A €180,000 investment over 5 years goes to the board for approval. A €3,000 monthly expense flies under the radar. It is exactly the same thing financially — but not politically. Cloud transformed investment decisions into recurring subscriptions. The result: organisations spending 10 times more than they would in CapEx, never building an asset, without the figure ever being aggregated and presented as a strategic decision.

In France, the share of large enterprise IT budgets allocated to cloud rose from 18% in 2019 to 41% in 2023. The budget allocated to internal technical teams fell by 22% over the same period. The "OpEx" newspeak made these two trends invisible to each other.

"You only pay for what you consume" is true in the per-second billing sense. It is false in the economic sense: you consume forever, you never build an asset, and the accumulated value of your payments enriches the vendor without ever belonging to you. Over 5 years, "pay-as-you-go" produces €180,000 of spending on a GCP instance — and €0 of asset. The same budget in CapEx produces a server, internal competence, and residual value.

"The platform scales automatically" made the question "who manages capacity?" superfluous. The implicit answer was "nobody — the platform does it". In reality, automatic upward scaling is real. Cost reduction downward is blocked by noncancellable commits. And the platform that scales automatically still needs engineers to configure it, monitor it, and resolve incidents. But those engineers were eliminated because "the platform scales".

"Cloud-native" created a skills hierarchy: those that are "native" (proprietary, recent, vendor-certified) and those that are not (neutral, deep, not certifiable by a hyperscaler). A Linux administrator with 15 years of experience is "not cloud-native". A junior AWS Solutions Architect-certified for 6 months is.

In France, job postings for "Linux system administrator" fell 34% between 2019 and 2023. Postings for "Cloud Engineer AWS/GCP/Azure" rose 187% over the same period. The market applied the newspeak hierarchy. And the average salary of an AWS-certified DevOps is 23% higher than that of an equivalent-experience bare-metal SRE — the proprietary certification is worth more than the neutral skill.

"DevOps" in its cloud version merged two roles by eliminating one. The developer absorbed operations — but not bare-metal operations, cloud operations. The "cloud DevOps" knows how to deploy on EKS, configure a GitHub Actions pipeline, and monitor with CloudWatch. They do not know what is under EKS. They cannot get out alone. The "DevOps" merger did not create more complete engineers — it created more vendor-dependent engineers.

Cloud certification spending tripled between 2020 and 2023 — massively funding proprietary AWS/GCP/Azure certifications. Each AWS certification is an AWS service (documented in the Terms of Service, subject to a separate CPA agreement). It is terminable by AWS. It is not transferable to GCP or a sovereign managed service provider. Cloud "upskilling" is not skill-building — it is public investment in private dependency on an extraterritorial actor.

"Digital transformation" turned an operational question (how do we improve our systems?) into an identity question (are we a modern company?). A company not "undergoing digital transformation" is backward. This identity pressure short-circuited cost-benefit analyses that should have been conducted.

McKinsey (2022): 70% of "digital transformation" projects fail to meet their objectives. The average cost of a failed digital transformation for a large European company: €45 to €120 million. These failures are not primarily technical — they are framing failures. People answered "are we modern?" rather than "does this work better?"

If resilience is "by design" — built into the architecture — why test it? This logic produced documented but untested BCPs. The AWS us-east-1 incident of December 2021 took down "multi-AZ" services because the control plane was centralised. The "design" and reality had diverged. Nobody saw it because nobody tested — because it was "by design".

Gartner (2023): 35% of organisations that massively migrated to cloud are considering partial repatriation for cost and resilience reasons. This figure was 8% in 2020. "Resilience by design" produced organisations that discover their resilience limits during real incidents — not during tests.

If the "best" is defined within the AWS ecosystem, then the "best" is always AWS. "Best of breed" means choosing the best tool for each need — but when all tools are evaluated within a single vendor's catalogue, "best of breed" becomes "maximum lock-in". The question "is this the best tool among all available tools?" was replaced by "is this the best tool in this catalogue?" The newspeak narrowed the comparison perimeter without anyone noticing.

"Our data is in the global cloud" sounds reassuring — modern, distributed, resilient. "Our data is on servers in Virginia, Ireland and Singapore, under US jurisdiction, subject to the CLOUD Act, with a legally possible non-disclosure order if US authorities require it" sounds different. It is exactly the same thing. The "global infrastructure" newspeak dissociated the physical question (where are the servers?) from the legal question (which law applies?) to reassure on the first by avoiding the second.

"Your data stays in Europe" is true in the physical sense — the bytes are on disks in Ireland or Germany. It is false in the jurisdictional sense — GCP Terms Section 14.12 stipulates that the applicable law is that of the State of California and the competent courts are those of Santa Clara County. "Data residency" created an illusion of physical sovereignty that masks the absence of legal sovereignty. European regulators took years to articulate this distinction — during which sensitive public administrations migrated to cloud "in Europe" believing they were protected.

"AWS is GDPR compliant" means AWS has put in place mechanisms to help its clients be GDPR compliant. It does not mean that using AWS is GDPR compliant — a nuance the "compliance" newspeak erased. Compliance is a property of the complete system (infrastructure + usage + contract + jurisdiction), not of the vendor alone. By designating the vendor as "compliant", the newspeak shifted verification responsibility to the vendor — and removed the client's incentive to verify for themselves.

Hyperscalers buy massively from the REC market — Renewable Energy Certificates. A REC means that somewhere in the electrical grid, a quantity of energy equivalent to the declared consumption was produced from a renewable source. It does not mean that the servers hosting your data run on green energy. It means the vendor's carbon accounting is balanced on paper.

The AWS us-east-1 datacentre in Virginia consumes several gigawatts. Virginia's electrical grid is approximately 30% powered by coal and natural gas. RECs purchased by AWS do not change this energy mix. They comptably offset real fossil fuel consumption. This is creative carbon accounting — exactly as OpEx is creative financial accounting.

