Foundations
Inside a data centre.
Follow the physical path from connectivity and power to cooling, racks and the compute hardware inside the data hall. The walkthrough shows general data centre infrastructure first, then how Allkira’s proposed off-grid model differs.
- Identify the main physical systems inside a data centre and how they depend on one another.
- Tell a conventional grid-connected data centre apart from Allkira’s proposed off-grid model.
- Recognise what changes between facilities, so a claim about one is not applied to all.
Two contexts appear in this walkthrough. Most scenes describe general industry infrastructure that applies to many facilities. The final scene describes Allkira’s proposed off-grid model, which is a design direction that requires project-specific validation, not an operating facility. The transition is labelled where it happens.
Inside a data centre
Where the cloudactually lives
Every message you send and model you talk to runs on real machines in a building like this.
The racks
Where computehappens
Rows of processors run the AI services you use every day. This is where the computing happens.
The meet-me room
Plugged intothe network
Fibre connects the building to the wider network. How many routes and how much latency depend on the site.
Backup power
Riding througha grid dip
In a grid-connected data centre, batteries bridge a power dip and generators can run longer, depending on the design.
Cooling
Keeping it cool
Computing produces a lot of heat. Cooling systems move that heat out before it slows the machines.
The Allkira model
Generatedon site
Allkira's proposed model generates power on site, from solar and storage rather than the grid.
The nerve centre
Always powered,always watched
Batteries bridge every flicker while engineers watch each rack and watt, around the clock.
The Allkira model
A proposedoff-grid model
A proposed model for off-grid, renewables-powered AI campuses, designed for Australian conditions.
Get in touchThe walkthrough, in words
What you saw, scene by scene.
A readable version of the walkthrough. For each part of the building: what it is, what it does, why it matters, and what changes between one data centre and another.
The data hall
- What is this?
- Rows of servers and processors, the computers that actually run AI.
- What does it do?
- Carry out the calculations behind every request you send and answer you receive.
- Why does it matter?
- This is the demand that everything else in the building exists to support.
- What varies between data centres?
- How densely racks are packed and how much power they draw differs widely between facilities and workloads.
Connectivity
- What is this?
- The point where fibre optic cabling connects the building to the outside network.
- What does it do?
- Carries data into and out of the facility.
- Why does it matter?
- Without connectivity the compute inside cannot be reached.
- What varies between data centres?
- The number of separate routes and carriers, and the resulting latency, depend on where the site is and which networks it can reach. Not every data centre has many carriers or the same latency.
Power and backup
- What is this?
- Transformers and switchgear, battery systems, and in many facilities backup generators.
- What does it do?
- Deliver a steady electricity supply and keep it running through disturbances.
- Why does it matter?
- Compute stops the moment power does.
- What varies between data centres?
- A grid-connected site typically uses short-duration UPS batteries to bridge seconds, and backup generators for longer outages, with the run time set by the system design and fuel on hand.
Cooling and water
- What is this?
- The equipment that removes heat from the building.
- What does it do?
- Keeps the hardware within a safe temperature range.
- Why does it matter?
- Uncontrolled heat slows down or damages the machines.
- What varies between data centres?
- Air cooling, liquid cooling and different heat-rejection methods use different amounts of energy and water, and the local climate changes the result.
The Allkira model
Allkira proposed model- What is this?
- Allkira’s proposed off-grid design: on-site solar generation, deep battery storage and integrated energy and compute.
- What does it do?
- Aims to run the campus on its own generation and storage rather than a grid connection.
- Why does it matter?
- It changes where and how a data centre can be built, and how it relates to the grid and the community.
- What varies between data centres?
- This is a proposed architecture that requires project-specific validation. It is a design direction, not an operating facility.
Two models
Conventional data centre, and the Allkira model.
Conventional data centre
General industry configuration
- Grid connection and transformers
- UPS batteries for short interruptions
- Backup generation for longer outages
- Cooling plant
- Fibre connectivity and data halls
Allkira model
Proposed design, requires validation
- On-site renewable generation
- Deep battery storage
- Islanded, off-grid operation
- Integrated energy and compute design
- Dry cooling and a zero-combustion design direction
The Allkira model is a proposed reference design. It reflects a current design direction and requires project-specific validation. It does not describe a secured, commissioned or operating facility.
In summary
One building, many dependent systems.
A data centre brings together compute, connectivity, power, backup and cooling, and each depends on the others. The details, from cooling method to backup duration, vary between facilities, so a figure measured at one site does not describe them all. Allkira’s proposed model changes the energy foundation by generating and storing power on site, a design direction that still needs project-specific validation.
Key terms
- Compute
- The processors and servers that carry out calculations.
- Data hall
- The room that houses the racks of servers.
- UPS
- Uninterruptible power supply: batteries that bridge short power interruptions.
- Firm power
- Electricity that is available on demand, around the clock.
- Latency
- The delay between sending a request and receiving a response.
- Islanded or off-grid
- Running on local generation and storage instead of a grid connection.