The Real Cost of AI in Business

The real cost of AI is more than its subscription price. Behind every prompt sits an energy and water cost in the data centre that answers it. Behind every tool sits a question of where your data lives and who controls it. And behind every poor use sits the cost of getting it wrong. Using AI well means being honest about all of it.

We believe that used intentionally, AI will help us solve some of the trickiest challenges and provide opportunities to democratise technology, however, there are important considerations to make before and during its use.

Key Takeaways

  • The monthly fee is the smallest part of what AI costs.

  • Every prompt consumes real energy and water in a data centre somewhere.

  • Most AI tools store your data on infrastructure you do not control, often overseas.

  • The largest hidden cost is capability: AI used badly is expensive, in money and in trust.

The visible and hidden cost

When people ask what AI costs, they mean the subscription. That is the visible cost, and it is the smallest one. There are three more that matter, and a business using AI seriously should understand all four.

1. The subscription cost.

The visible one. This is real for small businesses as is the increased cost of ‘tokens’; how AI usage is measured by the provider.

2. The environmental cost.

AI needs considerable ‘compute’ to complete responses. These are often located in data centres, and data centres consume significant electricity and water to run and to stay cool. A single AI query uses 10 times (Source) more energy than a standard web search. Across the world, data centre electricity demand is rising by 17% in April 2026 (Source). This is what we mean when we say every prompt has a cost. It does not mean stop using AI. It means use it with awareness, and not wastefully.

3. The data and sovereignty cost.

Most AI tools you use are owned by large overseas companies, running on their infrastructure, storing your data on their terms. For many tasks that is fine. For sensitive work, it is a question worth asking: where does this information go, and who controls it once it is there. We are excited by the feasibility of Sovereign AI and Local AI (Learn more about this in our free video series).

4. The capability cost.

The largest and probably the least discussed. AI used without judgment produces confident, plausible, wrong output, and a business that acts on it pays for the mistake. The cost is not the tool. It is the gap between having the tool and knowing how to use it well.

Every prompt has a cost

We say this often, and we mean it literally. The convenience of AI can make it feel free and weightless. It is neither. The energy and water behind the technology are real, and as a certified B Corp we think businesses should be able to use AI and be honest about that at the same time. Those two things are not in conflict. A capable AI user is also a considered one: clear about when AI genuinely adds value, and unwilling to burn resources on uses that do not.

Watch our video on Sustainable & Ethical AI Use

Where you data lives, and why it is starting to matter

For most of the last few years there was really only one option: cloud AI, run by a handful of large overseas providers. That is changing. Local large language models, regional data centres, and sovereign AI are emerging, and they shift the question from what can this tool do to where does my data live and who controls it.

You do not need to be technical to take this seriously. For an Australian small business, it is worth knowing which of your tools keep data onshore, which do not, and which of your tasks are sensitive enough that the difference matters.

Watch our video on Local LLMs, Data Centres and Sovereign AI

What to actually do about it

Being honest about cost does not mean avoiding AI. It means using it deliberately.

Three practical habits cover most of it.

  • Use AI where it genuinely adds value, and resist using it for things that do not need it. This manages both the environmental and the subscription cost.

  • Know where your data goes. Keep sensitive information out of tools that store it on terms you have not read, and prefer onshore options for work that warrants it.

  • Build the judgment to tell good output from bad. This is the capability that turns AI from a risk into an advantage.

Our position

We help businesses build the capability to use AI well, and we believe being clear about its cost is part of using it well, not a reason to avoid it. That honesty is the difference. We would rather you use AI intentionally for the things that matter than constantly for the things that do not.

If this is the way you want to think about AI in your business, two next steps.

Sign up for our newsletter for one clear, useful piece of thinking a week.

Find out more about Lead with AI

If you’d like to build the capability to use AI intentionally and govern it well, you’d benefit from our eight week capability building and support program Lead with AI.

Sign up for the weekly newsletter

Frequently Asked Questions

  • More than the subscription. It includes the energy and water consumed in the data centre behind every prompt, the question of where your data is stored and who controls it, and the cost of poor output when AI is used without judgment.

  • AI runs in data centres that use significant electricity and water. A single AI query uses ten times more energy than a standard web search, and sector-wide demand is rising. The practical response is to use AI deliberately rather than wastefully.

  • Sovereign AI refers to AI infrastructure and models run within a country, keeping data onshore and under local control, rather than on overseas cloud infrastructure. It matters when your work involves sensitive information and you need to know where that data lives.

  • Not necessarily. It means using it intentionally: applying it where it genuinely adds value, keeping sensitive data safe, and building the judgment to use it well. Often that means using it better, not simply less.

  • Apply it to real problems rather than experimenting broadly, choose tools that match the job and your data needs, keep a person reviewing output, and build the capability to use it well so you are not paying for mistakes.