Product1 distinct publisher3 min readPublished
The UK's 2024 Microsoft agreement runs five years, and Germany buys AWS through Bechtle. Europe's AI plan arrived in April 2025, roughly a year into terms it would have to unwind.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
Contract length is the number that decides this, and it is the number least often quoted. The UK signature landed in 2024 on a five-year term, which puts the earliest natural exit at 2029 [2][12]. The European Commission published its AI Continent Action Plan in April 2025, about a year into that term [6][13]. The policy is younger than the paperwork it would have to undo.
The plans also aim at the wrong half of the stack. AI factories and gigafactories put public supercomputers and datasets behind model training [6], but Rest of World's observation is that European startups trained on EU machines are still likely to deploy through Big Tech clouds [7]. Training is a grant. Deployment is a contract.
The emerging-economy versions have the same shape under different names. IndiaAI and Brazil's plan both lean harder on the public sector, and both still treat cloud as infrastructure as a service rather than decoupling from it [8]. The Indian advisory group that proposed the FutureLab AI Compute Infrastructure included Intel and Oracle employees, and if the recommendation is adopted the government builds it with Intel and Oracle, on top of edtech and healthtech services already running on Oracle infrastructure [9]. In Brazil, state-owned telecoms and data companies want an intermediate compute layer for the public sector, with US vendors continuing to supply the black-boxed core underneath [10]. Rest of World's conclusion on all three is that none has meaningfully expanded sovereignty against the cloud giants, and none touches intellectual monopolisation [11].
The 2024 outage is worth re-reading as a sizing exercise. A cybersecurity partner's error took out 8.5 million Windows devices, which the report puts at under 1% of the installed base [1]. Run that backwards and the base is north of 850 million machines [14]. That fraction closed airports, delayed airlines and reached US emergency services, banking, stock markets and hospitals [c1b]. It was the concentration lesson of the same year the UK committed to five more.
Two procurement details deserve more attention than the strategy documents get. Germany's arrangement runs through AWS and its German partner Bechtle AG [3], and New Zealand's Department of Internal Affairs holds an all-of-government deal with Google Cloud [4]. A reseller in the middle and a whole-of-government scope both turn substitution into a negotiation rather than a switch. Meanwhile the source of truth on what these deals actually contain is unreliable: Project Nimbus, worth $1.2 billion across Google and Amazon, became public after employees at both firms described what it was built for [5]. Governments that outsource control technologies to acquire them end up regulating vendors they depend on [15], and the dependency has a renewal date long before it has a policy answer.
Ranked by verification strength, evidence, and original report placement.
In mid-2024 a cybersecurity partner's mistake led to a global Microsoft Windows outage that affected 8.5 million Windows devices, which the report notes is less than 1% of all Windows machines.
The 2024 Windows outage shut airports, caused widespread airline delays, stopped media outlets broadcasting live, and affected US emergency services along with banking, stock markets, petrol stations, retail, hospitals and telecoms.
In 2024 the UK signed a five-year agreement with Microsoft for its digital transformation.
The German government has a similar agreement with AWS and its German partner Bechtle AG.
In New Zealand, the Department of Internal Affairs holds an all-of-government agreement with Google Cloud.
AWS is used by public bodies worldwide including the Australian Bureau of Statistics and Argentina's Provincia Bank, the public bank of the Buenos Aires province.
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Concrete procurement anchors, single-source analysis
The verifiable spine — the UK five-year Microsoft term, Germany's AWS purchase through Bechtle, New Zealand's Google Cloud all-of-government deal, AWS public-sector users, the $1.2bn Nimbus contract, the April 2025 EU plan, the 8.5 million-device outage — is specific and internally consistent, and the derived timeline (expiry no earlier than 2029) follows arithmetically. But the cluster has one publisher, cites no contract documents, plan texts or vendor responses, and its load-bearing conclusions about intellectual monopolization and failed sovereignty are asserted analysis.
Government cloud dependence documented across five continents
The phenomenon the story describes — public-sector reliance on hyperscaler cloud — is documented through concrete, named deployments spanning the UK, Germany, New Zealand, Australia, Argentina and Israel, plus an incident that demonstrated cross-sector operational exposure. What is not measured is uptake of the countervailing programs: no usage figures exist for EU AI factories, IndiaAI's FutureLab (only recommended, not adopted) or Brazil's intermediate layer.
Deflationary thesis outruns its own evidence
The cluster is a debunking piece rather than a promotional one, so it does not inflate a technology. Its overstatement runs the other way: the categorical judgments that decoupling is effectively impossible and that no initiative has meaningfully expanded sovereignty exceed what one outlet's unsourced assessment can carry, while the well-evidenced procurement and timeline facts are stated with appropriate restraint. Net: mildly overstated conclusions on a solid factual base.
Vendor-authored policy advice named; publisher stance explicit
The cluster surfaces a concrete incentive structure: the Indian advisory group that recommended the FutureLab public-private partnership included Intel and Oracle employees, whose firms would be the counterparties, against a backdrop of existing Oracle dependence in edtech and healthtech. It also documents governments' regulatory bind — they must police vendors they depend on. Offsetting this, the publisher writes from a declared anti-intellectual-monopoly position, and no vendor or government was given a right of reply.
Facts trustworthy, verdicts unverified
Confidence is moderate: the named contracts, dates and the outage figures are checkable and mutually reinforcing, and the derived timeline logic is sound. It is held down by single-publisher sourcing, the absence of primary documents or vendor comment, and the fact that the cluster's most consequential statements are interpretive judgments marked insufficient rather than measured findings.
invest
Callosum's $100m seed is a 10x on February, and a UK state fund's first cheque3 distinct publishers
product
Rillet's $100M reads as proof mid-market ERP is rip-and-replace, mostly at the cheap end1 distinct publisher
product
Made in EU starts at 5% concrete and says nothing about the software1 distinct publisher
product
Cisco and Nvidia go looking for the other third of AI spending1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 26, 2026