Information checked: November 7, 2026. Announced demonstrations, adopted rules, legislative proposals and company-reported experiments are distinguished below.
The route announced for a November 7 demonstration in Helsinki runs past Google’s office to Finland’s Parliament. It connects the company proposing an investment with the institution responsible for public rules. Organisers are demanding a pause in data-centre projects and tighter regulation. Finnish organisers’ announcement
The European Stop Dirty Data Centres campaign called for action on November 6–7 and a moratorium on large developments. Its programme is a record of announced events, rather than proof that every event took place. Its environmental and social allegations also need to be tested against particular projects. Campaign and event listings
In Amsterdam on November 6, protesters entered a construction site intended for Microsoft, Reuters reported. Microsoft said it respected peaceful expression but rejected unlawful or unsafe action; site owner Pure Data Centres called for dialogue. The industry’s case is that these facilities support demand for AI and cloud services. Reuters’ report and responses
A dispute over a building becomes a dispute over government when it involves access to electricity networks, water, land and permits. The public question is how those resources are allocated, how costs are shared and what residents can know before decisions become difficult to reverse.
What the lobbying documents show
The EU’s 2023 Energy Efficiency Directive already provided for protecting trade secrets in data-centre reporting. Confidentiality did not originate with Microsoft’s later intervention. Directive 2023/1791, Article 12
Microsoft’s January 2024 comments on the EU’s proposed data-centre reporting rules asked for an explicit confidentiality provision. The company also proposed deleting submitted information within a year. These were requests made by an interested party, not a description of what lawmakers had already agreed. Microsoft’s submission, pages 2–3
The 2024 delegated regulation requires reporting to a European database. Its Article 5 protects individual-centre information as confidential and provides public access to aggregated information. Regulators can therefore hold site-level reporting data that residents cannot obtain from the public database. Similarity between a company’s request and a rule does not establish who caused the final wording. Regulation 2024/1364, Article 5
The policy has continued to develop. A Commission delegated text adopted on September 21, 2026 sets out public rating labels from August 15, 2027, with an exception to confidentiality for label information. That is narrower disclosure than publishing all underlying site data. This article has not verified the Official Journal publication required for that text’s entry into force; its adoption is not evidence that labels are already available for every centre. Commission’s 2026 rating text
A separate Commission record of an October 29, 2025 meeting documents Microsoft asking for more consistent permitting, a single point of contact and limits on procedure times. The record lists no conclusion. It establishes access to a policy discussion and the changes sought, rather than an agreement to grant them. Commission meeting record
Together, these documents make corporate influence concrete enough to examine. A large provider can ask officials to change the procedures governing the infrastructure it wants to build and the information it must disclose. The further claim that a company dictated a government’s decision requires evidence of the decision-making process, alternatives considered and reasons given.
When faster construction becomes state policy
The European Commission’s Cloud and AI Development Act proposal of June 3, 2026 would create data-centre acceleration zones and a single advisory contact. It proposes a 12-month permitting limit for projects within those zones, starting with a complete application. An area’s basic permit would require the necessary environmental assessments; installation-specific permits remain separate. As of this article’s check, the proposal remained before Parliament, awaiting a committee decision. It is not an enacted general exemption from environmental law. Commission proposal, Articles 10–13, Parliament’s procedure record
In the United States, President Donald Trump’s July 23, 2025 executive order directs federal agencies to accelerate qualifying data-centre infrastructure through permitting measures, financial support and access to federal land. It sets out the administration’s own national-security and industrial-policy objectives. A presidential instruction is not proof that an individual project has obtained every approval it needs. Executive Order 14318
Governments therefore enter these negotiations with ambitions of their own. They want computing capacity, investment and a stronger position in a technology they regard as strategically important. That can align their interests with large suppliers before any particular lobbying meeting takes place.
The imbalance appears when an investment announcement is treated as sufficient evidence of public benefit. Construction spending, permanent employment, tax receipts and the cost of reinforcing a grid answer different questions. A credible decision compares them, identifies who carries the risk and leaves room to refuse or redesign a project. The size of a proposed investment does not perform that assessment by itself.
Who receives the revenue?
American companies dominate a major part of the business. Synergy Research estimates Amazon, Microsoft and Google held 26, 19 and 16 percent respectively of worldwide cloud-infrastructure service revenues in the third quarter of 2026: 61 percent combined. This measures a market containing AI and other cloud services, not all global AI revenues. Synergy’s October 29 market estimate
That is evidence of supplier concentration. It is not a calculation of the income retained by the United States as a country. A host region’s construction work, employment, taxes and infrastructure costs still need their own accounts. Company revenue, profit and a public budget are different measures.
