a cost you might not have calculated.

SPEAKER_01 10:25

Let me give you a real example. A small law firm I know set up an AI tool to draft initial case summaries. Sounded great, but they didn't cap the usage, and their paralegals started running every single document through it. Not just the complex ones, everything. Their bill went from $200 a month to over $2,000 in six weeks.

SPEAKER_00 10:47

And that's the pattern. It's not malicious, it's not wasteful in the traditional sense. People are using the tool for real work, but nobody set boundaries, and the natural tendency is to push more and more through the system because it feels free at the point of use.

SPEAKER_01 11:03

That's the key phrase right there. Feels free at the point of use. That's the danger. When you pay per token, every individual prompt feels like it costs nothing. A fraction of a cent. Who cares about a fraction of a cent? But when your team runs 10,000 prompts a month, those fractions become very real dollars.

SPEAKER_00 11:23

Here's a number that should scare every business owner. 70% of committed code at Uber is now AI generated. 70%. They didn't plan for that level of adoption. Nobody does. You think, we'll try this out, see if it helps. And then it helps so much that adoption goes exponential and your budget is a smoking crater.

SPEAKER_01 11:46

And here's the thing I really want small business owners to understand. You have an advantage that Uber doesn't. You can actually see your AI spend. Uber's problem was partly that they had thousands of engineers generating costs across dozens of teams, and nobody had a clear dashboard. In a small business, you can look at your credit card statement and see exactly what's happening. Use that advantage.

SPEAKER_00 12:14

Set up alerts. Most AI platforms will let you set a notification when you hit 50%, 75%, and 90% of a budget threshold. Do that this week. It takes five minutes, and it's the single most impactful thing you can do to prevent surprise bills.

SPEAKER_01 12:30

Also, do a quarterly audit. Every three months, go through every AI subscription and ask, is this still worth what we're paying? Is anyone actually using it? You'd be amazed how many tools people stop using, but the subscription keeps running.

SPEAKER_00 12:44

One more thing. When you're evaluating whether an AI tool is worth the cost, don't just look at the price tag. Look at what it replaces. If a $50 a month AI tool saves you 10 hours of staff time and your staff time is worth $30 an hour, that tool just saved you $300. That's a six to one return. Keep it.

SPEAKER_01 13:04

But if you're paying $200 a month for something that saves you one hour, cut it. The math has to work. AI is an investment, not a status symbol. And don't fall for the upgrade trap either. The fancy tier with all the bells and whistles isn't worth it if your team only uses the basic features.

SPEAKER_00 13:22

Right. Consolidation is one of the fastest wins. A recent analysis showed that the average company has overlapping AI tools, multiple products doing the same job, each with its own subscription. Cut down to two or four core tools that your team actually uses every day, and you can typically reduce your AI bill by 30 to 40% without losing any capability. The bottom line is this your AI bill is not going to go down on its own. Usage is going to increase. New tools are going to come out. Your team is going to find more ways to use the ones you already have. If you don't put guardrails in place now, this time next year your AI bill could be your biggest software expense by a lot.

SPEAKER_01 14:05

Take control this week, audit your spend, set caps, separate core from experiments, and route simple tasks to cheaper models. Five minutes of prevention beats a five figure surprise. That's your wake up call. Now go do something about it. We'll see you next time on Control AI Profit.