Services
Data Centre & Hosting
Avenue Group Australia collaborates with Green Racks Data Centre to relocate steady state IT workloads away from aging server environments and cloud pricing that comes with unpredictability and into an Australian owned facility based in Perth. Avenue does the environment design and migration process. Green Racks manages the floor space. Your data stays onshore.
REVIEWED ON
- Years10+
Field Experience
- Projects50+
Around the Globe
- Minute0
Response Time
- Client approval10%
Why This Decision Is Landing on Your Desk Right Now
Three separate pressures have converged, which is why infrastructure conversations that used to be a five-year hardware refresh are now something bigger.
- The server room is ageing. That part isn't new.
- The cloud bill stopped shrinking. Egress, inter-zone traffic, and orphaned storage volumes are now permanent line items, not one-off surprises.
- Regulators and boards are asking sharper questions. Data residency has gone from a nice-to-have to something a tender or an auditor will actually test.
The numbers behind that:
of Australian data centre demand in 2025 in 2025, were met by colocation instead of hyperscale self-build. Opting for colocation is not an alternative route; that’s how the majority of the market goes about it. (Source: Mordor Intelligence, Australia Data Center Market, 2026)
of IT leaders report that their infrastructure should remain in one nation only — a criterion not met automatically by any multi-regional cloud framework. (Source: Nutanix Enterprise Cloud Index, 2026)
the highest civil fine that can be imposed on anyone who interferes with another person’s privacy according to the new version of the Privacy Act, enacted since June 2025. (Privacy and Other Legislation Amendment Act 2024, Cth)
scoping calls
What we actually hear on scoping calls
Five-year-old hardware, out-of-warranty SAN, and capital expenditure competing with money-making projects. Why does this happen every five years for the infrastructure, asks the business.
Egress charges, inter-zone charges, gateway services, and orphan volumes were not considered in the business case. This estimate never tallies with the invoice.
Whether through a tender process or a board, the question asked is about where exactly the data resides and what legal obligation there is to produce it. A region setting in the console will not do.
A comms room designed to house four servers operates twelve at present. The summer season is considered the window of opportunity, and the failure occurs mostly on Fridays.
Backups are made regularly, however the last restore exercise is either not documented or took place more than a year ago. The recovery time target set out in the plan and the one practiced is two separate figures.
Information about the environment is held in someone’s mind. The absence, illness or vacation causes the usual modification to become a potential cause of an outage and prevents any other project.
There is something about the data that the law will not allow you to work with offshore. There is no way that you could use GPU resources somewhere outside of the country, and local GPUs require electricity and cooling.
A move of the office, fit-out or downsizing means the communications room is going to stay where it is. Your infrastructure decisions are now subject to another deadline.
How an Engagement Actually Runs
Migration goes through five stages within about six weeks’ time frame; they include work load assessment, target system design and quotation, contracting and continuity planning, provision and cutover, and stabilisation. Every stage is followed by creation of a document that will be yours regardless of whether you continue to the following stage or not. This is by design. Even if you decided not to migrate after assessment, you've still walked away with something useful.
Stage 1 (Week 1) — Workload Assessment and Placement
We audit what is being run, not what the asset registry tells us. Each workload is assessed based on four factors – predictability of demand, sensitivity of the data, tolerance to latency and urgency of the hardware refresh.
- Server and storage inventory, licensing details and dependencies
- Detailed assessment of network dependency and egress volume (what most business cases assume)
- Data classification according to Australian Privacy Principles and other industry-specific requirements
- Workload placement matrix with migration, cloud-based continuation, retirement or no action options
You walk away with: a workload placement matrix and a current state cost baseline, which is derived from actual network egress and staffing costs.
Stage 2 (Week 2) — Target Design and Firm Quote
Here is where rack allocation, network planning, IP addressing, backup targets, and recovery goals come into play. Recovery point objective and recovery time objective are mentioned for each workload and the quote you will receive is a fixed one and not indicative.
- Rack height, power requirement, and cooling footprint
- Network layout, addressing plan, and carrier route
- Backup targets, retention period, and DR tier for backup
- A fixed quote with onboarding cost listed separately
You walk away with: a target state design documentation, an RPO/RTO list, and a quote you can present to your board.
Stage 3 (Week 3) – Sign-Off, Contracting, and Contingency Planning
The contracts are finalized, and equally important, the plan is written for how things should be handled in case the cutover fails. This is the plan that really counts.
- Signing of service agreement and service level schedule
- Defined cutover runbook with owners and decisions points
- Rollback plan and abort threshold definition
- Stakeholder communication and notification during outage window
You walk away with: an executed contract, a cutover runbook, and a rollback plan with its specific abort criteria.