China’s response combines its own computing systems with a national network. Huawei’s March 2026 presentation describes Atlas 950 SuperPoD and an interconnect intended to make thousands of computing nodes work as one system, alongside its software ecosystem. These are the supplier’s specifications and claims; a maximum supported configuration is not proof that every installation has reached it. Huawei’s product announcement
China’s official account of the East Data, West Computing programme describes linking eastern demand to western computing hubs with wind and solar resources. It also acknowledges that network development is struggling to keep up. The response includes building centres, strengthening fibre links and coordinating workloads, rather than simply adding isolated processors. Official programme account
A September 2026 government account of the communications ministry’s five-year plan again prioritises an integrated nationwide computing network. That is a development objective, not evidence that capacity constraints have disappeared. Hardware supply, interconnection, power and usable software remain distinct tests of whether the announced capacity can deliver a service. China’s plan announcement
What has actually reached orbit
The search for another location is already producing experiments. Google’s October 30, 2026 update says its Project Suncatcher prototype ran a Gemma 4 model in orbit and produced a haiku. This is a company-reported computing result, not an independently audited account of a commercial data centre’s capacity, operating life or cost. Google’s October update
Google’s September technical explanation schedules a two-satellite optical-link test for 2027; its October update keeps that milestone ahead. Processing a request on one spacecraft and operating a closely connected computing cluster are different milestones. Google’s technical explanation
Starcloud says its Starcloud-1 spacecraft launched an NVIDIA H100 processor in November 2025 and trained a small language model in December. Its account demonstrates the kind of work the company reports having performed; it does not establish a large commercial service. The company’s Starcloud-2 page describes a further commercial mission whose full operation is targeted for 2027. Starcloud-1, Starcloud-2
China’s programme provides another distinction. In a February 2026 Xinhua report, Zhejiang Lab described tests involving its first 12 computing satellites, launched in May 2025. The applications discussed include Earth observation and astronomy. Processing data collected in space is a different service from replacing a terrestrial cloud campus. These are the laboratory’s reported results, not an independent performance audit. Xinhua’s report
These examples establish a change in the question. It is now possible to ask what orbital computers have done, alongside what their sponsors hope to build. They do not establish that the industry has demonstrated a replacement for large terrestrial cloud campuses.
How much solar power is available?
At Earth’s distance from the Sun, incoming solar radiation is about 1,361 watts per square metre. That is sunlight arriving at a surface facing the Sun, before conversion into electricity. NASA’s irradiance account
NASA’s spacecraft-power review describes multijunction solar cells with nominal efficiencies around 30 percent. Using 30 percent as an illustrative assumption, one megawatt of electrical output with perpendicular full illumination would require about 2,450 square metres of active cells: 1,000,000 divided by (1,361 × 0.30). NASA’s power review
That calculation answers a narrow question. It is an ideal collection area in sunlight, before gaps, structures, ageing, conversion losses and additional loads. It is not a satellite’s overall size or a design for a continuous one-megawatt computing service. Communications, control, thermal equipment and battery charging also require power.
Choosing an orbit with long illumination helps, but “Sun-synchronous” is not a universal guarantee of uninterrupted sunlight. DLR’s operational paper on TerraSAR-X describes seasonal eclipses on its dawn–dusk orbit. A proposed data centre needs its own orbital and seasonal analysis. If service is to continue when production stops, storage must bridge the gap: a one-megawatt load uses one megawatt-hour over an hour. DLR’s battery-operations paper
Solar power is therefore technically credible for orbital computing with existing cell technology. Whether it is sufficient for a particular service depends on the complete power and storage budget. A claim of commercial competitiveness needs much more than a panel’s peak output.
Heat and data still have to go somewhere
A computer in orbit cannot cool itself by blowing outside air over a heat exchanger. In a vacuum, heat must be carried to surfaces that radiate it away. NASA’s thermal-control guidance also accounts for sunlight, reflected solar energy and Earth’s infrared radiation heating the spacecraft. The environment is a heat balance, not a free refrigerator. NASA’s thermal-control review
Almost all electricity used by computing equipment eventually becomes heat. A megawatt-scale computing load therefore requires heat rejection of a similar order, in addition to other thermal loads. Radiator area is separate from solar-cell area. Hardware temperature limits, radiator orientation, pumps, structures and mass all constrain the system.
Radiation is another operating condition. NASA distinguishes accumulated damage from effects caused by individual particles; the appropriate protection depends on the hardware, orbit and intended life. A successful demonstration is valuable evidence, while long-duration reliability remains a separate test. NASA’s avionics review
Fast optical communications have also been demonstrated in space. NASA’s communications review discusses both high data rates and requirements involving accurate pointing, ground stations, weather and contact opportunities. A peak link rate is not the same as a continuous service. Processing observations where they are collected can reduce what must be sent down; serving users on Earth requires a network matched to their workloads. NASA’s communications review
The useful comparison is consequently a whole service delivered over a specified lifetime. It must include launch and deployment, power, cooling, communications, failures and replacement. Claims that an orbital centre will be cheaper or environmentally preferable remain claims until the assumptions and results can be examined.
The decisions still belong in public
An orbital processor may eventually avoid some demands on a local electricity network or water supply. It brings other decisions about launches, orbital safety, ground infrastructure and control of the resulting service. Changing the location does not eliminate the need to account for the system’s costs and risks.
On Earth, the immediate test is whether permit decisions make the important trade-offs visible: expected and actual electricity demand, grid access and reinforcement costs, water and cooling arrangements, credible local benefits and enforceable operating conditions. Commercial confidentiality needs a reasoned boundary if the public is to assess the burden placed on shared systems.
The protests have brought those choices into the street. The documents show companies seeking to shape the rules and governments actively pursuing more computing capacity. The central issue is how public institutions judge those requests, explain their decisions and retain the practical ability to change course.
Sources and method
This is an analysis of public documents, reporting and technical literature. Northern Casebook has not attended these demonstrations, interviewed the parties, reproduced orbital experiments or independently audited company performance. Company announcements establish what the companies report; proposals and adopted texts are identified by their actual stage. The solar-area example is our own idealised calculation, not a spacecraft design or a measured data-centre result.