Who shapes the rules you have to follow?
On lobbying capacity around European digital policy, and on the one case in which you have a direct interest.
Last checked: 4 September 2026
The figures
Corporate Europe Observatory and LobbyControl have been mapping the digital sector's lobbying expenditure in Brussels since 2021, drawing on the EU Transparency Register. Their analysis of October 2025 arrives at the following findings.
The sector spends €151 million a year on influencing European policy, against €113 million in 2023. That is a rise of 33.6 per cent in two years and 55.6 per cent since 2021. The number of companies and associations lobbying on digital policy grew from 565 to 733, partly through the arrival of AI companies.
The concentration is more striking than the total. The ten largest players account for some €49 million between them, a third of the whole. Those ten thereby exceed the ten largest spenders from pharmaceuticals, finance and automotive combined. Meta stands at the top, with a budget above €10 million.
In terms of manpower, the researchers counted around 890 full-time lobbyists working on digital policy, against 699 in 2023. Of these, 437 hold an access badge for the European Parliament. In the first half of 2025, 146 meetings with the Commission and 232 with Members of the European Parliament were registered.
Two qualifications belong with this. The researchers work with the lowest amounts declared in the register, so this is an estimate at the lower bound. And the figures date from October 2025; more recent ones are not available at present.
The case that concerns you: the Digital Omnibus
Abstract lobbying figures say little. For anyone preparing for the EU AI Act there is one concrete case, and it has just happened.
On 19 November 2025 the European Commission proposed the Digital Omnibus, a package amending the EU AI Act among other instruments. On 27 July 2026 it entered into force as Regulation (EU) 2026/1744. The principal effects: the obligations for high-risk systems move to 2 December 2027 for standalone systems and 2 August 2028 for AI in regulated products, and the AI literacy duty has been softened from an obligation to achieve a result to an obligation to use best efforts.
There are two readings of why, and both deserve a fair account.
The Commission's reading
The amendments followed months of consultation, in which, according to the Commission, European businesses were the primary voice heard. The Draghi report on European competitiveness is cited as the reason for acting. Substantively, there was a practical problem: the harmonised standards on which high-risk systems are to rely were not ready. One law firm put the objection sharply: if the Commission and the standardisation bodies do not meet their own deadlines, how can companies be expected to comply with requirements that have not yet been written. Another firm describes the amendments not as deregulation but as concessions at a practical level.
The critics' reading
133 civil society organisations and trade unions signed a joint call to halt the proposal even before publication. European Digital Rights described it as a considerable rolling back of European digital protection. Corporate Europe Observatory documented that specific amendments could be traced to lobbying points made by large technology companies. A recurring objection: the Commission carried out no impact assessment, while maintaining that the amendments would have no consequences for fundamental rights.
The Dutch government took an intermediate position: support for the simplification for smaller companies, but criticism of tying entry into application to uncertain decisions, with a preference for fixed dates and a shorter postponement. That last line prevailed.
We do not settle this discussion here. What is established is that the outcome landed between the two original positions: the deadline for high-risk systems has been extended and part of the administrative burden removed, while the risk-based structure, the literacy duty and most other dates remained in place.
Why this is a planning problem
For an organisation, the political question matters less than the practical one, which is: what date can I plan on.
Until the end of July 2026, lawyers advised continuing to work towards 2 August 2026, because the postponement was not yet legally settled. The package appeared in the Official Journal on 24 July and entered into force through an urgency procedure on 27 July, six days before the original deadline. Organisations that had prepared for August 2026 did not do that work for nothing, but they did it at a different pace from the one that turned out to be needed.
That is the real consequence of a file like this. Not that the rules disappear, because they did not. Rather that the date on which they apply was uncertain until days before the deadline.
A related but different phenomenon: the Amsterdam permit
What appears above concerns influence over legislation. What follows concerns something else: the room available within an existing definition. The two are often lumped together, and that is not warranted.
Since 2022 the Netherlands has restricted the construction of hyperscale data centres. The definition requires two criteria at once: a floor area from 100,000 square metres and a connection of more than 70 MW.
In the Western Port Area of Amsterdam, Pure Data Centres Group, financed by Oaktree, is developing a complex of three 85-metre towers, each with 26 MW of data halls, 78 MW in total. Microsoft is the tenant, not the applicant for the permit. The province of North Holland granted separate permits per tower, with the result that none of the three meets both criteria on its own.
The province later acknowledged that the project could be regarded as a single data centre when taken together, and pointed to the height of the buildings and to the fact that the plans dated back to 2016, before the restriction. Members of the provincial council spoke of a way round the rules. The responsible minister stated that there were no legal means available to stop the construction. A member of the Dutch parliament suggested changing the definition from "and" to "or".
This is not lobbying. It is a definition doing exactly what it says.