Stage 4 (Weeks 4–5) — Provisioning and Cutover
The rack, network, and managed infrastructure are provisioned and burned-in in advance of any activity. The actual cutover occurs during off-hours, in the window you specify, with engineers physically on-site at Osborne Park and a test restore having already been completed successfully.
- Provisioning and burning-in of infrastructure ahead of time
- Data seeding, with a successful test restore prior to the cutover
- Out-of-hours cutover, with engineers on-site
- Rollback available for the entire duration of the window
You walk away with: a cutover completion report, with test-restore documentation included.
Stage 5 (Week 6 onward) — Stabilisation and Steady State
The initial month is operated at increased monitoring vigilance levels, since what really counts is exposed by real production load rather than testing. The final review marks the end of the project and determines the subsequent operating tempo.
- Verification of post-cutover operation according to original design
- Initial month monitored carefully, with weekly reviews
- Scheduled and documented disaster recovery test
- As-built documentation transferred into steady-state support
You walk away with: as-built documentation, disaster recovery test documentation, and steady-state support schedule.
Service Models — Choosing Where the Responsibility Line Sits
The six service models below — colocation, managed colocation, managed servers, sovereign GPU, managed WordPress, and migration — don't differ by facility. They differ by how much of the operational stack Avenue Group Australia and Green Racks take on, versus what your team keeps.
Colocation — You Own the Hardware
Your equipment sits on our floor. Power, cooling, physical security, and connectivity stop being your problem, while your team keeps full control of the operating system, hypervisor, and applications. This is the model that changes the least about how your team already works.
Good fit if:
- Your hardware is recent enough to have real life left in it
- You have in-house infrastructure capability you intend to keep using
- Licensing is tied to specific hardware or socket counts
- An application vendor requires a configuration you need to control directly
Not the right fit if:
- Hardware is already out of warranty and due for replacement anyway
- Nobody on staff can administer the operating system layer
- The whole point of moving was reducing dependence on specific people
Responsibility split: Green Racks handles facility, power, cooling, physical security, and network. You handle hardware, OS, hypervisor, applications, and backups.
How it's priced: Rack unit or full-rack allocation, plus committed power and bandwidth — quoted against power draw and rack units, not per server.
Facility: Osborne Park, WA — modular, precision-cooled, fire-suppressed, under an hour from any Perth metro CBD address.
Managed Colocation — You Own the Hardware, We Operate It
Colocation with remote hands, monitoring, and maintenance layered on top. You keep your hardware and your licensing position, but you no longer need someone available to drive to a facility at 2am, or a shelf of spare parts sitting idle just in case.
Good fit if:
- Hardware still has useful life, but your team supporting it doesn't have the depth
- One person currently holds all the operational knowledge
- You want to retire the spares cupboard and the after-hours on-call roster
- Physical intervention is currently a single point of failure
Not the right fit if:
- Monitoring and alerting run against agreed thresholds
- Remote hands handle physical intervention — no site visit required from your side
- Firmware and hardware maintenance run on a schedule
- A named escalation contact, not a support queue
Responsibility split: Adds hardware operations, monitoring, and remote hands to the colocation baseline. The application layer stays with you.
How it's priced: Colocation base fee plus a managed operations fee, scaled to device count and monitoring depth.
Managed Servers — No Hardware, No Refresh Cycle
Fully provisioned, monitored, and maintained as a service. Nothing sits on your balance sheet, and there's no capital refresh looming every five years — the environment is specified to your workload and billed monthly. This is the model most enterprises land on when a refresh quote was the trigger for the whole conversation.
Good fit if:
- Hardware is at or past end of life and a refresh is the realistic alternative
- Finance would rather see operating expenditure than a capital request
- You want one accountable party for the entire stack
- Workload demand is steady-state and predictable rather than spiky
Not the right fit if:
- Managed server or VMs on isolated tenant networking
- Backups to a defined retention schedule
- Monitoring, patching, and support down to the base operating system
- A website audit, scaled to package tier
Responsibility split: Avenue and Green Racks cover everything from the facility through to the base operating system, patching, monitoring, and backups. You own your applications and your data.
On licensing: Microsoft 365, RDS CALs, and Windows licensing aren't included — they're itemised separately. Customer-specific applications stay supported by their own vendor unless separately agreed.
Published pricing: See the managed hosting packages — $599, $1,199 and $2,490 per month ex GST by user band.
