Compute, cloud and AI capacity: start with the workload

By 1Gov advisory team · Published

The right compute arrangement depends on the applications and services an LGU must run. Begin with a workload inventory rather than a server brand, cloud platform or AI hardware specification.

Measure what the workload needs

Record users, transaction volumes, storage growth, peak periods, dependencies and acceptable downtime. Separate applications that must run continuously from batch jobs and short experiments.

For AI workloads, distinguish model training from inference. Define a specific use case, expected quality, response time and usage volume before reserving accelerator capacity. AI infrastructure is useful only when it supports a justified service need.

Compare owned, hosted and cloud options

Owned servers bring operating and replacement responsibilities. Hosted infrastructure can move facility responsibilities to a provider. Cloud services can offer flexible capacity, but the LGU must still define configuration, monitoring, security and recovery responsibilities.

A hybrid option may fit different workloads. Compare options against application compatibility, connectivity, staff capability, performance and data requirements rather than assuming one platform suits everything.

Make recurring costs controllable

Price the whole workload: compute, storage, backup, networking, data transfer, licences and support. Model peak usage as well as the baseline. Include migration and the cost of leaving.

  • Budgets, alerts, spending caps where supported and approval for expansion
  • Capacity reservations and minimum commitments versus variable usage
  • Responsibility for shutting down unused resources
  • Recovery testing, backup retention and restoration costs
  • Portable data exports and application dependencies at exit

Keep governance with the LGU

Classify data and review applicable privacy, security and data-location requirements before choosing the arrangement. Document access permissions, encryption, logging, incident response and provider responsibilities. Outsourcing infrastructure does not remove the LGU’s governance responsibilities.

For personal data, review the current NPC rules, including Circular 2023-06; the NPC states that it repealed Circular 16-01. Keep human review and defined data-handling rules in any AI pilot. A subscription alone does not establish approval to use sensitive data.

Discuss your ICT project