Fridays, 2.30 pm (Lisbon) · Artificial intelligence on NOW →Fridays, 2.30 pm · AI on NOW →
Miguel Sousa SoaresMSS
← Writing
LinkedIn · Essay · 19 March 2026

Portugal, the United States’ Digital Trading Post

Sines is the emblematic case of a broader shift: Portugal is becoming critical infrastructure for Europe’s digital economy. The question is how much power it retains.

At the Web Summit in November 2025, Microsoft President Brad Smith gave an interview to Jornal de Negócios and said, triumphantly:

“Countries across Europe are competing for European Union public funding to build an AI gigafactory. The competition is intense. Portugal has already won the bid with Microsoft, because we decided to build this gigafactory in Sines.”

The statement was celebrated. The government applauded. The press reported it. And with good reason: Microsoft committed $10 billion to install 12,600 of the world’s most advanced graphics processors in Sines, the NVIDIA Blackwell Ultra GB300, designed specifically to train large-scale artificial-intelligence models. It is the largest single investment in AI computing ever announced on European soil.

The enthusiasm is understandable. What is missing is the question no one asked Brad Smith: on what terms?

“Portugal has already won” can be read in two ways. The first, the one celebrated, is that the country has established itself as a leading destination in Europe’s digital economy. The second, less comfortable, is that it has won the privilege of hosting critical infrastructure whose control, jurisdiction and value remain elsewhere. The difference between these readings is not semantic. It is strategic. This article explores that difference.

A quiet transformation, and its true scale

Portugal now has around 47.5 megawatts of data-centre capacity in operation. More than 1,800 megawatts are in the pipeline, under construction or at an advanced planning stage. That means eleven new data centres, operated by thirteen entities, most of them foreign-owned.

The arithmetic is revealing: Portugal is on the verge of multiplying its digital infrastructure forty-fold in less than a decade. To gauge the scale of this transformation, consider that Start Campus is building a six-building campus in Sines with a total capacity of 1.2 gigawatts; on its own, it is larger than all the capacity currently installed in Portugal. Construction investment in the campus exceeds €8.5 billion. Microsoft is adding $10 billion in computing equipment and infrastructure on top of that base. These are two distinct investments (one in real estate and energy, the other in silicon and software) overlaid on the same territory.

Sines is therefore not just a project. It is the emblematic case of a broader shift: Portugal is moving from being merely a user of the digital economy to becoming one of its physical support bases. The decisive point is that hosting infrastructure is not the same as controlling the value chain built on it.

It is tempting to read these figures as proof of success. But the real question is not how many gigawatts are coming to Portugal. It is what Portugal retains when they arrive.

Data centres are not data warehouses. They are digital factories. And those who control the factories control a growing share of the global economy’s cognitive capacity.

These centres process the data that power online services, healthcare systems and financial platforms, and increasingly the artificial-intelligence models reshaping work, medicine and geopolitics. The oil analogy is inadequate: oil is extracted, transported and consumed. Data is processed, learns and reproduces itself. Each processing cycle deepens the advantage of whoever controls the infrastructure.

But infrastructure alone does not capture the essential value. Value accumulates in the intellectual property of the chips, the software that orchestrates computing, the cloud platforms where services run, the business contracts that monetise that capacity, the models trained on it and the recurring revenues those models come to generate. That is where it is decided who hosts and who commands. In Sines, the infrastructure will be in Portugal. Computing power, intellectual property, monetisation capacity and jurisdiction will, to a large extent, be elsewhere.

Who pays the bill: the Irish lesson

Before discussing legal sovereignty, we need to talk about electricity. Portugal’s transformation has a very concrete physical cost. Ireland has already paid it.

Dublin was Europe’s first major data-centre hub. For years, the model looked perfect: massive foreign investment, skilled jobs and tax revenue. But in 2021 EirGrid, Ireland’s electricity-grid operator, issued an unprecedented warning: the system could not accommodate more data-centre load in Greater Dublin without putting supplies to homes and essential services at risk. A de facto moratorium followed and lasted four years.

The figures that have since emerged are alarming. According to data from EirGrid and the CRU (Ireland’s Commission for Regulation of Utilities), cited in a confidential Irish government memorandum reported by RTÉ in August 2025, data centres and technology companies account for 89% of projected electricity-demand growth over the next decade. In its review of network tariffs, the CRU estimated that household bills could rise by between 8% and 21% over the next five years to fund the network upgrades and generation required by this demand. By contrast, large consumers, including the data centres themselves, could see their bills fall by 14%.

In Ireland, those who create the demand pay less. Households paying their electricity bills finance the grid upgrades. This is not a regulatory accident. It is the predictable result of failing to set conditions before letting investment in.

