2025, Shaping Lorentz, A Different Path for AI Factories
Looking back at 2025, one thing stood out in almost every conversation about AI: scale dominated the narrative.
Bigger models, larger investments, hundreds of thousands of GPUs, gigawatts of power, national AI initiatives, and the push for ever-larger AI factories. Much of the debate revolved around numbers, capacity, size, rankings and megawatts.
Important metrics, but often detached from the question that mattered most to me throughout the year: what does this compute actually need to achieve, and who is truly in control?
We chose a different path with MDCS.ai
When I reflect on 2025, Lorentz marks a turning point. What began as an idea about compute (inspired by Sweden) became something far more concrete, not a product or a single deployment, but a working model for building an AI capability with intent. A model where sovereignty is not claimed through size, but realized through ownership and control, with tenants firmly in the driving seat.
We created Lorentz as a deliberate alternative to the “bigger is always better” reflex. Lorentz is a regional and thematic AI factory concept:
- Regional, because AI capacity needs to be close enough to be governed, owned and trusted.
- Thematic, because real innovation emerges when ecosystems form around shared challenges.
Instead of concentrating everything in one massive facility, Lorentz connects innovation hubs that work on the same problem domains, each supported by its own AI factory, aligned within a broader ecosystem where tenants actively shape priorities, access and evolution.
This approach is intentionally smaller in scale, but sharper in purpose. It is not about maximum capacity, but about relevant capacity. Compute that is accessible, accountable and directly connected to the people solving real problems, with decision-making power where it belongs.
This matters to me personally
Over the years, I have seen too many strong ideas stall because access to AI compute was either too generic, too distant, or effectively controlled by someone else. In 2025, that dissatisfaction turned into motivation. Lorentz is our answer to that pattern. It treats sovereignty as a design principle rather than a reaction. It preserves real choice over where and how AI workloads run. And it positions AI infrastructure as a shared engine for innovation, rather than just another IT asset on someone else’s balance sheet.
Early in the year, I had moments of doubt.
Not about the need for Lorentz, but about whether stakeholders were ready to embrace shared ownership and long-term thinking. Regional and thematic models demand patience. They are harder to finance, harder to govern and harder to explain than simply buying capacity somewhere else.
What changed that doubt into confidence were the conversations that followed.
Throughout 2025, we discussed Lorentz with the parties required to make it real. With financial institutions willing to take measured risk to enable AI compute at a regional level. With NVIDIA as a core technology partner, contributing architectural depth to help shape AI Centers of Excellence. With specialized AI service providers capable of supporting tenants from early ideas to production-grade algorithms. And with entrepreneurs who pushed the concept forward, challenged assumptions and helped turn abstraction into momentum.
These discussions were grounded in reality. With universities under pressure to scale AI research without losing academic autonomy. With public stakeholders who understand the strategic importance of AI, yet remain cautious about long-term financial exposure. And with private organizations that demand performance and predictability, while also caring deeply about sovereignty, compliance and control over their data and models.
One insight became especially clear to me in 2025. Lorentz does not replace other ways of accessing compute. It fits alongside them.
In a world where everyone talks about sovereignty, Lorentz operationalizes it at the regional and thematic levels. It complements national AI initiatives, cloud platforms and other access models by adding what is often missing in practice: transparency, governance and tenant ownership. A layer where shared infrastructure enables collaboration and ecosystem building, rather than isolated contracts and one-off deployments.
By the end of 2025, Lorentz was no longer hypothetical. The foundations are there, technically, organizationally and, most importantly, in mindset.
I want 2026 to be the year where sovereignty proves its value in practice.
Where regions and themes translate ambition into durable ecosystems. And where AI infrastructure becomes a strategic asset that serves innovation, resilience and long-term competitiveness.
That is the work ahead. And that is precisely what Lorentz was built for.
All the best for 2026,
Niels
Read this post on LinkedIn.
| Niels van Rees Co-Founder & Chief Operations | MDCS.AI |
