IX · Mēden agan · Nothing in excess
Infrastructure should fit the work.
AI infrastructure should be workload-specific.
Cloud, private, or hybrid, sized from measured workloads.
In practice
What it means for your system.
Cloud, private, or hybrid: the answer depends on the workload. We size infrastructure from measured demand and show its cost over time before you commit.
- Deployment follows the workload’s volume, sensitivity, and cost.
- Scenario models show cost over time, under stated assumptions.
- Hardware is sized from measured workloads, and you own what you buy.
The word for it
Harmonia
ἁρμονία
Harmonia is Greek for “a fitting together, first a carpenter’s joint.” Infrastructure is chosen to fit the work it carries.
Why it matters
Infrastructure is a measurement before it is a purchase.
Where AI runs sets much of its long-term cost, and it decides who can see your data. Those are commitments that outlast any single project.
Oversizing locks capital into capacity the work never uses. Undersizing slows the work and pushes sensitive tasks toward options nobody chose on purpose.
So the decision starts from the workload: its volume, its peaks, its sensitivity, and its growth. The infrastructure is then chosen to fit, with the cost over time shown before you commit.
Read it accurately
Scope of the rule.
- Capacity follows measured demand.
- Capacity is sized from measured demand, with growth stated as an assumption. More capacity follows when the measurements show the need.
- Each workload is placed on its own merits.
- Each workload is placed on its own merits. A hybrid setup can keep sensitive work private while other work runs in the cloud.
- Sensitivity and privacy weigh alongside cost.
- Data sensitivity, privacy obligations, and who can see the work weigh alongside cost. The scenario model shows the trade-offs in one place.
How it is measured
Every rule is something you can check.
Scenario models of cost over time; hardware sized from measured workloads.
In an engagement
Where the rule is applied.
The rule is checked at each stage of the work, from the first design to the system in operation.
In the design
Each workload is measured for volume, peaks, sensitivity, and growth. Self-Sovereign AI work turns those measurements into deployment options.
In the build
The system is designed so a workload can move between cloud and private deployment. Capacity is set from measured demand.
At acceptance
Running costs and capacity are checked against the scenario model’s assumptions. Differences are recorded, and the model is updated.
In operation
Measured demand is reviewed against capacity over time. Any move or expansion goes to your owner with its cost stated first.
Check your own system
Five questions to ask this week.
Each one has a yes or no answer. A no marks where to start.
- Do you know how much AI work your business runs, and when it peaks?
- Is each AI workload classified by the sensitivity of the data it touches?
- Can you see the expected cost of your current setup over time?
- Are the assumptions behind that cost written down where your team can review them?
- Was your current capacity chosen from measured demand?
Delivered through
The services that apply this rule.
Mēden agan
Nothing in excess.
No more model, spend, or autonomy than the work requires.
Apply the rule
Check it against your own system.
Questions
Do you recommend private hardware for every client?
No. Many workloads fit the cloud well. Private or hybrid deployment is recommended only where the measured workload, sensitivity, or cost calls for it.
Can we start in the cloud and move later?
Yes. Systems are designed so a workload can move when the measurements support it. Your owner approves any move, with its cost stated first.
What goes into a scenario model?
Expected volume, growth, model choice, and support needs, each stated as an assumption. Changing one shows how the cost over time moves.
How do we check this rule on our own system?
Ask for the scenario model of cost over time behind your current setup. Then confirm any hardware was sized from measured workloads.
Who owns hardware bought for a private deployment?
You do. Hardware is itemized separately, and your business buys and owns it.
How is a workload’s sensitivity assessed?
By the data it touches and the obligations attached to that data. Data Protection & Readiness maps those flows, and the result guides where the workload runs.