The response came in December 2025, when the CRU approved a new grid-connection policy for large consumers (CRU/2025236). The rules are now demanding: a new data centre above certain power thresholds can only connect if it funds dispatchable generation or storage capable of covering 100% of its maximum import capacity; commits to sourcing at least 80% of its annual energy from new renewable generation in Ireland within six years; and accepts that the regulator may reject projects in areas with grid constraints. The logic is explicit: the investor brings the demand, and the energy.

Portugal is now where Ireland was in 2019. The pipeline is growing rapidly. The electricity grid (which supported consumption of 51.4 TWh in 2024, 71% of it from renewable sources, but with a significant net-import balance) cannot be reinforced at the pace of the market. It is reinforced at the pace of the state. And the Portuguese state has not yet publicly defined who will pay for that reinforcement.

The current regulatory framework points in a different direction. Decree-Law 84/2024 transposed European reporting and energy-efficiency obligations for data centres into Portuguese law. It is a serious step, but essentially a transparency and compliance regime. The dominant signal remains one of competitive attraction: the Electro-intensive Customer Status offers large consumers with a green profile reductions of up to 85% in general economic-interest costs within grid-access tariffs. Ireland requires the investor to bring energy. Portugal offers a discount to those who are already efficient. These are two opposing philosophies, and only one protects household consumers.

The sovereignty geography cannot guarantee

The energy question is visible and measurable. The legal question is deeper and less discussed.

A belief has taken hold in European public debate: if a server is in Portugal and data resides in EU territory, it is protected by the GDPR and European territorial sovereignty. At best, this belief is incomplete; politically, it is convenient for those who do not want to ask difficult questions.

The US CLOUD Act (Clarifying Lawful Overseas Use of Data Act), in force since March 2018, authorises US authorities to require American companies to provide data stored anywhere in the world, regardless of local law. The obligation falls on the company, not the server. An American company operating a data centre in Sines remains subject to US law, wherever the data it processes is physically located.

This is not a hypothetical scenario constructed for argument’s sake. The case of United States v. Microsoft Corp. demonstrated it in practice: the US Department of Justice demanded access to emails stored in a Microsoft data centre in Dublin, and the dispute reached the US Supreme Court. The CLOUD Act was passed precisely to resolve this legal tension. It resolved it unequivocally in favour of American extraterritorial reach.

The Court of Justice of the European Union responded twice with equal resolve. In October 2015, the Schrems I judgment (C-362/14) invalidated the Safe Harbour regime. In July 2020, Schrems II (C-311/18) invalidated the Privacy Shield. In both cases, the Court found that US surveillance programmes, particularly FISA Section 702 and Executive Order 12333, were incompatible with fundamental European rights. The third arrangement, the EU-US Data Privacy Framework (DPF), was adopted by the European Commission in July 2023 and upheld at first instance by the General Court in 2025 (Latombe v Commission, T-553/23). It is formally in force. Since then, however, the new US administration has weakened some of the oversight bodies on which the framework rests, and several European legal scholars warn that its foundations have been undermined. For many, a third Schrems case is only a matter of time.

A server’s geographical location does not, by itself, determine its legal sovereignty. When Microsoft operates in Sines, the data is technically in Portugal. The company processing it remains subject to US law. In a conflict between the two jurisdictions, there is currently no fully binding, automatic and tested mechanism that guarantees, in practice, the primacy of European law.

The point is not that Microsoft is a hostile actor; it is not. The point is structural: the legal architecture of the global digital economy was built so that sovereignty is determined not only by the physical location of servers, but above all by the jurisdiction of their operator. Portugal may have the servers. The United States retains significant legal instruments over those who control them.

The trading post, and what history teaches

In the sixteenth century, Portuguese trading posts were obligatory waypoints in global commerce. There was movement and visible wealth, and Portuguese people managing and navigating the routes. There was genuine national pride, and genuine economic benefit. But capital was often Venetian, Flemish or Genoese. The routes were Portuguese; the intermediation was Portuguese; yet decisive value was often accumulated elsewhere. When the centre of gravity of world trade shifted, the trading posts emptied, not because Portugal had failed technically, but because it did not sufficiently control the links in the chain where financial and commercial power was concentrated.

The analogy is imperfect; no historical analogy is perfect. But the structure is recognisable.

Today Portugal offers territory overlooking the Atlantic, inexpensive renewable energy, connectivity through submarine cables linking three continents, and an EU regulatory framework that provides legal certainty. In return, it receives investment, jobs and tax revenue. These are real benefits, and it would be intellectually dishonest to deny them.

But the chips are NVIDIA’s. The software is Microsoft’s. The platforms are American. Artificial-intelligence models, the most valuable cognitive assets of the twenty-first-century economy, are developed, trained, owned and commercialised outside the country. Portugal participates in the value chain but does not control its most concentrated centres of economic power.

