For years, the debate around digital ownership has focused on one question:
What happens when something digital can no longer be bought, but still exists?
Classic Game Boy, PlayStation and Nintendo games are perhaps the best examples. Millions of people grew up with them. They required vast amounts of engineering, creativity, manufacturing, logistics and natural resources to bring to life.
Today, many of those titles are impossible to purchase legitimately. The companies may still own the copyright, but the games themselves have effectively disappeared from the commercial market.
This has led to years of debate.
Some argue that downloading those games is always wrong because copyright remains in force.
Others argue that if there is no longer any legal way to purchase a title, preserving it causes no commercial harm and may even protect an important piece of digital history.
I’m not trying to settle that debate.
Instead, I think it points us towards a much bigger conversation that is only just beginning.
Artificial Intelligence.
AI is different from traditional software
Most enterprise software is designed around consistency.
Microsoft Excel doesn’t deliberately change how SUM() works every time a new version is released.
Databases don’t intentionally alter how SQL behaves.
Operating systems offer Long-Term Support (LTS) releases because businesses value predictability almost as much as innovation.
AI is different.
Every new model behaves differently.
It may reason in a different way.
It may explain answers differently.
It may become better in one area while performing worse in another.
It may refuse prompts that previously worked.
Even when benchmark scores improve, compatibility isn’t guaranteed.
That’s perfectly understandable while AI technology is advancing so quickly.
But it raises an important question.
What happens when businesses depend on a specific model?
Imagine a healthcare provider spends twelve months validating an AI model.
Every response is tested.
Every workflow is reviewed.
Clinical staff are trained.
Auditors approve the process.
The model becomes part of day-to-day operations.
Now imagine the provider announces:
“This model will be retired in 90 days.”
A replacement exists.
It’s faster.
It scores higher on public benchmarks.
But it behaves differently.
Everything must be tested again.
The hospital isn’t simply buying software.
It’s rebuilding trust.
The same applies to legal services, insurance, financial services and any regulated industry where consistency matters.
AI is becoming infrastructure
When we think about infrastructure, we usually think about electricity, water, roads or telecommunications.
Increasingly, AI belongs in that conversation.
Many organisations are already using AI to support customer service, analyse documents, draft communications, identify fraud, summarise conversations and assist with decision-making.
As reliance grows, AI models stop being interesting technology and start becoming operational infrastructure.
Infrastructure is expected to evolve.
It isn’t expected to disappear overnight.
The environmental question we rarely discuss
Training frontier AI models requires enormous resources.
Powerful GPUs.
Large-scale data centres.
Cooling systems.
Electricity.
Engineering expertise.
Years of research.
The industry often focuses on the environmental cost of training models.
Much less attention is given to how long those models remain useful.
If a model delivers reliable value for ten years, that investment looks very different from one retired after eighteen months.
The most sustainable AI may not always be the newest AI.
Sometimes the most sustainable technology is simply technology that continues to serve a useful purpose.
It’s the same thinking behind repairing devices instead of replacing them.
Innovation isn’t the enemy
To be clear, providers have very good reasons for releasing newer models.
Technology improves.
Costs reduce.
Safety mechanisms become stronger.
Capabilities expand.
None of that is a problem.
Innovation should continue.
The question is whether innovation and continuity need to be treated as competing priorities.
Software has already shown us that they don’t.
Consumers often choose the latest version.
Enterprises often choose stability.
Both markets coexist successfully.
Could AI evolve in the same direction?
A future of AI long-term support
Imagine a future where AI providers offered two clear options.
One is the cutting edge.
Always improving.
Always changing.
Designed for experimentation and discovery.
The other is an enterprise-grade Long-Term Support model.
Its behaviour remains stable.
Security vulnerabilities are addressed.
Performance is maintained.
Outputs remain predictable for five or even ten years.
Businesses could innovate when they choose to, rather than because a retirement date forces them to.
Open models could change everything
Open-weight AI models introduce another possibility.
If an organisation can host a validated model within its own infrastructure, retirement becomes less disruptive.
The provider may move on, but the organisation can continue using the version it has already tested and trusted.
That doesn’t remove every challenge.
Models still age.
Hardware evolves.
Security expectations change.
But it gives organisations far greater control over technology that increasingly underpins their operations.
From ownership to dependence
For decades we asked:
“Do I own this game?”
Digital distribution changed the question to:
“Do I own this licence?”
Cloud AI introduces something even bigger.
Can I own the intelligence my organisation depends on?
Or am I simply renting access to it until someone decides it should no longer exist?
That isn’t just a commercial question.
It’s becoming a strategic one.
The conversation we should start having
AI governance discussions often focus on safety, transparency and bias.
Those conversations are essential.
But perhaps another topic deserves a place alongside them.
Continuity.
How do we preserve trusted AI systems?
How do regulated industries avoid repeating expensive validation exercises every time a model changes?
How do we balance rapid innovation with long-term reliability?
And perhaps most importantly…
Who should ultimately decide when proven AI systems are allowed to disappear?
I don’t have the answer.
But I think it’s a conversation worth having now, before AI becomes as fundamental to society as the technologies we already take for granted.
I’d genuinely be interested in hearing perspectives from people working in AI, healthcare, legal services, software engineering, regulation and enterprise technology.
Are we thinking about AI in the right way—or are we repeating the same debates we’ve already had around digital ownership, only this time with far higher stakes?