01 Assess · Compute Blueprint
Make the infrastructure decision before making the investment.
Peregrine analyses your workloads, performance requirements, utilisation forecasts and infrastructure constraints to design an appropriate AI compute architecture — before capital or cloud commitments are made.
The problem
Why organisations struggle to size AI infrastructure.
AI infrastructure decisions are made under uncertainty: workloads change, accelerator supply shifts and cloud bills are hard to predict. The Compute Blueprint is designed to replace assumptions with analysis.
The outcome is a documented recommendation your organisation owns — whether you proceed with Peregrine, another provider, or internally.
GPU AVAILABILITY
Accelerator supply varies by architecture and timeframe. Decisions made without a clear view of what can be sourced, and when, often lead to compromise.
UTILISATION UNCERTAINTY
Teams rarely know how heavily infrastructure will be used before it exists. Over-provisioning strands capital; under-provisioning stalls projects.
CLOUD COST UNPREDICTABILITY
Consumption-based GPU pricing is flexible, but sustained workloads can make monthly costs difficult to forecast.
TRAINING VS INFERENCE MISMATCH
Training favours tightly coupled, high-bandwidth clusters. Inference favours distributed, latency-sensitive serving. One architecture rarely fits both.
DATA RESIDENCY
Where data and models live affects which infrastructure options are viable, particularly for regulated and public-sector organisations.
Scope
Questions the Compute Blueprint answers.
- 01How many GPUs do we actually need?
- 02Which accelerator architecture fits the workload?
- 03Should we buy GPUs or rent capacity?
- 04Should we use public cloud?
- 05What utilisation level makes dedicated infrastructure economical?
- 06How much networking bandwidth is required?
- 07How much storage is required?
- 08Do we require NVLink / InfiniBand / Ethernet?
- 09Where should the infrastructure be located?
- 10How will capacity requirements change over 12–36 months?
- 11How should training and inference infrastructure differ?
- 12What data residency requirements exist?
- 13What will the infrastructure cost?
Deliverables
What you receive.
The Compute Blueprint is a documented recommendation your organisation owns. Use it to proceed with Peregrine, with another provider, or internally.
EXECUTIVE RECOMMENDATION
A clear, board-ready summary of the recommended infrastructure approach and why.
WORKLOAD ANALYSIS
Characterisation of training, fine-tuning and inference workloads, including growth assumptions.
GPU SIZING
Estimated accelerator count, architecture and memory requirements for the target environment.
INFRASTRUCTURE ARCHITECTURE
Compute, networking, storage and platform design for the recommended deployment.
COST COMPARISON
Estimated infrastructure economics across buy, rent and cloud options.
CAPACITY ROADMAP
How capacity could be phased in over 12–36 months as demand changes.
DEPLOYMENT OPTIONS
Premises, Australian data centre, colocation, Peregrine capacity, cloud or hybrid — with trade-offs.
RISK & CONSTRAINT ANALYSIS
Supply, power, data residency, skills and operational risks that affect the recommendation.
Process
From workload to recommended architecture.
WORKLOAD
GPU REQUIREMENT
INFRASTRUCTURE
COST MODEL
RECOMMENDED ARCHITECTURE
- Engagement
- Typical engagement: approximately 2 weeks.
- Commercial model
- Fixed fee
- Deliverable
- Compute Blueprint document
- Next step
- Build, Operate, Capacity — or none
Request a Compute Blueprint.
Tell us about your workloads and constraints. We will scope the assessment and confirm the fixed fee before any work begins.