HV Design and Commissioning for Hyperscale, AI and Tier III/IV Data Centres
Australia’s data centre sector has shifted scale. The traditional 5 to 50 megawatt colocation facility is no longer the dominant build pattern. Hyperscale campuses at 100 MW and above are now the standard for major cloud providers and AI infrastructure operators. AI training campuses are pulling forward 200 MW and 400 MW connections that did not appear in the country’s grid planning forecasts even three years ago. The capital pipeline is well above $70 billion and the constraint on the sector is no longer demand. It is high-voltage grid capacity.
This shift changes what HV commissioning looks like for data centres. A traditional Tier III colocation facility typically connects at 22 kV or 33 kV through a single feeder substation. A hyperscale or AI campus connects at 132 kV or higher, often with two or more feeders, often with on-site generation, and often with dedicated protection schemes that didn’t exist in the colocation playbook. Megavar provides end-to-end HV capability across the full spectrum, from a single Tier III substation through to multi-feeder hyperscale campuses.
Megavar’s Launceston engineering team is engaged on Project Southgate, Firmus Technologies’ AI factory campus in northern Tasmania. Project Southgate is Australia’s flagship sovereign AI infrastructure programme, expanding from its Launceston site to a planned national network of approximately 1.6 GW across Tasmania, Melbourne, Sydney, Canberra and Perth by 2028. The project is delivered in partnership with CDC Data Centres and NVIDIA, powered by Tasmanian renewable energy, and built around fully liquid-cooled AI factory modules. Megavar’s geographic alignment in Launceston and across Australia positions the firm to support hyperscale and AI campus delivery across the national rollout.
Hyperscale and AI: A Different HV Profile
Hyperscale and AI training campuses impose HV requirements that traditional data centre commissioning experience does not fully cover:
- Direct grid connection at 132 kV or 220 kV. Hyperscale campuses bypass the medium-voltage distribution network and connect directly to the transmission grid through dedicated transmission-class substations.
- Two or more independent grid feeders for resilience. The redundancy requirement at hyperscale is at the feeder level, not just the substation level.
- Dynamic and bursty load profiles. AI training workloads can swing from 30 percent to 100 percent within seconds. Protection schemes, transformer thermal modelling, and harmonic management must accommodate this.
- Liquid cooling on the IT load. Liquid-cooled GPU systems shift the cooling power split. The power infrastructure and the protection ratings must reflect the new mechanical load distribution.
- Power purchase agreements with on-site behind-the-meter generation. Solar, BESS, and gas peaking are increasingly part of the HV architecture, which means commissioning includes generator paralleling, BESS commissioning, and the protection coordination between behind-the-meter sources and the grid feeder.
- Grid connection lead times of three to five years. Substation design, transformer ordering, and protection scheme finalisation must start before the IT design is locked. The HV programme is on the project’s critical path.
Megavar’s commissioning engineers work on these campuses across Australia. Our combined design and commissioning capability lets us run the HV workstream as a single thread from the early connection studies through to energisation, rather than handing off between separate firms.
AI Training Campuses: What's Different
AI training is a distinct workload from cloud or colocation. The HV implications are specific:
- Power density is two to three times that of traditional cloud workloads. A typical AI training rack now exceeds 100 kW. Some Nvidia GB200 NVL72 systems exceed 130 kW per rack. The supply substation must be sized for this density.
- Synchronised compute means synchronised load. Thousands of GPUs running a single training run draw power in lockstep. The grid feeder sees a load profile that looks more like a steel mill than a traditional data centre.
- Power conditioning matters more. AI workloads produce harmonics from the GPU power supplies that interact with the transformer secondary. Harmonic studies and filter design are part of the HV scope.
- Resilience to failed training runs. A failed training run is a multi-million-dollar event. The HV system must support deterministic restart sequences and instrumented load shedding.
- Liquid cooling on the IT load. Liquid-cooled GPU systems shift the cooling power split. The power infrastructure and the protection ratings must reflect the new mechanical load distribution.
