Almost every company today rents its AI by the question, from a model somebody else owns. That is starting to change, because open models have closed enough of the gap to run privately, and when the model moves inside your business, three things move with it: your costs, your visibility of what is actually being used, and where your legal exposure sits.
You are paying by the question
Everything your people use today runs on somebody else’s model. OpenAI, Anthropic, Google, Microsoft. Closed, owned and metered. Every email drafted, every spreadsheet cleaned up, every document analysed.
Think about what that means. The better your adoption goes, the more you pay. You are being charged for succeeding at the thing you asked your organisation to do.
For a while there was no alternative. The open models were not good enough, and no serious business runs on second best.
That has changed
The open models, several of them Chinese, are now close enough to the closed ones that for most ordinary business work you would struggle to tell the difference. The way they are built underneath is changing too, so the machines needed to run them privately are getting smaller rather than larger. The gap between open and closed models is now measured in months, not years: since January 2026, the most capable open-weight models have lagged the closed frontier by an average of just four months on Epoch AI’s Capabilities Index, down from close to a year in late 2024.
I am not going to give you a date. Anyone who does is guessing, and the date is always the thing people use to dismiss the argument. But the direction is clear enough to plan against now.
I keep picturing the horse and buggy. Before that it was your own back, carrying goods to market. Then a horse, which got tired. Then a carriage, and suddenly you carried more for less effort. What we are doing with AI today is the back-carrying stage. It works. It is just the most expensive way we will ever do it.
What changes when the model lives in your house
Your costs stop being a meter. They become something your finance people already know how to handle. Whether that is cheaper depends on how much you use, and if you are using AI seriously, it will be.
You can finally see what is running. Right now a good deal of the AI inside your company is invisible to you. Personal accounts. Tools nobody approved. That is shadow AI, and it exists for reasons worth understanding before you try to stop it.
And that is where it gets legally interesting. The moment you can see what every employee is asking, you are handling personal data about your own staff. In Norway that is already covered by personopplysningsloven and the rules on control measures at work. It did not wait for the AI Act. Datatilsynet has published guidance on monitoring employees’ digital activity, and it applies to this whether or not anyone intended surveillance.
Running AI locally does not remove your exposure. It moves it into your house, where it is yours. Almost nobody writing enthusiastically about local AI mentions that, which is one reason boards keep asking better questions than they are given answers to.
AI has not disappointed you. Strategy has.
I meet leadership teams who are quietly angry about what they spent and what came back. They have decided the technology was oversold. That is almost never what happened.
AI was bought as a project and run as a hobby. A bit here, a pilot there, something so the company would not be left off the list of companies doing AI. Nobody was ever given the job of owning the whole thing, and nobody understood both the business and the technology well enough to say what should stop.
If your AI has not delivered against its budget, it is not the technology’s fault and it is not yours. You were badly advised. The same pattern sits underneath every pilot that quietly went nowhere.
What I would do this month
The decisions coming look technical and are not. Where the money goes. What you build inside the business and what you keep buying. Which vendors are worth their contract term.
You cannot answer any of them until someone can tell you three things. Which AI systems are actually running. Which of them touch a decision about a person. And who owns each one, by name.
Most companies cannot answer that today. That is not carelessness. Nobody was ever given the job. It is the question I put to every board I work with, and the one I keep returning to on stage.
Questions this article answers
Are open source AI models good enough for business use?
For most ordinary business work, yes. Drafting, summarising and analysis are now close enough to the closed frontier models that the difference rarely matters in practice.
Will running AI locally save money?
It converts a per-query meter into a fixed cost with maintenance attached. Whether that is cheaper depends on volume, and for organisations using AI seriously it usually will be.
Does running AI in-house reduce legal risk?
No. It relocates it. Seeing what employees ask means handling personal data about staff, which in Norway is already governed by personopplysningsloven and the rules on control measures at work.
Why has our AI investment underdelivered?
Almost always because AI was bought as a project and run as a hobby, with nobody given ownership of the whole picture.
What should a board ask first?
Which AI systems are running, which touch a decision about a person, and who owns each one by name.
This edition is adapted from the Clarity at the Top podcast.
If you are leading your organisation through this, this is the kind of work I do with a limited number of senior leaders each quarter. → https://anglero.com/the-reset/
Strategic AI Advisor, MerkabaPhi AS, Oslo. 450+ keynotes across 30+ countries.
Listen to the Clarity at the Top podcast on Spotify.
Enquiries: anglero.com