
AI / GPU Data Centre Delivery — Phase-Gate Reference | RegionCo Model
This playbook governs how we deliver data centres — from investment thesis to operating asset — across nine phase gates.
Brief new members fast with a single source of truth.
Structured phase gates keep every site on track.
Consistent delivery from 30MW to 300MW.
Nine phases. Nine gates. One standard.
Each gate must be cleared before the next phase begins — no exceptions.
Fix the thesis before the site. This phase converts an opportunity into a defined product with bankable economics.
Define retail colo / hyperscale / dedicated AI-GPU; set target IT load and staged capacity (30 → 100 → 300MW).
AI workloads run 40–130kW+, forcing liquid cooling. Tier target: AIDC typically Tier III (concurrently maintainable).
Lock take-or-pay economics that make the asset financeable before any capital is committed.
Run the go/no-go Investment Committee screen. Gate: IC screen approved. Owner: CEO, CCO, CFO.
The make-or-break phase — a data centre is a power-and-connectivity play in a real-estate wrapper. Screen and secure the parcel.
Power availability & time to energisation; water for cooling; fibre/latency; zoning; hazard/climate exposure.
Negotiate option/SPA; confirm title and buildable capacity. Acquire land and M&E rights.
Due diligence on title provenance, seller UBO and consideration flow before committing.
Secure the utility connection and lock the environmental design basis that lenders and tenants underwrite.
Secure grid connection or PPA. Options: grid, LNG, solar, battery; hydrogen as differentiator.
Best-in-class AIDC ~1.1–1.2 with liquid cooling. WUE and carbon intensity targets set at this phase.
Renewable overlay, water strategy, green certification path — what lenders and hyperscale tenants underwrite.
Signed grid/PPA term sheet + ESG design basis. Owner: CTO, CFO.
DesignCo takes the scheme from concept to issued-for-construction, de-risking a modular, certifiable build.
Utility → MV → transformers → UPS → PDU → rack, with N+1 / 2N redundancy per Tier target.
Select CRAH vs. direct-to-chip liquid vs. immersion for GPU density. Fix PUE/WUE design.
DCIM stack, fire and life-safety, structural. Favour modular / prefabricated / containerised to compress schedule.
Fix the Uptime design + facility certification path. Gate: IFC design freeze + Tier design cert. Owner: Group CTO.
Close the capital stack. Lock long-lead equipment. Don't let procurement become the critical path.
Advance rate on Compute SPV GPU debt
Typical lead times for generators, switchgear, transformers & chillers
Equity + construction debt + GPU-collateralised debt
Close equity + carry (SPV level), construction/project debt, and GPU-collateralised debt (~60% advance) for the Compute SPV.
Order generators, switchgear, transformers, and chillers early — these carry 12–24 month lead times and are the schedule's critical path.
Execute the EPC / design-build contract at JV/SPV level — not as a GP subcontract — to preserve role separation. Gate: Financial close + EPC executed. Owner: CFO, CEO.
The owner's-rep PMO governs the build against the M&E critical path. Manage schedule, cost, quality, safety and change control; hold monthly cost-to-complete and risk-register reviews. Track long-lead delivery and run factory acceptance tests (FAT) on critical equipment.
Gate: Mechanical completion. Owner: Local SPV CEO, PMO.
The phase amateurs skip — proving the facility performs to SLA under load. Five levels: L1 factory → L2 site component → L3 subsystem → L4 integrated systems → L5 IST. Full-load black-building / pull-the-plug tests on generators and UPS under simulated IT load.
Gate: L5 IST sign-off + RFS. Owner: CTO, Operator.
Our operating bench takes the asset live and holds it to the SLA.
Stand up round-the-clock Network Operations Centre, O&M, DCIM/digital-twin monitoring, and incident/change control.
Automated KPI scorecards tied to take-or-pay SLA — Tier III targets ~99.982% uptime.
Continuous PUE/WUE/carbon reporting to LPs and tenants. Gate: SLA in force + first tenant live. Owner: OpCo.
One playbook. Infinite sites. The McDonald's model for data centres.
First site — prove the playbook.
Split SPVs, add PMO headcount, standardise procurement.
Full platform scale — standardised design, procurement and operations shorten each subsequent build.
Six roles. One chain of command. Zero ambiguity.
I Build Concepts (System) Pty Ltd — This is the definitive framework for delivering world-class AI/GPU data centres at scale.
From thesis to operating asset
Scalable from 30MW
With liquid cooling
Critical M&E equipment
Data Centre Development Playbook