What the VM does: it runs. It consumes. It sends a bill on the 1st of the month. It does not take holidays. It does not ask for a raise. It does not get sick. It does not resolve incidents.

What the VM does not do: it does not configure itself. It does not detect the anomaly at 3am before it becomes an incident. It does not understand why the payment service has been slow for 20 minutes. It does not write the post-mortem. It does not train the junior. It does not read the Terms of Service before signing. It does not tell the CIO that the architecture is drifting toward irreversible lock-in.

What the SRE does: everything the VM cannot do — plus the VM itself.

Quarter 3. Management decides to optimise costs. Two lines are on the table:

Line A: "Cloud Infrastructure — 47 GCP instances — €142,000/month." → The CIO says: "We cannot touch that, it is our entire infrastructure."

Line B: "SRE Team — 4 people — €16,800/month gross with contributions." → The CHRO says: "We could reduce to 3 people and outsource the rest."

The €3,000/month VM is untouchable. The €2,800/month net SRE is optimisable. The machine is protected by its complexity. The human is exposed by their visibility.

Imagine a landlord who says: "I am renting you this flat for €3,000/month. You will never be able to buy it. I can revise the rent whenever I wish, with no guaranteed notice period. If you want to move out, you pay transfer fees for every piece of furniture you take. The contract is subject to California law. And if the US justice system wants to visit without telling you — the law sometimes prevents me from informing you, but not always."

No CFO in the world would accept these conditions for a flat. Every CFO in the world accepts them for cloud infrastructure. Because it is called "OpEx" and "flexibility" instead of what it actually is: an eternal rent with unfavourable contractual conditions.

A Dell PowerEdge R750 Refurbished Grade A — same specs, 3 years younger, tested and certified — costs €4,000 to €6,000. Depreciated over 3 years: €140 to €170/month. With opex (power, network, colocation): approximately €270/month. Over 3 years: €9,720.

The GCP VM over the same period: €108,000.

Ratio: 11×. For that price, you could hire a full-time senior SRE to manage your Refurbished infrastructure — and you would still have €87,000 of annual budget left for something else.

Cloud pulled off a remarkable accounting sleight of hand: it transformed a CapEx expense (server purchase) into an OpEx expense (monthly subscription). Finance directors love OpEx — it smooths spending, simplifies accounting, avoids balance sheet capitalisation. What finance directors did not see is that cloud is the most expensive OpEx in the history of computing.

A server purchased as CapEx appears on the balance sheet as an asset, then depreciates — reducing the tax base each year. A VM as OpEx is simply an expense — with no asset counterpart, no depreciation, no residual value. It is the OpEx that enriches the vendor and impoverishes the client, without the client ever accumulating anything.

When an organisation compares "cloud vs on-premise", it often compares the visible cost of cloud (the monthly bill) to the total cost of on-premise (purchase + staff + maintenance + datacentre). It forgets that cloud also has a staff cost — AWS-certified DevOps cost more than classic system administrators. It forgets egress fees, Premium support, migration consulting fees. And it always compares an already-operational cloud to a hypothetical on-premise yet to be built — never to an already-built on-premise.

"If I took the €3,000/month I pay for this GCP VM and invested it in a bare-metal server, what would happen?"

Months 1 to 5: buy the server (€15,000 — 5 instalments of €3,000). Same monthly budget.
Month 6: the server is paid off. Monthly cost drops to €150. Monthly saving: €2,850.
Over 5 years: total saving of €156,000. With this money, you can recruit an additional senior SRE, buy two Refurbished backup servers, fund three years of training for your team, and still have enough left for the coffee machine.

For €3,000/month at GCP: an n2-highmem-32 VM. You do not own it. It runs under California jurisdiction. When you want to leave, you pay egress fees. In 5 years you will have spent €180,000 and you will have nothing.

For €3,000/month for 5 months + €150/month thereafter: a Dell PowerEdge R750 bare-metal server you own, under your jurisdiction, in your colocation or with your sovereign managed service provider. After 5 months, your monthly cost drops to €150. In 5 years you will have spent €24,000 and you have an asset.

For €5,000/month (VM + €2,000 extra): the Refurbished Grade A server + a full-time senior SRE who manages it, documents it, trains your team and sleeps better at night. In 5 years you have sovereign infrastructure, a competent team, and an SRE who knows your infrastructure from the inside.

If Refurbished Grade A logic were applied to hiring: recruit an SRE with 10 years of experience (Refurbished — has already resolved hundreds of production incidents) rather than an AWS-certified junior (new, brilliant but without lived experience). The senior SRE costs €3,500 net instead of €2,800. Delta: €700/month.

What the senior SRE brings extra: they prevent on average 2 to 3 incidents per year that would each have cost 4 to 8 hours of downtime × business impact. They optimise the infrastructure to reduce costs. They document what they do. They train juniors. They sleep better at night — and so do you.

ROI of the €700/month delta: incalculable. But certainly positive.