If AI got significantly more expensive tomorrow what would happen?

Work with us
Back to SERVICES & INDUSTRIES
Back to Blogs
I do not know if AI is a bubble, but I do not know if people on the side of ‘It is a bubble’ or ‘It isn’t a bubble’ have any concrete evidence they are right either, and that's why it's currently a debate.What I do know as a small business owner and tech bro is that I think it is the wrong question. The question I am asking myself at the moment, and I think other people should too, is:If AI got significantly more expensive tomorrow, or the model you use was no longer available, how much of the new productivity gains your business has experienced would roll back?

For the last couple of weeks, whenever I have been caught scrolling on Instagram, LinkedIn, and TikTok, my algorithm has been showing me more and more content asking, ‘Is AI a bubble?’ I am not a Finance Bro, so I don't know if I am equipped to answer that question; however, I am a Tech Bro, so I want to think about it differently.

A lot of the content I have seen is about how Nvidia is guaranteeing hundreds of billions to OpenAI. OpenAI is signing trillion-dollar deals with Microsoft, AMD and Oracle. This circular passing of billions and trillions of dollars between a handful of companies does concern me.

I do not know if AI is a bubble, but I do not know if people on the side of ‘It is a bubble’ or ‘It isn’t a bubble’ have any concrete evidence they are right either, and that's why it's currently a debate.

What I do know as a small business owner and tech bro is that I think it is the wrong question. The question I am asking myself at the moment, and I think other people should too, is:

If AI got significantly more expensive tomorrow, or the model you use was no longer available, how much of the new productivity gains your business has experienced would roll back?

What Is Actually Happening in the AI Market?

The money really is moving in circles

This diagram from Bloomberg highlights that there are plenty of two-way relationships between the handful of companies propping up AI valuations. Nvidia invested $30 billion into OpenAI. OpenAI is spending its funding with Microsoft and Oracle. Microsoft and Oracle then spend it back with Nvidia on GPUs.

Studies have looked into this, and one found that approximately $1.8 trillion of off-balance-sheet AI debt could be sitting across Alphabet, Amazon, Meta, Microsoft, and Oracle, on top of the $1.3 trillion already on their balance sheets (source).

Goldman Sachs has put AI-related market cap at $18 trillion ahead of their estimates of the profits that AI can actually generate… They have said that they aren’t calling it a ‘classic’ bubble, but as a non-finance techie, that means nothing to me, and I do not know how confident I am in something that seems to be too good to be true.

What does this mean for AI pricing?

I am sure that if you have used AI, you have noticed the price increases, or that ChatGPT's free tier gives you less and less in recent months. You are not going crazy, don't worry. The price of frontier models has doubled in the last 7 months. However, the mid-tier and budget model pricing has fallen ~30% in the last year.

I am not going to pretend to understand the unit economics of AI, I don’t. Yet, as a business, if I need to make more money because I have to pay off debt, there are only a few ways to do that… one is to raise prices for customers. If prices have doubled in the last 7 months and these companies are still heavily subsidised by investors and these two-way deals, I would think it's possible to expect them to double again, and by this time next year could we see it reaching 6 or 8 times the current price…

Don't get me wrong, I understand this is doom-and-gloom talk, but if it turns out to be true, a £ 20-a-month tool today could be £150+ a month per person.

Why This Matters More for SMEs Than for Big Tech

If a big company's thoughts on AI are wrong by 25%, they are fine and might not even notice. If, as an SME business owner, one of your expenses doubles or quadruples and you are anything like me, you’ll notice.

McKinsey reported in a survey that ~80% of people feel more productive using AI. However, only roughly a third of businesses using AI can directly attribute any EBIT growth to it. This is an interesting point: people feel like they're moving faster, but maybe not in the right direction.

Recently at Shape.tech, we have worked with or talked to more and more businesses who are using more and more AI in their business workflows. However, when we dive a bit deeper, we find that:

  • Frontier AI models are glued into business-critical workflows
  • The workflow is no different from the AI model following it
  • No one is keeping track of which vendor they are using
  • People are not tracking how much they are spending on these workflows, never mind a plan for a 2x price rise

It is like having a Swiss Army knife person on your team who is great at everything, doing their own processes and demanding their own salary. You cannot control them because if they go sick or leave, no one else can fill in.

What to Build Instead: Thin AI on Durable Automation

We are now suggesting to founders and business owners who come to us asking about AI that they first build a workflow that works and then use AI to make that workflow faster.

We use a catchy line internally: “Own the workflow. Rent the model.”

The workflow - This is the data model, the integrations, the business rules, the UI, the code. This should all be yours and work without any AI.

The model - ChatGPT, Claude, DeepSeek, whichever AI model is doing the work inside the workflow.

How I found this out the hard way

A couple of weeks ago, I ran the experiment on myself. I swapped our coding practice from frontier models (Codex, Claude Opus) to a cheaper mid-tier model, just to see if the output held up. It didn't. I'm back on frontier pricing now, roughly 3x the cost.

The uncomfortable finding wasn't the quality gap. It was that my workflow wasn't really mine; it was the model's. The prompt I had written to get the output I wanted only worked because the model I was using was so good. Once I took the frontier model away, the output turned to rubbish.

We are now re-engineering the workflow internally to ensure that if we do take the model away to use a cheaper model because of price rises, our output is just as good, if not better.

The Thin-AI Test

Before we add AI to anything for a client, we now ask 5 questions:

  1. Would the workflow still deliver value if the AI output was 50% worse? If not, you are betting the business on a supplier you don't control.
  2. Can we swap the underlying model in a day? If it takes a week, the AI is not a layer; it is a crutch.
  3. Do we own the data and the pipeline? Or does it only exist inside a vendor's product?
  4. Do we know what happens at 2x the price? If a price change would kill the unit economics, that is a written-down risk, not a footnote.
  5. Is there a manual or automated fallback path? Not necessarily fast or cheap, but real.

If a build passes all 5, AI is a genuine accelerant. If it fails 2 or more, we usually recommend investing in the underlying automation first and treating AI as phase 2. Digital Transformation used to be the buzzword and still has tremendous value for businesses that have not taken that step yet. We encourage partners not to get distracted by the shiny new thing.

The Honest Version

I am not trying to predict a market crash. I do not know if there will be one or a correction, a slow deflation, or a soft landing into a productive economy; a recent MIT paper said that these are all still possible outcomes.

What I do know is that the SMEs who will be fine either way are the ones who built durable automation and used AI as a swappable layer on top of it. Those who tied their business to one AI vendor will find out how brittle that is when the bill arrives, or the model changes without their control.

Want to work with us?

Shape is a Newcastle-based software development agency. We specialise in rapid prototyping, user-centred design, and custom software. Technology shaped to fit your business, not the other way round. That approach is exactly the antidote to bubble-era, AI-shaped decisions.

Book a call: https://tidycal.com/shapephil/product-accelerator-discovery

Shape Technical Consulting | www.shape.tech | enquiries@shape.tech | +44 (0)191 535 9615

Phil