APPROACH
How does AI data centre actually behave?
{The question no one can independently answer, until now}
AI compute is grid-critical infrastructure. Some facilities draw
more electricity than the cities beside them.
How they behave is no longer an engineering detail. It's a
financing question, an insurance question, and a regulatory question.
The answer today comes from the operator.
We think it should come from the evidence.
The Architecture.
{Patent Pending - filed May 2026}
{01}
Independent Witness
The entity producing the evidence has no stake in the outcome.
{02}
Bounded Scope
Verified at a point the operator does not control.
{03}
Signed Record
A tamper-evident, auditable record that survives downstream review.
{04}
Open Rails
Built on the attestation standards institutions already trust.
We work alongside regulators, network operators, lenders, and insurers.
{Six problems. One evidence layer.}
1
Financing and covenant evidence
GPU-backed debt costs 11–15%. Traditional DC debt costs 3–5%. The difference is uncertainty. We eliminate it.
2
Insurance and risk pricing
Parametric triggers need an independent oracle. Operator dashboards are not independent. We are.
3
Regulatory compliance
For AI-grid connections subject to new performance standards across AEMC, AEMO, and equivalent jurisdictional frameworks.
4
Cross-jurisdictional evidence
Evidence that travels, across regulators, capital markets, and export-control regimes. One record, every jurisdiction.
5
System-stability witnessing
When a gigawatt load misbehaves, the post-event review needs evidence that existed before the event. We produce it.
6
Sustainability and disclosure
Scope 2 attribution for AI compute. Not what the operator claims. What was independently measured.
Every GPU-hour is sold. Every capacity claim is financed. None of it is independently verified.
{Until now.}
THE REASON
Why we built
Canus
{Reason 3}
No one independently verifies what AI data centres actually do to the grid. They self-report.
Regulators accept it. Lenders price around it. Insurers exclude it.
We think that's a structural gap, not a feature.
{Reason 1}
AI infrastructure should be verifiable, not just visible. Dashboards show data.
We produce evidence.
{Reason 2}
GPU-backed lenders pay 11–15% where traditional data centre debt costs 3–5%.
That's not a risk premium. That's the price of flying blind.
Independent evidence compresses the spread. That spread is our entire market.
From the Canus research programme.
{Submissions, technical positions, and case studies on AI compute infrastructure}