Sovereign GPU — Pricing on Application
GPU capacity hosted onshore, built for training and inference on data that legally or contractually can't leave Australia. What's usually driving this conversation isn't cost — it's that legal has already ruled out an offshore region, and on-premises GPU means power and cooling density the building simply can't deliver.
Good fit if:
- Training or inference data carries a residency restriction
- Moving training sets in and out of cloud has become the actual bottleneck
- You need predictable GPU access rather than spot-instance availability
- Your building can't support high-density racks or liquid cooling
Worth knowing:
- GPU rack densities now routinely exceed 80kW — which is exactly why this is a facility question, not a procurement one
- Sovereign AI capacity in Australia is capacity-constrained right now, so lead time matters more than rate
- Training and inference have different load profiles and are quoted separately
Configurations and rates are quoted per engagement, against the model, dataset size, and utilisation profile. Bring the workload to the scoping call — a rack count on its own isn't enough to price against.
Managed WordPress — Included in Managed Packages
Business website hosting on Australian infrastructure, with the platform, updates, and backups all handled for you. Included at Business tier within the Essential package and Enterprise tier within Premium, or available standalone for $79/month ex GST.
- Platform, core, and plugin updates managed on a schedule
- Backups aligned to your package's retention schedule
- Hosted on the same onshore infrastructure as the rest of your environment
- Unlimited managed sites at Premium tier, under a fair use policy
This one matter more often than it sounds like it should. For organisations working through a data residency question, a marketing website still hosted offshore is frequently the last unresolved piece of the estate.
Migration — Six Weeks Indicative
The structured five-stage migration described above, with business continuity built into the sequence rather than added afterward. Onboarding covers rack, network, physical install, and cutover assistance.
- Assessment, design, contracting, cutover, and stabilisation as five discrete, document-producing stages
- A rollback plan with a defined abort threshold, held live through the entire cutover
- Out-of-hours cutover windows, with engineers physically on site at Osborne Park
- A proven test restore before the production cutover, not after
Onboarding is quoted separately from the recurring fee and invoiced ahead of cutover. Ask for it itemised — a migration quoted as "included" is usually a migration that was never properly scoped.
- Pricing models
What it costs, and how it is structured
Managed hosting is published pricing, starting at $599/month ex GST for 1–10 users. Colocation, sovereign GPU, and enterprise-scale environments are quoted individually, because their cost drivers are power draw and rack units, not user counts — publishing a flat rate for those would just be a guess with a dollar sign in front of it.
Managed hosting packages. All figures AUD per month, excluding GST.
| Inclusion | Basic | Essential | Premium |
|---|---|---|---|
| User band | 1–10 | 11–50 | 50–100 |
| Monthly, ex GST | $599 | $1,199 | $2,490 |
| Compute | Managed business VM, isolated tenant network | Dedicated managed server or VMs | Performance managed server, specified at audit |
| Storage included | 200 GB | 4 TB | 10 TB |
| Backup frequency | Nightly | Daily, off-site | Hourly |
| Retention | 7 days | 30 days | 90 days |
| Disaster recovery | — | DR-ready / DRaaS, cold standby | Full DR / DRaaS, warm standby |
| Managed WordPress | Add-on, $79/mo | Included, Business | Included, Enterprise |
| Website audit | Basic (RRP $299) | Advanced (RRP $499) | Included, specified at audit |
| Support | Monitoring, patching, business hours | Monitoring, patching, business hours | Priority, named contact |
| Scaling note | — | To 30 users; +$500/mo for 31–50 (second node) | Unlimited managed sites, fair use |
Four services are quoted rather than published, because the cost driver is not user count.
Colocation
Priced on rack units, committed power draw in kW and committed bandwidth. Two customers in identical rack space can differ several-fold on power. Bring your current draw or your nameplate figures to the scoping call.
Managed colocation
Colocation base plus managed operations, scaled by device count and monitoring depth.
Sovereign GPU
Quoted against model, dataset size and utilisation profile. Training and inference are quoted separately.
Onboarding and migration
Quoted per environment and invoiced at the pre-cutover stage. Covers rack, network, physical install and cutover assistance.
Enterprise above 100 users
Beyond the Premium band, environments are specified at audit and quoted against the target-state design produced at Stage 02.
Three commercial shapes, depending on how you would rather carry the cost and the risk.
Monthly recurring
Fixed monthly fee in Australian dollars. No capital request, no depreciation schedule, no currency exposure. The default, and the reason most refresh conversations end here.
Term commitment
A committed term in exchange for rate certainty across the term. Worth considering where the workload is genuinely long-lived and finance values a fixed forward number.
