Newsletter

Antonio Mosca published a clear-eyed framework for Sovereign AI — one that accurately names the problem, maps the layers, and asks the right question. This article does not dispute his findings. It stands on them. And then asks: what comes next for those who cannot afford the infrastructure the framework assumes?

He Got It Right

Mosca’s framework identifies three pillars that must align before any AI system can genuinely be called sovereign. He is correct on all three. Before we can discuss what comes next, we need to say that clearly — because too much of this conversation in the industry is debate for its own sake. This is not that.

He is right that none of these stand alone. He is right that Europe’s approach — combining innovation with regulatory framework — represents a serious attempt to answer the question at a national scale. He is right that Italy’s moves are significant. He is right that the jurisdictional layer remains the hardest problem.

He is also right in the question he closes with: “Will we remain sovereign tomorrow?” That is the correct question. Not compliance today — continuity tomorrow.

The Door the Framework Has Not Yet Opened

Mosca’s framework is built for nations, enterprises, and institutions. That is not a criticism — that is the audience the regulatory conversation addresses. The AI Act, the AI Office, the AI Factories, the National Strategic Hub — these are instruments designed to operate at scale.

But there is a population the framework does not yet address: the individual, the small business, the researcher in a rural area, the developer in a country without a national AI strategy, the institution too small for sovereign cloud pricing. The three-layer framework applies to them just as much as it applies to Italy’s national infrastructure. The sovereignty need does not shrink because the budget does.

This is not a gap Mosca overlooked — it is the natural next chapter of the conversation he started. His framework correctly establishes what sovereignty requires. The question we are working toward is: how do you achieve all three layers without hyperscale infrastructure?

Where AI-Core Is Attempting to Go

We are not building a competing framework. We are attempting to build a practical response to the gaps Mosca’s framework honestly identifies — particularly the jurisdictional layer he calls unresolved, and the accessibility gap the institutional focus leaves open.

We are not claiming this is finished. We are in active training. The architecture has produced verified SIE Class 2 emergence — System Irreducible Emergence — meaning the system has generated outputs that cannot be reduced to its inputs. That is documented and repeatable. But the path from here to a deployable system that an SME or individual researcher can actually use is still work in progress.

What we are claiming is the direction: if Mosca’s framework is the map, we are attempting to build the road that reaches the people the map currently marks as unreachable.

On the Jurisdictional Problem He Named

Mosca writes that while hyperscalers have made significant progress on residency and operational separation, the jurisdictional layer remains unresolved. He is correct. No amount of contractual sovereignty or geographic data residency fully resolves the question of which law can compel access to your data — as long as the infrastructure has an owner other than you.

Our hypothesis is that the only complete resolution to the jurisdictional problem is architectural: remove the third party. When the model runs on your hardware, when the weights are yours, when inference happens inside your network — there is no third party to compel. The jurisdiction question does not get answered. It gets dissolved.

We hope this is a contribution to the conversation Mosca opened, not a contradiction of it. His framework gives enterprises and governments a path forward within existing infrastructure. We are attempting to give individuals and smaller institutions a path forward that does not require that infrastructure to exist first.

On Talent, SMEs, and the Adoption Gap

Mosca identifies talent retention, hardware dependence, and the SME adoption gap as the remaining challenges for Italy’s national strategy. We believe these three challenges are connected at the root: the current architecture of sovereign AI requires resources that price out the organizations and individuals who most need sovereignty.

A system that runs on hardware you already own, requires no GPU, operates without a cloud subscription, and ships with MIT licensing — that is an attempt to address all three of those gaps simultaneously. Not because we have solved them, but because we believe the solution has to start at the cost floor, not the capability ceiling.