THE STRATEGIC SO WHAT
Ask better questions of the infrastructure.
Perspectives on capacity, delivery risk and usable AI compute for policy, media, investors and the AI community.
Demand, workloads and commercial commitments.
Read the latest source-checked evidence on electricity demand, model training and inference, disclosed hardware and customer commitments. Each claim retains its geography, date and scope so a portfolio announcement cannot become a site-level GPU estimate.
Explore demand and compute research →From ambition to delivery.
A large announced pipeline describes ambition. To assess delivery, follow each project through evidence of planning, power availability, financing, construction, fit-out and commissioning. The atlas separates the capacity stages that our present sources support.
For policy and investment decisions, the useful question is how much incremental IT capacity can be delivered, where and when, and what remains uncertain. The answer needs dated milestones and explicit physical scopes. A campus headline cannot safely be added to the same campus’s funded phases.
Differences between operators’ disclosure practices can dominate comparisons; a larger disclosed subtotal does not establish a larger operating footprint.
Inspect the construction records → · Read the counting rules →
From power to usable AI compute.
MW measures electrical capacity, not the number or capability of accelerators. A useful compute inventory also needs GPU or accelerator model, installed quantity, operational status, interconnect, cloud or private access, location and an observation date.
Before making a compute estimate, distinguish an announced purchase from an installed deployment, and a provider’s global inventory from equipment demonstrably operating in Australia or New Zealand. Conversion assumptions should be published, not hidden inside a headline GPU number.
The compute research register separates sourced deployments, advertised capabilities and commercial commitments. It does not publish an aggregate GPU fleet total or derive GPU counts from MW.
From a location to an assurance claim.
A physical facility location answers one question. Ownership, service operation, certification, data storage, processing and access controls answer different questions. A credible assessment needs a defined service and workload as well as a map pin.
The atlas records exact Australian HCF facility and enclave scopes, including visible unknowns. An enclave entry must not be presented as certification of the whole building, and an Australian framework should not be assigned to a New Zealand site.
Explore matched Strategic scopes → · Understand the scope boundaries →
Using the research.
Find source-backed facility records, inspect comparable capacity claims and see where disclosures remain incomplete. Site-level subtotals describe the named facilities and capacity scopes in the evidence. They do not represent an exhaustive national supply estimate, GPU census or commissioning forecast.
Compare the source date, approval stage, power requirements and commercial commitments for each project. These details help separate near-term delivery from longer-term development potential.
Explore the evidence base →