- Behind-the-meter generation is increasingly common. Operators are signing PPAs with on-site solar and BESS to manage exposure to grid prices and to support sustainability commitments.
These are HV engineering problems before they are IT problems. Getting the substation design and commissioning right is a precondition for the campus operating at all. Project Southgate in Tasmania is an Australian benchmark example: a 100 percent liquid-cooled AI factory campus, powered by Tasmanian renewables, designed around fully integrated power, cooling, and compute in modular form. Australian AI campus delivery in 2026 and beyond will follow this engineering pattern, and Megavar’s combined design and commissioning capability is positioned to support it.
Services for Data Centres
- HV substation design: Complete primary and secondary design for data centre incoming supply substations from 33 kV through to 132 kV transmission-class. Includes single-line diagrams, protection coordination studies, earthing design, insulation coordination, cable schedules, transformer specification, and detailed design documentation. Sized for hyperscale and AI campus loads as well as Tier III and Tier IV colocation.
- Protection engineering: Protection scheme design and coordination for dual-feed and triple-feed substations, bus section arrangements, transformer protection, automatic changeover systems, and the integration between grid feeders and behind-the-meter generation. Designed to isolate faults rapidly without cascading to redundant feeds.
- HV commissioning: Full pre-commissioning and commissioning of data centre HV infrastructure including transformers, switchgear, protection relays, SCADA integration, generator paralleling systems, automatic transfer switches, on-site BESS where deployed, and the integration with the local network operator’s protection schemes.
- Power system studies: Load flow analysis, fault level studies, harmonic analysis (critical for AI workloads), arc flash assessments, and motor starting studies for liquid cooling pumps. Also includes dynamic load modelling for AI training profiles where the operator requires it.
- Grid connection support: Engineering support during the connection negotiation with the local network operator, including connection studies, model validation, and protection scheme review. The grid connection process is the single longest lead time in the project and the most common source of delay.
- Ongoing maintenance: Scheduled maintenance programmes for data centre HV assets, designed to fit within concurrent maintainability windows without impacting the IT load. Transformer condition monitoring, protection relay testing, switchgear maintenance, earth system verification, and thermal imaging across the substation and supply infrastructure.
Tier III and Tier IV Colocation HV Requirements
Tier III (concurrently maintainable) and Tier IV (fault tolerant) colocation facilities remain a core part of the Australian data centre sector. Megavar designs and commissions HV infrastructure for both classifications:
- Tier III requires redundant electrical paths so any component can be maintained without taking IT load offline.
- Tier IV requires independent, physically separated distribution paths with automatic failover.
- Both require protection coordination that prevents cascading trips between redundant feeds.
- Both require automatic transfer scheme response times within the operator’s defined window.
Megavar designs and commissions HV systems that meet the Uptime Institute’s Tier requirements, supporting the path to certification.
Addressing the Grid Connection Bottleneck
Grid connection is the defining constraint for hyperscale and AI campus development in Australia. Connection lead times of three to five years for transmission-class connections are now standard. Transformer and switchgear delivery windows have stretched as global demand has surged. Megavar works with developers and their head contractors from the connection negotiation phase, helping to identify grid risks early, optimise substation configurations to suit the available connection capacity, and plan commissioning programmes that align with tight construction schedules. Our combined design and commissioning capability lets us manage the HV infrastructure timeline as a single workstream.
Equipment for Data Centre Commissioning
Frequently Asked Questions
for Data Centres
Does Megavar commission hyperscale and AI data centres?
What's different about HV commissioning for AI training campuses?
Can Megavar work at 132 kV and 220 kV transmission class?
What Tier levels does Megavar design and commission to?
Can Megavar help with the grid connection process?
Does Megavar commission BESS and behind-the-meter generation at data centres?
What makes Megavar different from other data centre commissioning providers?
Need High Voltage Testing or Equipment?
Talk to our engineers about your project requirements.
Get In TouchTrusted by top enterprises for exceptional results
Explore the latest insights
We’re incredibly excited to share a big milestone in Megavar’s story—we’ve officially opened our new Newcastle branch, located at 22 Gavey Street, Mayfield,
