As an someone who continuously tracks the IT industry, I shake my head sadly when I hear some of the tech-related stories propagated by politicians and the media.

A narrative that needs challenging at the moment, for example, is the one that frames the AI discussion as some kind of winner-take-all arms race in which either the USA or China will ultimately emerge victorious. The implication is that the loser will suffer economically, as well as militarily and in other ways.

Europe stuck in the middle?

Here in Europe, many worry about us being stuck in the middle, with debates about digital sovereignty and whether we can muster enough investment to compete with the big dogs. The fear is that we won’t be able to catch up and will end up over-reliant on American tech giants to meet our AI needs.

I don’t want to get into a geopolitical discussion here, not least because I don’t think this prevailing narrative is helpful. If the concern is regional and economic success, the key question, surely, is not how we can take on two superpowers, but how we can enable European businesses more broadly to exploit AI effectively.

Focusing on the opportunity

If you look at any mainstream business, in pretty much any sector, and ask yourself where AI can have a positive impact, the chances are that most of the use cases you come up with are not that glamorous. Contrary to the rhetoric, the latest developments in AI can make a lot of things better, but they don’t universally “change everything” as some would have us believe.

Beyond software development and some other specialist areas where generative and agentic AI can be genuinely transformative, it’s mostly about increasing the efficiency of core business processes, professional/creative workflows, and customer/supplier interactions. Don’t read this as me downplaying the impact and value of AI – it’s the total opposite. The point is that the most high-impact use cases, when viewed through a business lens, are actually quite boring, and not hugely demanding from a requirements perspective.

This impacts the types of model that are most relevant.

Most things don’t need a frontier model

Over the past couple of years, we’ve spent a lot of time at Freeform Dynamics looking at generative AI use cases in a mainstream business context, and the truth is that the vast majority can be easily handled by smaller, less sophisticated models.

In fact, if you throw an expensive frontier model at a lot of requirements, not only would it be overkill, you might also end up with more challenges and risks to work through. The more general the model, the less predictable its behaviour, and the more you have to do around cost control, data handling and governance to keep it in check.

By contrast, smaller models, especially if they have been tailored to a specific domain, tend to be more predictable, cheaper to operate and easier to govern, and they can often run close to the data they depend on.

And there’s no shortage of options here, including ‘open weight’ (similar to open source) models that are free to download, and hardened models trained on tighter, more rigorously controlled data sets, which allows vendors to indemnify customers against IP infringement claims.

But what are the practicalities for mainstream businesses?

No longer the Wild West

It used to be that straying from the AI hyperscalers meant taking a DIY approach, with the hassle of buying the right hardware, installing and configuring models, then running the whole thing on an ongoing basis. Even if you have internal IT resources, you probably want them focused on more productive things – identifying AI use cases, specifying and building applications, and generally supporting the business.

What’s needed is to stop thinking about AI as something special in fundamental systems terms. LLMs are really no different to anything else in that there’s an ecosystem of providers out there offering hosting, implementation services, managed services, support, maintenance and everything else we are used to for other types of systems.

Keeping options open

A lot of ecosystem players also take a model-agnostic stance, helping customers match LLM options to specific requirements. This often also includes helping customers get the most from frontier models where they do add value, or implementing a hybrid approach that combines different models across on-premises, hosted and hyperscaler boundaries.

This raises another important consideration. Nowadays, the term AI is closely associated with LLMs and companies like Anthropic, OpenAI and the other big players in this space. This overlooks not only the rich history of AI more broadly, but the incredible advances still being made in machine learning, computer vision, robotics, world models, expert systems and other areas of AI that are evolving in parallel with LLMs.

Recognising the breadth of the AI domain is important because the non-deterministic nature of LLMs means they’re just not suitable when absolute repeatability and predictability are required. This is why, if you scratch the surface of many AI-enabled automation solutions, you find a blend of LLM technology and other types of AI engine implemented side by side.

All of this underlines the point that frontier models really are just part of the equation, and that we need to take the “AI arms race” narrative with a pinch of salt. As a simple analogy, obsessing over the latest developments in Formula 1 racing doesn’t help you decide which trucks to buy as you modernise your fleet of goods delivery vehicles.

Europe’s secret weapon

Pull all of the above together and it soon becomes clear that European business success with AI has more to do with the strength of our AI ecosystem than whether or how we try to compete in the frontier model space.

As someone who spent a large part of their early career developing and running distribution networks for American tech firms, it strikes me that ecosystem building is something we’re very good at here. Channel partners often succeed by taking a core offering from a big vendor and turning it into a full solution by adding complementary technology, services and local and/or industry-specific know-how. And with open-weight and model-agnostic options making the core increasingly interchangeable, it’s those who assemble the solution who hold the control and much of the value – not whoever supplies the engine.

So let’s not take our lead from people more interested in posturing, sensationalism and scaremongering. Working together, the European AI ecosystem and our mainstream businesses have everything they need to compete effectively on the world stage!