By Micky Irons, founder and CEO of Mickai.
Every finance director who has signed off on a large cloud AI bill knows the quiet dread of the second invoice. The first month is a proof of concept, priced like a courtesy. The tenth month, once the models are woven into contracts, claims, code and customer service, arrives with a number nobody forecast and nobody can fully explain. The workload did not change. The meter did.
This is the central deception of the per-token economy. It is sold as efficiency, pay only for what you use, but it is really a lease on a capability you can never own. At Mickai we take a different view. We believe the honest question is not what artificial intelligence costs this month, but what it costs to own outright, across five years, with no landlord standing between you and your own data. That is the true total cost of ownership, and it is time somebody stated it plainly.
The meter you cannot read
A per-token contract is a variable-rate mortgage on your own intelligence. The price per unit looks trivial in isolation, a fraction of a penny, but usage is not something you control once a model is embedded in daily operations. Every prompt a colleague writes, every automated retry, every retrieval step that stuffs a document into context, every agent that calls another agent, all of it turns the meter. Success becomes the thing that bankrupts you: the more valuable the system, the more it is used, and the more it is used, the larger the bill.
Worse, the unit itself is opaque. Token counts depend on tokenisation you do not control, on hidden system prompts you cannot see, on model versions that change under you without notice. You are being charged by a scale whose calibration lives on someone else's premises. No regulated business would accept a utility contract written this way, yet this is precisely how most organisations now buy the capability they call strategic.
Prometheus gave fire once and for all, not by the spark; a capability you own is a gift, not a lease.
Capital you own versus rent you never stop paying
There is an older and more honest way to acquire a strategic capability. You buy the asset, you install it on hardware you own, and you depreciate it over its useful life. The cost is large and visible at the front, and then it falls. This is how organisations have always treated the things that matter: the presses, the reactors, the ledgers, the vaults. You do not rent the vault by the coin.
Mickai is a Sovereign Intelligence Operating System, a SIOS, built to be owned in exactly this sense. It runs on hardware the customer owns, air-gapped or on-premise, with zero data egress. The brains, the subsystems and the studios sit inside your boundary, not behind someone else's login. The economics invert. Instead of a meter that rises forever with your success, you hold a depreciating capital asset whose marginal cost of another million tokens is the electricity to run it. Ownership converts an unbounded operating liability into a bounded, forecastable line on the balance sheet.
The costs the invoice hides
An honest total cost of ownership has to count what the per-token invoice quietly omits. The first hidden cost is egress and exposure. When your prompts leave your building, they carry your commercial secrets, your patient records, your legal privilege and your trading positions into an environment you do not govern. The price of that is not on any bill until the breach, the regulator or the discovery request arrives, and then it is the only number that matters.
Atlas bears a weight that never lightens; a rented meter that rises with your success is the same burden by another name.
The second hidden cost is compliance drift. Under the EU AI Act, the Digital Operational Resilience Act (DORA), the General Data Protection Regulation (GDPR) and frameworks such as ISO 42001 and the NIST AI Risk Management Framework, you must be able to prove what a system did, on what data, under whose authority. A rented black box cannot furnish that proof on your terms. Every audit you cannot answer cleanly is a cost, paid in remediation, in delay, and in trust.
The third hidden cost is lock-in. A capability you cannot move is a capability you do not price. When the rates rise, and rented rates always rise once you depend on them, you have no exit, because the exit was never yours to design.
What sovereignty actually buys
Sovereignty is not a slogan, it is a set of engineering guarantees you can hold in your hand. In Mickai, every action a brain takes is wrapped in an Operation Attestation Record, an OAR, which is signed before the action executes, not after it is noticed. Those records are chained with SHA-3-512 hashes into a tamper-evident ledger and signed with post-quantum signatures under the FIPS 204 ML-DSA-65 standard, so the audit trail survives even the arrival of quantum computers that would break older cryptography.
Themis weighs each act on its own merit; a sovereign ledger lets every action be weighed and proven on your own terms.
That machinery pays for itself the moment a regulator, an auditor or a court asks the awful question: prove it. Verification runs offline, inside your walls, with no call home. High-stakes actions require multi-brain and voice-biometric approval, and any brain can be revoked the instant it misbehaves. These are not features you can bolt onto a rented endpoint. They are the reason ownership is cheaper than it looks, because the cost of not being able to prove your own actions is the cost that ends careers and companies.
The five-year picture, told honestly
We will not pretend sovereignty is free. Owning your intelligence means capital outlay, hardware, deployment and the discipline of running a system rather than merely calling one. That cost is real and we state it plainly, because the whole point is honesty about total cost. What we will say is that this cost is bounded, front-loaded and falling, where the rented alternative is unbounded, back-loaded and rising.
Across a five-year horizon the two curves cross, and they cross sooner than most boards expect, because the rented curve bends upward exactly when adoption succeeds. The rented model is cheapest in the pilot that nobody depends on and most expensive in the production that everybody does. Ownership is the reverse. The right question for any board is therefore not which is cheaper to try, but which you would rather be holding in year five, when the capability is load-bearing and the meter, if you are still renting, has never once ticked down.
Allies, not adversaries, at a different layer
None of this is a quarrel with the public cloud. The hyperscalers built extraordinary infrastructure and they remain allies for everything that belongs in the open, the elastic, the shared. Mickai sits at a different layer, the regulated boundary the public cloud cannot cross on the customer's own terms. There will always be workloads that are fine to rent. The argument here is only that your most sensitive, most strategic and most heavily used intelligence should not be one of them, because for that class of work, renting forever is the most expensive option there is.
Hephaestus forged what the gods relied upon in his own workshop; the capability that matters most is the one you build and hold within your walls.
The bottom line
The true cost of owning your artificial intelligence is a capital number you can see, forecast and control. The cost of renting it is a per-token number you cannot read, cannot cap and cannot escape, and it grows with your own success. Mickai exists so that regulated organisations can own the capability outright, prove every action cryptographically, and keep their data inside their own walls. Buy the vault. Stop renting it by the coin.
Written by Micky Irons. Originally published at https://mickai.co.uk/articles/the-true-cost-of-owning-your-ai. More from Micky Irons and Mickai at mickai.co.uk.
Top comments (0)