What the sustainability director validated: "By migrating to AWS, we reduce our carbon footprint through resource pooling." What they actually did: relocate energy consumption to US datacentres partially powered by fossil fuels, with virtual carbon offsets purchased on a certificate market.

A poorly designed Serverless architecture (Lambda cold-starting on each request, containers running on standby, monitoring jobs querying APIs every 30 seconds) consumes compute continuously — and therefore energy continuously. This inefficiency is invisible in the cloud console: it appears as a cost line, not as a carbon footprint. The bare-metal architect who dimensions their server once optimises it to last 5 years. A serverless architecture can consume 10 times more CPU cycles for the same workload — and the sustainability team will never know because nobody measures kWh per request.

The mechanism is structural and does not imply individual bad faith:

Step 1: the consulting firm recommends a "cloud-first" strategy based on market analyses (Gartner, IDC) it cites abundantly. It invoices the scoping study: €150,000 to €500,000 depending on organisation size.

Step 2: the hyperscaler certifies the firm as a "top-tier partner" (AWS Premier Partner, Google Cloud Partner, Microsoft Gold Partner). This certification requires a minimum number of cloud certifications in the firm's teams — thus funding the proprietary certifications documented in Section 2.

Step 3: the firm conducts the migration. It invoices integration man-days. Complexity is its ally — the more complex the architecture, the longer the migration, the more man-days accumulate.

Step 4: the migration produces gaps vs estimates. The firm invoices remediation man-days. Then training. Then optimisation (FinOps). The relationship is structurally perpetual.

What the loop produces: the firm has no financial interest in recommending bare-metal or a local managed service provider — there is no "partnership" with these actors, no certification, no complex migration man-days. Advice is structurally oriented toward complexity.

A server purchased for €15,000 is entered on the company's balance sheet as a fixed asset. It is depreciated over 3 to 5 years. Each year, a depreciation allowance (€3,000 to €5,000/year) is recorded as an expense — reducing taxable income. At the French corporation tax rate of 25%, this allowance generates a tax saving of €750 to €1,250/year.

Over 5 years, a €15,000 server generates €3,750 to €6,250 in cumulative tax savings. The real net cost of the server is not €15,000 — it is €15,000 minus the tax saving = approximately €8,750 to €11,250.

The €3,000/month GCP VM generates €36,000 of deductible expenses per year — but without creating an asset. Expenses are deductible in both cases. The difference: bare-metal creates an asset that appears in equity, increases the company's book value, can be pledged as loan collateral, can be resold, and can be presented to an investor as a sign of patrimonial solidity.

A company that migrates 100% of its infrastructure to cloud carries out a silent balance sheet destruction: fixed assets disappear, replaced by recurring expenses. In management accounting, this company has become structurally lighter — but also more fragile. It no longer has productive assets of its own. If the vendor raises prices, it has no patrimonial negotiating leverage. If it seeks acquisition, the buyer will find no infrastructure on the balance sheet — only subscription contracts and noncancellable commits.

"Cloud empties the treasury without creating patrimonial value" is not a metaphor. It is an accounting reality.

On 7 December 2021, AWS us-east-1 suffered a major incident affecting numerous services — Kinesis, Lambda, API Gateway, AppSync, EventBridge, and others. Thousands of SRE teams worldwide simultaneously received cascading alerts on their own services. The root cause: an anomaly in AWS's internal control plane.

What the SRE could do: look at their dashboards, see everything falling simultaneously, read the AWS status page ("We are investigating"), and wait. They had access to no internal AWS logs, no control plane metrics, no AWS runbook. The black box was closed. The SRE was on the front line of organisational stress — their phone was ringing, their manager was asking for an ETA — but they had zero technical leverage on the root cause.

What this reveals: "Zero-ops" also means zero visibility on incident root causes. The cloud-native SRE is an expert in consequences — they see the effects, they cannot diagnose causes when those causes are inside the vendor's infrastructure.

A bare-metal SRE facing a 3am incident has access to everything: kernel logs, hardware metrics, network traces, system calls. They can go down to the physical level if needed. Their diagnosis can reach "the NVMe disk in slot 3 has had elevated error rates for 48h" or "server 7's network card is dropping packets on the production VLAN".

A cloud-native SRE facing the same incident has access to their own application metrics — and stops at the surface of the vendor's API. They cannot see what happens under Lambda. They cannot diagnose a DynamoDB degradation. They cannot trace a request beyond CloudWatch logs. The lower layer is opaque by design. And this opacity is presented as an advantage ("managed service — you don't have to worry about it") until it falls — and you still don't have to worry about it because you cannot.

"Digital transformation" creates the identity urgency → "Cloud-native" validates the direction → "OpEx" makes the spend invisible → "Pay-as-you-go" normalises owning nothing → "Upskilling" funds proprietary certifications → "Cloud-native" devalues neutral skills → "Zero-ops" justifies eliminating engineers → "Data residency" reassures on sovereignty → "Compliance" de-responsibilises the client → "Resilience by design" removes the incentive to test → "Strategic partnership" lock-ins the organisation → "HR optimisation" eliminates the last engineers who could exit.

Each term reinforces the next. Exit becomes harder at every step. This is not a conspiracy — it is an ecosystem. And ecosystems do not need conspiracies to be effective.

Opération Dindon documented cognitive capture, the infrational loop, loss of sovereignty, amputation of technical knowledge. This study demonstrates that these phenomena are measurable: −34% Linux jobs, ×3 training spend on proprietary certifications, 70% digital transformation failure, 35% cloud repatriation being considered. Newspeak is not the sole cause — it is the vector. It made possible decisions that would have been impossible if they had been named correctly.