The future of AI demands a hybrid foundation

Every day, I talk to IT executives, commercial leaders, and partners who are facing the exact same pressure: How do we move from AI experimentation to true enterprise production?

If you glance at the tech headlines today, you’d think the answer is simple. The industry has become utterly obsessed with the concept of agentic AI—the promise of autonomous software agents executing complex corporate workflows with the flick of a switch. The mainstream narrative treats AI like a standalone magic box. Buy the right proprietary model, plug it in, and watch your operational headaches vanish.

But out in the real world, the leaders I meet across our global ecosystem are dealing with a far more grounded reality. They know that the margin for error has dramatically decreased, and they are asking a fundamental question: What do we actually build on? Not which vendor, not which cloud, and certainly not just which model—but what is the underlying foundation that lets us innovate at pace without breaking what we cannot afford to lose?

The truth is, the conversation is shifting from "how big is your model?" to "how flexible is your platform?" While frontier models continue to dominate the mainstream conversation, operational reality isn't structured around these goliaths. To move from a flashy desktop demonstration to an accountable enterprise asset, we need to stop looking at AI as a centralised monolith and start realising that the future of enterprise AI isn't size—it is choice.

The myth of the “one-size-fits-all” model

The rush to adopt massive, proprietary, frontier AI models has created a false sense of security. However, these large language models (LLMs) are often complete overkill for everyday enterprise tasks, driving up costs while locking you into a single provider's roadmap.

True enterprise value doesn’t come from a single vendor. Just as there isn’t one single environment powering enterprise IT, enterprise AI’s future is a mixture of proprietary, open source, and highly domain-specific models. Smaller, open source models are ideal for solving specialised, niche business problems where a frontier model’s power and scale would simply be wasted.

For high-stakes industries like finance, healthcare, and automotive, relying blindly on an external, closed-source entity isn't just a business risk—it’s an impossibility. By maintaining smaller, open source models "as-a-service" in-house, enterprises can fine-tune on their proprietary data, maximise performance, and dramatically minimise the risk of data leaks. Context is king, and to optimise your investments, the AI model must map to your specific use case, not the other way around.

Decoupling software from silicon

When I look at the enterprise tech landscape, the biggest hurdle to scaling AI isn't the code or the capability of the model itself. It’s the constraints of the underlying infrastructure.

Right now, compute needs democratisation. Enterprise AI cannot be gated by GPU availability or singular silicon providers. If your entire automation strategy is shackled to a single hardware architecture, you are exposed to massive supply chain risks and skyrocketing costs.

To win the race to production, we must decouple the AI platform from singular hardware families. Choice in AI means having an underlying platform backed by a broad hardware ecosystem—incorporating major providers alongside specialised architectures. Your infrastructure needs to be a consistent abstraction layer. Whether your workloads require edge accelerators on a factory floor or heavy-duty data center workhorses, a unified platform allows you to shift workloads between models and hardware seamlessly as business needs dictate.

Control is currency

As automation becomes deeply embedded in core workflows, organisations are waking up to a critical challenge: digital autonomy. Control is the ultimate currency in modern IT, and AI must follow that exact same path.

Sovereignty has evolved to become more than just a regulatory formality, but a critical business priority for retaining absolute command over your technology supply chain, your operations, and your intellectual property. Because your data is inherently distributed—living on-premises, in multiple public clouds, and out at the edge—your AI workloads need to reside closest to where that data is actually born.

If you don't control the platform where your AI is trained, tuned, and served, you don’t control your business. This is why a sovereign AI strategy demands regional and country-specific boundaries to data ingestion and training. By anchoring your strategy in an open platform, you gain the audibility, transparency, and freedom to inspect and modify code, helping you meet your specific operational needs—not your provider’s.

The foundation for what comes next

We need to stop chasing the frontier swirl and focus on what actually drives commercial and operational success.

The organisations that will thrive in this next era of automation are not those searching for a singular, proprietary magic trick. The winners will be the leaders who focus on building a trusted, flexible foundation designed for choice—allowing them to run the right model, on the right hardware, in the right location. By choosing speed, openness, and platform flexibility, we don't just solve today's complexity; we unlock the unmatched freedom to change, flex, reinvent, and scale whatever breakthrough comes next.