Founder notes
Voice 1
Short memos from Will and Matt on what we're building and why — written as the decisions get made, published as they land.
Coming soonAllkira Voices
AI is physical. Learn the system behind it.
A source-led learning series for students, educators and communities exploring the compute, energy, cooling, land and people behind artificial intelligence.
Start here
Inside a Data Centre
Follow the physical path through a modern data centre, from fibre and electrical infrastructure to racks, cooling, controls and the compute hardware inside the data hall.
What you will learn
Start the walkthrough- Identify the main physical systems inside a data centre.
- Understand how compute, power, cooling and connectivity depend on one another.
- Distinguish a conventional grid-connected data centre from Allkira’s proposed off-grid model.
Follow the system
A path from the building to the grid, and back to the community.
Each module builds on the last. Start inside a data centre, then follow how compute becomes power demand, how energy and heat are managed, and how location and community shape the whole system.
From Workload to Electrical Load
Coming nextHow compute behaviour becomes power demand.
Generation, Storage and Firm Power
Coming nextHow energy systems support continuous operation.
Cooling, Heat and Water
Coming nextWhere heat comes from and how different cooling systems remove it.
Location, Land and Fibre
Coming nextWhy workload, energy, connectivity and site conditions affect location.
Infrastructure and Communities
Coming nextThe questions communities should ask about land, water, noise, energy, employment and accountability.
Question the claim
AI needs better questions, not simpler answers.
Claims about AI often collapse a whole system into a single number. Here is how Voices tests one. Open each claim to see what is established, what varies, what is being measured and what stays uncertain.
“AI uses a fixed amount of water per query.”
Established
Running AI hardware produces heat, and removing that heat can use water, so there is a real water dimension to AI.
Varies by design or location
The model and task, the hardware, utilisation, the cooling architecture, the climate and location, and the energy source all change the figure.
Measurement boundary
A per-query number depends on whether it counts only on-site cooling or also the water used to generate the electricity, and over what period it is measured.
Remaining uncertainty
Published per-query figures are usually estimates drawn with different boundaries, so they are rarely comparable.
“More efficient chips will reduce total electricity demand.”
Established
Newer hardware can do more computation for each unit of energy.
Varies by design or location
Energy per task can fall while total demand still rises, because adoption, model scale and computationally intensive uses are growing at the same time. The IEA frames AI energy demand as a combination of efficiency, uptake and changing capabilities.
Measurement boundary
Demand can mean one task, a data hall, a company or a whole grid, and the answer is different at each scale.
Remaining uncertainty
How far future efficiency, adoption and new capabilities offset one another is not settled.
“All data centres have the same environmental footprint.”
Established
Every data centre uses energy and carries some environmental footprint.
Varies by design or location
Workload, power density, energy supply, cooling, water use, utilisation, location and operating model all change the outcome.
Measurement boundary
Footprint depends on whether you count electricity only, or also water, land, materials and the local grid mix.
Remaining uncertainty
Comparisons are only fair when the boundary and time period match, which is often not the case.
Who Voices is for
Built for the people who use, teach and live alongside this infrastructure.
Students
Explore how the digital services you use depend on physical systems, engineering and energy.
Educators
Use source-led explanations, diagrams and supporting material to introduce the infrastructure behind AI.
Communities
Understand the practical questions to ask about energy, water, land, noise, connectivity, employment and local participation.
Educator and community material is being developed. To talk about using Voices with a class or a local group, start a conversation with Allkira.
Why Allkira Voices exists
Understanding should come before advocacy or opposition.
- AI is physical, not purely software.
- Infrastructure decisions carry engineering, environmental and community consequences.
- People should be able to test claims against evidence.
- Australia needs capability across energy, computing, engineering, environmental and community disciplines.
- Understanding should come before either advocacy or opposition.
In our own words
Notes, podcasts and interviews from the build.
Where Allkira publishes what it says out loud: founder notes, podcast episodes and interviews. Perspectives from engineers, educators, students and communities will sit alongside them as they are recorded.
How Voices is published
Every claim carries its evidence.
Voices content is written to a single standard, and each module states these openly so you can check the work.
- The source
- Geography
- Timeframe
- System boundary
- Evidence status
- What varies by design or location
- Allkira-specific modelling
- Remaining uncertainty
- Publication and review dates
Applied to Allkira’s own work: field note, July 2026
- EstablishedOff-grid, zero combustion, no mains water; the honest premium over grid-firmed supply. Supported by design doctrine and cited standards.
- In ValidationThe modelled energy mix, reliability, reserve sizing and firm LCOE: screening estimates, credible and reproducible, not yet closed.
- OpenThe historical weather backtest (WP-1), the EMT stability study (WP-2) and site-specific yield: the artefacts that move the modelled figures to established.
Established evidence is supported by a dated output or a cited standard. In Validation covers design decisions and modelled results not yet closed by their gating study. Open marks questions that still need site data, testing or modelling. Allkira reference designs are labelled as proposed, not as operating projects.
Educational approach
Infrastructure context
Allkira primary sources
Keep learning the system.
Start with the walkthrough, ask a question, or tell us how you would use Voices.