AI models need compute. Compute needs energy. For SMEs that is a cost and planning topic, not a buzzword.
Chapter 1: Energy and data centres
The International Energy Agency (IEA) describes that data-centre electricity demand is rising due to AI [1]. Cooling is part of the demand.
What that means for SMEs
If you use AI, you should know the energy and cost frame. That is a planning question, not a cause for alarm.
Chapter 2: Cybersecurity
Phishing, stolen credentials and persistent access are known attack patterns. Whether they were the cause in a concrete case must be checked.
Operational recommendations
Protect access with multi-factor authentication
Test backups, not just create them
Review permissions regularly
Document an incident response plan
Own data over blanket frequency claims
Which attack path occurs most often in your organisation is shown by your own logs and documented incidents. Review both regularly.
Chapter 3: Products and roadmaps
Product announcements and roadmaps change. Check the current vendor documentation before purchase decisions.
Chapter 4: Putting E-E-A-T in perspective
Google describes E-E-A-T as a quality frame, not as a single ranking factor [2]. Visibility comes from checkable, helpful content.
Worked example: separate energy from cost
Illustrative assumptions, not GlasBox measurements: a server needs an additional average 0.25 kW for a workload running 80 hours per month. That gives 0.25 kW × 80 h = 20 kWh. At an assumed electricity price of CHF 0.30/kWh, the electricity cost would be CHF 6. Idle consumption, cooling, hardware, maintenance and VAT are excluded.
With an external API, you usually see usage fees rather than directly attributable kWh. Token charges cannot be converted into energy without provider and measurement data. Record invoice cost, volume, answer quality and latency for an operational decision. Compare systems on the same task rather than nominal model size alone.
| Measure | Unit | Boundary |
|---|---|---|
| Additional average power | kW | Measure against a defined idle baseline. |
| Time at that load | Hours per month | Do not confuse with general server uptime. |
| Energy | kWh | Disclose the boundary, including cooling. |
| Operating cost | CHF per month | Separate electricity, infrastructure, API and maintenance. |
A security plan with testable outcomes
A small team can start with four acceptance checks: MFA sign-in, restoration of a test file, revocation of a departing user’s account and a simulated incident report. Record dated results. An existing backup without a restore test does not establish recoverability.
| Control | Concrete test | Evidence |
|---|---|---|
| Access | Test account loses the intended permissions | Time and affected systems. |
| Backup | Restore a test file in an isolated environment | Readability and measured restore time. |
| Incident | Who revokes keys and informs the owner? | Tested contact and escalation path. |
Conclusion
AI, energy and cybersecurity are planning topics. GlasBox reviews setups project by project.
Kacper Ruta · owner of Ruta Group · GLASBOX Studio
Sources
IEA — energy demand from AI and data centres
E-E-A-T — Google Search Central
Review the AI setup?
GlasBox reviews your AI setup: data flow, security and operational requirements — including energy costs as a planning factor. No prices beyond the public list.
Request Consultation