Designs
How we engineer an energy system for a workload and a place.
Allkira develops reference designs — complete generation, storage, cooling, fibre and firming configurations sized to a defined customer, workload, reliability requirement and set of site constraints. A reference design shows how the method applies to a scenario. It is not evidence that a site has been secured, approved, financed or commissioned.
The design method
Infrastructure becomes financeable one decision at a time.
Every design advances through the same gates. Each is tested before further development capital is committed; a design advances only while every critical test remains open or passed. This is how Allkira retires risk — a method, not proof that any design or site has been validated.
- 01Compute suitabilityIs the workload energy-seeking, and can a share of it flex with supply?
- 02Energy resourceIs the solar resource — and the land — strong enough at this location?
- 03Reliability designCan the system stay firm through the worst season, with no grid behind it?
- 04Land & approvalsIs the site securable and consentable under a State Significant Development pathway?
- 05ConnectivityIs diverse, low-latency fibre reachable?
- 06DeliverabilityCan it be built from track-record components with a credible EPC?
- 07Commercial bankabilityDoes the offtake and cost stack finance as infrastructure?
Reference designs
Worked examples of the method.
Each reference design applies the method to a defined region, workload class, capacity and grid relationship. More will be added as the method is applied to new scenarios.
Reference design 01
Sunline
Sunline is a hypothetical energy solution designed for a neoscaler seeking to deploy approximately 10 MW of energy-seeking AI compute in regional New South Wales. It shows how the Allkira design method applies to a defined customer and location. It is not evidence that a site has been secured, approved, financed or commissioned.
Intended customer & use
- Intended customer
- A neoscalerCustomer dependent
- Intended use case
- AI training, batch processing, fine-tuning and asynchronous inferenceCustomer dependent
Scenario
- Location context
- Orana region, Dubbo–Wellington corridor, Central-West NSW (~32.2°S)Site dependent
- Indicative IT capacity
- ~10 MW IT, 24/7; facility ~10.9–13.1 MW with ambient overheadDesign assumption
- Offtake shape
- 70% firm / 20% flexible (≤1h shed) / 10% preemptibleDesign assumption
Proposed architecture
- Generation
- ~100 MWp fixed-tilt (35° north) solar, ~115–125 haDesign assumption
- Storage
- ~240 MWh usable grid-forming LFP; dual-track (second-life base case, new-cell priced fallback)Design assumption
- Firming reserve
- ~14 MW renewable-fuel (HVO) block (6 × 2.8 MW, N+1) with a 420 kL EN 15940 fuel farmDesign assumption
- Power spine
- 33 kV AC collection ring → rectifier plant → ±400 VDC hall bus → 48 V racksDesign assumption
- Cooling
- Direct-to-chip liquid + dry (air-cooled) heat rejection; DX trim above ~32–35 °C; zero mains waterDesign assumption
Reliability, water & land
- Reliability target
- Firm-tranche LOLE ≤ 1 h/yr; ≥ 99.5% monthly availability to the firm trancheDesign assumption
- Modelled reliability
- Firm-tranche LOLE < 0.05 h/yr at the 420 kL fuel store (screening)Modelled output
- Water use
- No mains water; sealed glycol/treated loops with kilolitre-scale trucked top-up; static firewaterDesign assumption
- Land
- ~115–125 ha for the solar field on cleared, low-value grazing landDesign assumption
Modelled outputs (screening estimates)
- Energy mix
- 47% solar-direct / 52% via storage / 1% reserve-directModelled output
- Renewable-fuel share
- ~2.9% of annual energy; ~199 reserve-generator hours/yrModelled output
- Curtailment
- ~42% of generation — the structural price of winter-firm islanding, and feedstock for the preemptible tierModelled output
- Indicative cost
- Capex ~A$225M (second-life) / ~A$244M (new-cell); firm LCOE ~A$348/MWh (SL) / ~A$378 (new) at the IT busModelled output
Requires site-specific validation
- Resource adequacy
- Historical multi-year weather backtest (WP-1) at pre-registered acceptance thresholdsRequires validation
- Stability & protection
- EMT (PSCAD-class) study of the grid-forming island (WP-2)Requires validation
- Site yield
- Site-specific solar-yield data package (currently anchored to an operating farm ~50 km away)Requires validation
- Storage
- Second-life fade coefficients and aged-module fire testing for the fire safety studyRequires validation
- Commercial
- Firm supplier quotes, final planning conditions and customer-specific requirementsRequires validation
Sunline is an indicative reference design. Figures are screening estimates from a reproducible model calibrated to verified regional climatology; they are labelled as such and would be superseded by site-specific studies before any project decision. No site has been secured, approved, financed or commissioned. Design-direction note: Allkira's current model direction is combustion-free — backup from array oversize and deep distributed battery reserves. This published reference design carries a small renewable-fuel reserve and its modelled outputs reflect that configuration; a combustion-free revision is in design.
In plain terms
What a reference design is — and is not.
It is a demonstration of how Allkira's method applies to a defined customer, workload and location — with every material line labelled as a design assumption, a modelled output, or something that requires validation.
It is not a live project. It does not mean a site has been secured, approved, financed or commissioned. When a real requirement and location are in play, the same method produces a site-specific design, validated at each gate.
For compute partners
Bring capacity online without the grid queue.
For compute customers and data-centre operators with energy-seeking workloads: Allkira develops and owns off-grid, renewables-powered campuses and contracts firm, 24/7 power — energised in months rather than years.
Time-to-power and firm 24/7 supply are design targets in validation In Validation, closed by the WP-1 weather backtest and WP-2 stability study.
- Which workloads fit
- Energy-seeking compute — AI training, batch processing, fine-tuning and asynchronous inference — where longer-running jobs can locate near strong energy rather than near end users. Latency-bound workloads stay on the metro grid.
- How capacity is staged
- Each campus is built from standardised, factory-integrated blocks that are independently energised and replicated to scale — a proof unit of 5–20 MW growing to a portfolio of hundreds of megawatts. The first block earns while the tenth is still being delivered.
- How firmness is demonstrated
- Firm power off-grid is treated as engineering, not a footnote: solar oversized for the worst season, right-sized grid-forming storage on the DC bus, deep distributed battery reserves engineered statistically for extended low-sun periods, and a contracted flexible tranche. No combustion on site. Reliability is tested at each development gate before capital advances (validation comes from the gating studies, not the gates).
- Cooling and water
- Direct-to-chip liquid with dry heat rejection and no mains water — a durable answer where cooling water is a consent flashpoint.
- Fibre and connectivity
- Designed for diverse, low-latency fibre appropriate to energy-seeking workloads, assessed per site.
- The commercial relationship
- Allkira builds, owns and operates the whole power stack and contracts firm, 24/7 power — optionally with colocation — under a long-term take-or-pay agreement. You buy capacity and outcomes, not a construction project.
- Where you can influence the design
- Early. Workload profile, flexibility, reliability tranches and expansion pathway are design inputs — the earlier you engage, the more the reference design is shaped around your requirement.
Saying hello takes one line — detailed workload, capacity, reliability and deployment requirements come after the first exchange.