Hybrid placement
Steady-state workloads placed in the facility, elastic workloads left in hyperscale cloud, connected and managed as one environment. The most common end state, and usually the right one.
Why we quote in Australian dollars and why that matters
Hyperscaler list pricing for Australian regions is published in United States dollars. Your invoice converts at the prevailing rate, so infrastructure cost moves with the currency even when your usage does not change.
The Australian dollar moved roughly 7.9 per cent against the US dollar over the twelve months to August 2026, trading around 0.703. On a $2,000 monthly cloud spend, a movement of that size is meaningful and entirely outside your control.
A fixed monthly fee in Australian dollars removes that exposure. It is not the largest line in the comparison, but it is the one most business cases omit completely.
What's not included
- Licensing — Microsoft 365, RDS CALs, and Windows licensing are itemised separately.
- Customer-specific applications — patching and support cover the base operating system. Your line-of-business applications stay with their own vendor unless separately agreed.
- Individual file restores — backups cover recovery from internal failure. Individual restore requests are subject to fair use.
- Onboarding and migration — quoted and invoiced separately from the recurring fee.
- Hardware purchase — where colocation needs new hardware, that's your procurement. We'll specify it, but it doesn't sit inside the monthly fee.
- Application remediation — if an application needs rework to run properly, that's scoped separately, and we flag it at Stage 1 rather than discovering it at cutover.
The Cost Model — With the Assumptions Visible
Based on a sample environment of 45 users, 6 servers/VMs, 4TB storage, and 1TB monthly egress, on-premises hosting runs approximately $4,262/month, hyperscale cloud approximately $5,108/month, and the Avenue × Green Racks model approximately $2,869/month — around 44% below cloud, or roughly $26,867/year in savings. Most provider pages just assert a saving. This one shows the working, because a cost comparison only earns any trust if you can open it up and argue with the assumptions inside it.
Your environment
Data leaving the network. Most business cases guess this. Measure it.
Indicative monthly cost, AUD ex GST
Five-year total cost
Sixty months at the figures above, before any hardware refresh event in year five.
Where the cloud number comes from
Monthly breakdown at your inputs. Compute is rarely the largest share.
Every assumption in the model above
Currency. Hyperscaler list pricing is published in USD. Converted at AUD 1 = USD 0.7030, the rate on 7 August 2026. Green Racks figures are AUD and are not affected by currency movement.
Verified published rates used. Azure internet egress from an Asia Pacific region at USD 0.12 per GB for the first 10 TB, with 100 GB per month free. AWS at USD 0.09 per GB on the same tier. AWS gp3 block storage at USD 0.08 per GB-month. NAT gateway at USD 0.045 per hour plus USD 0.045 per GB processed.
Compute sizing. Modelled on an average instance of 4 vCPU and 16 GB at approximately USD 0.192 per hour on demand — not the largest instance in the estate. Sizing every VM at 8 vCPU / 32 GB is the most common way a comparison like this gets quietly rigged, and it inflates the cloud column by roughly double. If your average instance is genuinely larger, the real gap is wider than shown.
Storage. Blended at USD 0.06 per GB-month, between premium SSD and lower-tier storage, on the basis that a real estate does not put bulk data on its fastest tier. Backup modelled at 25 per cent of the storage line and disaster recovery at 25 per cent of compute, reflecting a cold standby rather than a warm duplicate environment.
Regional uplift. Published Asia Pacific region pricing runs approximately 20 to 30 per cent above US East. The model applies an uplift within that band. This is a modelled figure, not a quoted one.
On-premises. VMs consolidated onto hosts at four per host, host hardware amortised over 60 months with maintenance, storage amortised per terabyte, plus power, cooling, UPS, floor space and insurance as a facility overhead, plus your internal labour at the rate you set. It excludes the refresh event itself in year five, which is why the five-year on-premises figure is conservative rather than inflated.
Labour. Applied at your stated hours and rate to both on-premises and cloud. The managed model retains 35 per cent of those hours as residual internal oversight rather than dropping to zero — a comparison that assumes no internal effort after migration is not honest, and this is the single assumption most worth arguing with.
Green Racks. Derived from published managed hosting packages by user band, plus storage above the included allowance and additional server or VM instances. Colocation and GPU are not modelled here because they are quoted on power draw.
What this model does not include. Software licensing, which is broadly comparable across all three options and would add noise. One-off migration or onboarding cost. Application remediation. Reserved instance or savings plan discounts, which can reduce hyperscaler compute materially on a multi-year commitment and are the strongest counter-argument to the figures above.
