AIGP Study Guide
Module 7: Governing AI Deployment · BoK IV.A

Where the model lives - three environments

The deployment environment depends on budget, IT expertise, model purpose and data type. Cloud scales, on-prem controls, edge localises - each buys one thing and costs another. Models are then packaged via containerisation and made accessible by exposing the model through REST APIs or embedding.

The right option depends on budget, IT expertise and resources, model purpose and computational needs, and the type of data processed.

Mnemonic

Cloud scales · On-prem controls · Edge localises - the trade-off triangle the exam tests; each environment buys one thing and costs another.

The three deployment environments
EnvironmentAdvantagesDisadvantages
Cloud-based deployment → third-party provider hosts and handles infrastructureEasy to scale up or down · reduces hardware investmentPotential latency issues · security risks when a third party handles the data
On-premise deployment → the organisation's own servers and hardwareGreater control over infrastructure → key for sensitive data or regulated sectorsGreater upfront hardware investment
Edge deployment → hosted on edge devices like smartphonesDecreased latency, greater data privacyLimited by the device's hardware → limited computational power
  • Containerisation → packaging the model and everything it needs to run, into a self-contained unit, reducing compatibility issues and easing deployment across environments.
  • Exposing the model → letting systems and applications interact with it, commonly via REST APIs or embedding into an application.

Key terms - quick answers

What is “Cloud-based deployment”?
Hosting where a third-party provider handles the infrastructure; easy to scale but adds latency and third-party data risk.
What is “On-premise deployment”?
Hosting on the organisation's own servers and hardware; greater control but greater upfront cost.
What is “Edge deployment”?
Hosting on edge devices like smartphones; low latency and privacy but limited compute.
What is “Containerisation”?
Packaging a model and all its dependencies into a self-contained unit to ease deployment and reduce compatibility issues.