Text decision tree

Adapted from Chapter 88 of Decision Trees: Frameworks for Technology Choices. Sources reviewed 2026-11-07.

START: A sourcing choice is conditional. Benchmark the actual workload and compare full cost.
Q1. Physical control required by policy?
  Yes -> Evaluate controlled infrastructure. Owned DGX or an approved location. [Q2]
  No -> Otherwise continue to the demand question. [Q2]
Q2. Demand is uncertain or short lived?
  Yes -> Start with flexible GPU capacity. GPU clouds offer compute capacity. [Q3]
  No -> For predictable demand, consider who will operate the infrastructure. [Q3]
Q3. Team can operate the GPU systems?
  Yes -> Compare owned and reserved capacity. Operator inputs: staff, energy, downtime. [Q4]
  No -> Otherwise evaluate a managed service and verify its actual operating responsibilities. [Q4]
Q4. Final check: workload fit and full cost
  Yes -> Pilot the shortlisted capacity. Verify fabric, data location and recovery. [END]
END: Validate GPU availability, interconnect, data movement and recovery. Keep a tested capacity fallback.

AI-generated chapter illustration

Conceptual illustration for Choose how to source GPU compute

The illustration groups named sourcing categories in one row. The graph applies control, demand and operating gates in order, then validates the shortlist. A logo or category label in the illustration does not establish purchasable capacity or contract terms.

The illustration was generated in BookWriter. Its visual arrangement is interpretive; use the explicit nodes and routes in the graph and text tree for the decision.

Primary references