Use this to frame a conversation, not to sign anything. Bring your actual egress volume and utilisation data to a scoping call and we will run it against real quotes from all three directions, including telling you when cloud wins.
When cloud is genuinely the better answer
A page that claims colocation always wins isn't worth reading, so here's when it doesn't:
Stay in hyperscale cloud when:
- The server room is ageing. That part isn't new.
- The cloud bill stopped shrinking. Egress, inter-zone traffic, and orphaned storage volumes are now permanent line items, not one-off surprises.
- Regulators and boards are asking sharper questions. Data residency has gone from a nice-to-have to something a tender or an auditor will actually test.
Move to this facility when:
- Demand is predictable and runs at a steady baseline
- Data residency is a contractual or regulatory requirement, not a preference
- Egress is a persistent, growing cost line rather than an occasional one
- A hardware refresh decision is due within the next twelve months
- You need physical access, or a specific hardware or licensing configuration
- Readiness Assessment
Is your environment ready to move?
Eight questions. They are the same ones we work through in the first twenty minutes of a scoping call, so answering them here means the call starts further along. Nothing is submitted anywhere — this runs in your browser.
1. Where is your primary hardware in its lifecycle?
2. Do you have a current, documented inventory of workloads and their dependencies?
3. When did you last complete a full restore test?
4. Is your workload demand predictable or spiky?
5. Do you know your actual monthly egress volume?
6. Do any of your workloads carry a data residency requirement?
7. Are documented recovery point and recovery time objectives agreed with the business?
8. Is there executive sponsorship and a budget owner for this change?
Your result appears here as you answer. It is an indication of migration readiness, not a quote.
FAQ
Frequently Asked Questions
What's the difference between colocation and managed hosting?
In colocation, you provide your own hardware, but Green Racks provides the infrastructure, power, cooling, and network connections, and you manage the operating systems and applications yourself. In the case of managed hosting (also called managed servers), you have no responsibility except for applications and data; Avenue and Green Racks own everything down to the operating system itself, including updates, monitoring, and backup.
How much does data centre hosting cost in Australia?
Avenue Group Australia offers managed hosting starting from $599 per month ex. GST for 1-10 users and going up to $2,490 per month for 50-100 users. Colocation, Sovereign GPU and enterprise hosting are individually quoted because they are priced based on rack units and power usage.
Is colocation cheaper than AWS or Azure?
Yes, but only in case of steady-state loads which are predictable, as cloud exit cost, gateway processing, and labor are not considered while computing costs. Colocation becomes expensive and even irrelevant in case of burst or seasonal loads; calculate both sides, do not make any assumptions.
What does data sovereignty mean, and is Australian data residency enough?
Data residency indicates that data will reside locally in Australia. Data sovereignty, on the other hand, indicates that the data will be regulated through Australian laws, not through any compulsion powers from abroad. Data residency by itself, with the Australian region selected in the cloud console, does not equate to data sovereignty, especially in cases where the cloud service provider is subject to another jurisdiction’s laws. It is all about who owns and runs the cloud.
How long does migrating to a data centre take?
A typical six-week duration will cover the process, with five phases of assessment, design, contracting, cutover, and stabilisation. The presence of multiple locations, bespoke application, or acquisition of hardware will stretch this duration.
What happens to our data during the cutover if something goes wrong?
A rollback plan will be written and agreed upon prior to the cutover, and maintained throughout the whole process. Test restore will be validated before the actual production cutover — not after.
Do we need to be Hosting Certification Framework certified?
This depends on the industry we operate in and the client base we serve. The best way would be to verify this through a scoping call.
Why not just refresh the hardware we have?
A hardware upgrade would solve the issue of aging hardware, but it would just reset the cycle of five years for capital expenditure and would ignore the cloud cost pressure. A hardware upgrade can actually become the very reason for adopting managed hosting.
Can we run a hybrid model instead of moving everything?
Indeed – the standard end state is hybrid placement, with steady-state workloads remaining within the facility, and elastic workloads remaining within hyperscale cloud, all as one inter-connected and managed environment.
What are the pitfalls of a migration like this?
The two key ones: assuming network egress, without proper measurement, and considering restore testing as optional. Both of these are specifically covered by Stage 1 and Stage 5 of the process laid out above.
Who owns the relationship if we came through our MSP?
Via the IT Partners program, you keep the client relationship, reselling on your own margin – Avenue does not compete for that relationship within the account.
How do we get out of it if it does not work out?
Speak about exit clauses as a part of the service agreement during Stage 3 of onboarding process, it's a legitimate thing to ask during the scope call.