The sovereign AI that pays part of its own power bill
Every Mickai deployment includes the Fleet Energy and Efficiency Advisor, a subsystem that measures your compute estate and, every month, shows exactly where to cut electricity cost, with the kilowatt hours, the pounds, the payback and the carbon sealed to an audit record.
On premise AI raises one question, and we built the answer in
When a firm moves its AI off the metered cloud and onto hardware it owns, the biggest saving is obvious. The cloud AI bill stops. The next question a finance director asks is just as fair. Will all of this compute run up our electricity bill? We built the answer directly into the Sovereign Intelligence Operating System. It is called the Fleet Energy and Efficiency Advisor, and it is included in every deployment. It watches the compute estate the customer owns and, every month, tells them exactly which hardware and power changes will cut their electricity cost, with the numbers spelled out and sealed.
The Advisor is a subsystem of Mickai, not a separate tool to buy or another cloud to trust. When a client installs Mickai in their server room and rolls it out to their workstations, every machine already carries its own identity and department. The Advisor turns that fleet into something the system can measure and optimise for energy, cost and carbon, entirely on hardware the customer owns, with nothing leaving the building.
What it measures, and what it does about it
The Advisor works in four movements. First it discovers and measures. It scans every workstation and server for its processors, memory and power supply, then reads real power draw per machine, per department and per site on a live, on premise dashboard. Graphics processor wattage is measured directly. Whole system figures are estimated, and are always described as estimates.
Then it finds the waste. It flags machines drawing power while doing nothing, which is the single largest source of waste in an unmanaged fleet. It flags processors doing the same work at far higher wattage than a modern part, and low efficiency power supplies, with the upgrade payback calculated. It finds heavy machines running light workloads that could be right sized or redeployed.
Then it recommends the fixes. Sleep scheduling and power capping that do not disrupt work. Moving idle workstation inference onto a central batched tier. Choosing the best split between graphics and general processing per machine, because Mickai lets the operator choose. Shifting heavy jobs into cheaper off peak tariff windows. Planning hardware refreshes by return on investment rather than by habit.
Finally it reports. Every month the customer receives kilowatt hours saved, cost saved at their own tariff, payback period and carbon avoided, per department and fleet wide, with a Scope 2 energy and carbon figure suitable for sustainability reporting. Every report is sealed to a post quantum Open Audit Record, so finance, sustainability and audit teams can trust and present the numbers.
How it turns into money
The Advisor attacks four levers, in rough order of size.
- Idle elimination. Workstations left on around the clock waste a large share of their annual energy doing nothing. Sleep scheduling and power capping remove most of it, and this is usually the biggest single lever.
- Consolidation. Idle local graphics processors are retired onto a central batched tier that runs far more efficiently per unit of work.
- Hardware efficiency. Replacing the least efficient processors and power supplies cuts the energy cost of the work that remains.
- Right sizing and scheduling. Matching machines to workloads and moving heavy jobs to off peak tariffs trims the rest.
Each of these cascades into lower cooling load in the server room, and each is quantified and sealed so the customer can act with confidence.
What it can save, on a conservative five year model
The figures that follow are a modelled illustration, not a guarantee. Real savings depend on the customer's hardware, usage pattern and tariff. We assume one AI workstation draws about 1,550 kilowatt hours a year in an unmanaged fleet, roughly 450 watts while working and about 100 watts left idle around the clock, plus a modest 15 percent facility overhead for cooling and distribution. We price electricity at 25 pence per kilowatt hour in year one, rising 5 percent a year in line with commercial energy inflation, and we let the saving ramp from 20 percent of facility energy in year one to about a third by year three as the customer acts on more recommendations. Grid carbon is taken at 0.20 kilograms of CO2 equivalent per kilowatt hour.
On those assumptions, a single workstation saves about 89 pounds in year one and about 190 pounds by year five, roughly 747 pounds and 2,673 kilowatt hours over the five years. Because the saving scales per seat, the fleet numbers are substantial.
- A 500 workstation organisation could save on the order of 45,000 pounds in year one, rising to around 95,000 pounds by year five, roughly 375,000 pounds over five years, and cut about 267 tonnes of CO2 equivalent.
- A 1,000 workstation organisation could save on the order of 89,000 pounds in year one, rising to around 190,000 pounds by year five, roughly 750,000 pounds over five years, and cut about 535 tonnes of CO2 equivalent.
These are deliberately conservative. At a higher tariff of 30 pence per kilowatt hour, or with an older always on fleet, the five year figure for a thousand workstations moves comfortably past 900,000 pounds. For a fleet that already runs many thin clients into a central tier, the per seat figure is lower, but the central tier is exactly where the Advisor's consolidation and batching gains are largest, so the fleet level saving holds up.
A bonus, not the pitch
The reason a firm chooses Mickai is sovereignty and the end of the metered cloud AI bill. The Efficiency Advisor sits on top of that and does three things for the buyer. It gives money back, as a rebate against the cost of ownership, so the system helps pay for itself. It de risks the board conversation, because the honest answer to will this run up our power bill is a monthly, audit sealed report that measures the bill and drives it down. And it hands sustainability a defensible Scope 2 reduction with no new tool to buy.
No cloud AI service can offer this, because none of them run on the customer's hardware and none of them can see or optimise the customer's estate. Only a system that lives inside the building can measure it, and only a system the customer owns can be trusted to report on it honestly.
“Sovereign by architecture, efficient by design. A compute estate that optimises itself, and hands you the audit sealed proof every month.”
See it on your own estate
The Fleet Energy and Efficiency Advisor is part of the operations capability inside every Mickai deployment. To see how it would map your own fleet and model your own saving, read the capability in full at mickai.co.uk/capabilities/operations.