This is where the analogy stops being literary and becomes structural. A trading post was not a place without activity; it was a place where activity did not equate to command. Portugal’s risk is not being shut out of the digital economy. It is entering it in a functional, useful and short-term profitable position that remains strategically subordinate.

The risk is not irrelevance; Portugal is too well placed to be ignored. The risk is comfortable subordination: a dependency that takes root precisely because it benefits both sides in the short term, and becomes costly only when it can no longer be reversed.

What Portugal can and should demand

Portugal is not powerless in this equation. On the contrary, it is in a negotiating position that will not recur: the pipeline has not yet been built, contracts are still being negotiated, and demand for this location is structurally high. The window for setting conditions is open. It is impossible to say for how long.

An active digital-sovereignty policy (neither protectionist nor hostile to investment, but genuinely strategic) would entail at least three concrete requirements.

The first is simple: those who create demand must internalise the costs they impose on the system. This means mandatory grid-cost-sharing clauses in all connection agreements for large electricity-intensive consumers. The post-2025 Irish model is the benchmark: the investor brings the demand and finances the firm capacity needed to support it. Without this condition, the country privatises the immediate benefit and socialises the structural cost through household tariffs.

The second follows directly from today’s legal vulnerability: projects of this scale should operate through Portuguese legal entities with autonomous operational responsibility, stronger governance and audit requirements, and clear regulatory accountability. This does not eliminate exposure under the CLOUD Act (no unilateral measure can do that), but it reduces asymmetries, creates additional layers of protection and makes it harder for public debate to confuse physical location with effective sovereignty.

The third may be the most decisive: tax benefits, accelerated licensing and access to public land should be conditional on a genuine transfer of capabilities to Portugal. Not just operational jobs (maintenance technicians, facility managers and support teams) but engineering, research and development, advanced training, university partnerships, systems-design capabilities and a skilled presence in the upper layers of the value chain. The goal is not to replace foreign investment with domestic capital that does not exist today. It is to avoid allowing foreign presence to occupy space without leaving lasting capabilities behind.

None of these measures is radical. They all follow the same logic: if Portugal provides land, energy, connectivity, regulatory stability and European legitimacy, it has every right to demand value retention, internalisation of costs and the development of domestic capabilities. This is not hostility to investment. It is mature negotiation.

The question, then, is no longer whether Portugal should welcome this investment. It should. The question is whether it welcomes it as a strategic partner or as a passive platform.

The choice that remains open

Portugal is approaching a tipping point: the moment when data centres stop being a marginal opportunity and become a force shaping the country’s economic, energy and geopolitical system. Before that point, conditions can be negotiated. Afterwards, they are accepted.

Portugal can become an intelligent node in Europe’s data economy: demanding about terms of entry, able to retain value, internalise costs and build its own capabilities from foreign presence. Or it can become an efficient platform: useful for energy, legally exposed and strategically dependent on decisions made in Seattle or San Jose.

The word used in official discourse is “hub”. It is neutral. Technical. It implies no power. It implies passage.

In the sixteenth century, there was another word for a place through which wealth passed without command remaining there.

Portugal knows that word better than any other European country. The question is whether it recognises the model when it is applied to itself.

Methodological note

Data on Portugal’s electricity system is based on public sources from REN and ERSE. Figures on installed data-centre capacity and the pipeline in Portugal come from recent sector reports by Research and Markets and Arizton. They are used as orders of magnitude, not as measurements taken at perfectly comparable points in time.

The Start Campus investment refers to construction CapEx for the SINES DC campus; Microsoft’s announced investment refers to the computing and AI layer to be installed at the same site in partnership with Nscale. These are distinct components of the same physical project and must not be added together as separate investments.

The reference to Brad Smith is based on statements reported by several media outlets, citing Jornal de Negócios. The figure of 12,600 Blackwell Ultra GPUs is based on the official announcement of the Start Campus-Nscale partnership.

The comparison with Ireland draws on EirGrid and CRU data and information reported by RTÉ on data centres’ impact on electricity demand and grid-connection policy. The legal analysis draws on the CLOUD Act, United States v. Microsoft Corp., the Schrems I and Schrems II judgments, the EU-US Data Privacy Framework, and public regulatory and legal sources on Portugal and Ireland.

Translation of the text in the author’s archive. Original publication: Portugal, a Feitoria Digital da América · LinkedIn · Essay · 19 March 2026.
Original bibliographic details (Portuguese)

Ensaio de Miguel Sousa Soares, publicado originalmente no LinkedIn a 19 de março de 2026. Esta página é o arquivo pessoal do autor.

Miguel Sousa Soares is an economist and analyst of geoeconomics and artificial intelligence, with more than twenty years’ experience in strategy and corporate finance. He writes for Visão, Jornal de Notícias, Expresso and Correio da Manhã. He comments on artificial intelligence on NOW on Fridays.