Meta's Brussels AI symposium and the case against EU battery-replaceability rules for smart glasses

What Meta staged in Brussels

On March 24, Meta hosted an event it called 'Build to Lead: The Brussels AI Symposium.' The company published the opening remarks on April 7, delivered by Markus Reinisch, its VP of Public Policy for Europe.

The guest list is the substance of the announcement. Meta convened European Parliament President Roberta Metsola, US Ambassador to the EU Andrew Puzder, Italian Vice Minister Valentino Valentini, and UK AI Adviser Matt Clifford, alongside companies including EssilorLuxottica and Circus Group. The framing question was blunt: can Europe seize the AI opportunity?

Reinisch's answer sorts Europe's options into three: technological sovereignty, regulatory simplification, and a 'third way' of backing European design, engineering, and hardware. He is dismissive of the first, skeptical of the second's execution, and stakes Meta's pitch on the third.

The specific complaint: battery replaceability in AI glasses

Most of the speech is high-altitude, but one passage is concrete and product-specific. Reinisch cites Meta's partnership with EssilorLuxottica on wearables and singles out a single EU requirement as an obstacle: battery replaceability.

But that uniqueness is undermined by requirements like battery replaceability, requirements that don't fit the reality of this new category and don't help Europe's position as a technology leader.Montana Labs

This is the frontend argument hiding inside a frontier-AI speech. Reinisch is not asking Europe to build its own foundation models. He is arguing that the device on your face — the interface layer where AI actually meets a user — is the layer Europe can win, and that a general-purpose sustainability rule written for phones and appliances is being applied to a device category it did not anticipate.

It is a narrow, testable claim rather than a slogan. Whether replaceable batteries are genuinely incompatible with the industrial design of AI glasses is a real engineering and policy question — and it is the kind of friction that shapes which physical products can ship in a market.

'Relevance crisis' as the organizing frame

Reinisch's rhetorical move is to reject sovereignty as the primary lens. He grants a case for 'certain forms of digital sovereignty' but argues that waiting to 'replicate every layer of the stack' means missing the moment. His phrase is that Europe faces 'a relevance crisis,' not a sovereignty crisis.

He is equally pointed about simplification, the reform agenda Europe has already embraced. His line — that the same people asked to reform the framework 'are often the ones who just finished writing it' — is a critique of execution, not intent. He calls current simplification efforts 'not yet ambitious enough.'

Both positions serve the hardware argument. If sovereignty is a distraction and simplification is underpowered, the remaining lever is removing specific rules that slow specific products — which is where the glasses example does its work.

What this signals for teams building AI hardware in Europe

The honest read is that this is an interested party's argument. Meta is an American company, as Reinisch states plainly, and the passage ends with 'When Europe succeeds, Meta succeeds too.' The battery example advances a product line Meta co-owns with EssilorLuxottica.

But the underlying observation is useful independent of the messenger: as AI moves from browser tabs and APIs onto worn devices, the binding constraints shift from model access to hardware regulation — battery design, radio, sensors, physical form factor. Those rules were largely written for a device landscape that predates AI wearables.

For applied teams shipping AI into physical products in the EU, the takeaway is concrete rather than ideological: the frontend is increasingly a hardware-compliance problem, and the specific regulatory categories a product falls into can matter as much as model performance. That is the real implication of Reinisch's speech — the fight he is picking is at the device layer, not the model layer.