WAR MACHINE
Complete Anatomy · English Version
based on nineteen legal analyses of AWS · GCP · Azure terms
addressed to the CMA · FCA · ICO · European regulators
Public document · CC BY-NC-SA 4.0 · Not legal advice
These contracts are not poorly drafted. They are perfectly drafted for a precise objective that is not the customer's: to maximise exit costs while minimising entry costs. This is not a side effect — it is the primary function. Every clause analysed becomes immediately explicable when read with this objective in mind rather than through the lens of a good-faith buyer.
Reserved Instances cost 30–40% less than on-demand. The saving is real. It comes at the price of non-cancellability. The customer trades freedom for a price reduction — a rational short-term choice that becomes a medium-term trap. UCTA 1977 s. 3: unreasonable exclusion of liability = unenforceable. CRA 2015 Sch. 2: terms creating significant imbalance = unfair. EU Data Act art. 25: contractual obstacles to switching cloud providers = unenforceable.
The threat of suspension is sufficient to deter any challenge. Azure MOSA Section 3.c provides 30 days' notice for non-payment — the one point where Azure outperforms AWS on this criterion. CA 1998 s. 18 · CRA 2015 Sch. 2 para. 1(b): right to suspend must not be exercised in a disproportionate manner.
Entry is free, exit is priced. The machine deliberately separates contractual from pricing to maximise opacity at signature. EU Data Act art. 25 · CA 1998 s. 18: obstacle to switching = potentially unenforceable.
Proprietary APIs: Lambda/Cloud Functions/Azure Functions = non-standardised. Migration requires full rewrite.
Managed services: BigQuery/Redshift/Azure Synapse = proprietary SQL dialects. 10 years of queries = months of migration.
Superlinear effect: each additional service increases the cost of exiting all others.
I — Scale: 65%+ global market → potential abuse of dominance. CA 1998 s. 18.
II — Information asymmetry: egress fees outside T&Cs · Directive 2018/1972 waiver in English only · constitutive, not accidental.
III — Positive law violated: UCTA 1977 · CRA 2015 · CA 1998 s.18 · EU Data Act art. 23–25.
IV — Deliberate design: drafted by top-tier legal teams · tested across jurisdictions · the waiver clause is not in English only by accident. CMA 2023 Study identified switching barriers — no enforcement followed.
An unfair clause is a problem for the civil courts. A system deliberately designed to create dependency, relying on unlawful clauses, at the scale of a 65% concentrated market — that is a problem for the competition regulator. The CMA holds the Competition Act 1998 s. 18. Its own 2023 Cloud Market Study identified switching barriers as a primary concern. No formal enforcement has followed. This document and the nineteen analyses underpinning it constitute the two faces of the same case: clause by clause, and as a coherent system. Both are necessary for regulators to act.