Five operators build or run GPU-first neocloud capacity on Australian soil today: Firmus, Sharon AI, IREN, ResetData and Polaris, with Macquarie Data Centres hosting the tier rather than operating in it and SCX competing from outside the GPU model on SambaNova ASICs. The committed pipeline exceeds 1,600MW, anchored by Firmus’s 1.6GW Project Southgate, and IREN has added an 800MW campus at Bundey in South Australia that energises from 2028. Power availability and grid connection set the pace at which that capacity comes online.
What is a neocloud?
A neocloud is a cloud provider built from the ground up around GPU compute for artificial intelligence and high-performance computing workloads. Unlike hyperscalers such as AWS, Microsoft Azure and Google Cloud, which offer broad service portfolios spanning storage, databases, enterprise software and compute, neocloud operators focus almost entirely on deploying tightly integrated clusters of advanced GPUs, high-bandwidth memory, low-latency networking and efficient cooling.
The term has gained traction globally since 2024 as demand for AI training and inference infrastructure outpaced what enterprises could procure cost-effectively from the major public clouds. Equinix defines a neocloud as “a vendor that offers AI-specific infrastructure and services”, noting that because GPUs are the dominant processor type in AI today, most neoclouds specialise in delivering GPU-as-a-Service (GPUaaS). DriveNets describes them as providers that primarily offer GPU-as-a-Service and emerged in response to the AI gold rush. ABI Research forecasts that North America accounts for 88% of total neocloud GPUaaS revenue in 2026, with that share dropping to 72% by 2030 as other regions, including Asia-Pacific, scale sovereign cloud capacity.
In Australia and New Zealand, the Integral Advice 2025 Neocloud Study found that 84% of organisations surveyed currently use GPUaaS or intend to do so by 2027. The study identifies three emerging provider types, infrastructure players, platformers and aggregators, and forecasts a 17% compound annual growth rate in enterprise compute demand through 2027.
Hyperscalers in Australia
Australia’s hyperscale layer is anchored by the three global cloud majors, the tier neoclouds are now building alongside. Amazon Web Services operates its Sydney and Melbourne regions and committed A$20 billion to Australian data centre infrastructure through 2029, announced in June 2025. Microsoft Azure runs Australia East in Sydney and Australia Central in Canberra, and added AI capacity at its Kemps Creek campus in Western Sydney. In April 2026 Microsoft committed A$25 billion to Australian AI infrastructure, security and skills over the same window, the largest single corporate technology investment Australia has received. Google Cloud operates Sydney and Melbourne regions and funds local infrastructure and renewables through its A$1 billion Digital Future Initiative, announced in November 2021. We set out the AWS and Microsoft pledges in detail in our AFR AI Summit 2026 preview.
Rather than own most of their floor space, these hyperscalers lease heavily from local operators such as AirTrunk, NEXTDC and CDC, the same Certified Strategic operators that anchor Australia’s sovereign capacity. The neocloud tier now competes with them for that capacity, for grid connections, and for the GPU supply NVIDIA confirmed Australia can access as a sovereign deployment market, a shift we covered in our NVIDIA GTC 2026 recap.
Who are the neocloud players in Australia?
Five operators are actively building or operating GPU-first neocloud capacity on Australian soil. Their approaches, capital structures and facility partnerships differ significantly.
Firmus Technologies
Firmus Technologies is the largest-capitalised neocloud play in Australia. Through Project Southgate, a strategic alliance with CDC Data Centres and NVIDIA, the company has committed to 1.6GW of planned AI factory capacity across Australia, with first stages under construction in Tasmania and Melbourne. The Melbourne deployment is built around approximately 18,500 NVIDIA GB300 GPUs under a multi-billion-dollar customer contract, which Firmus describes as the first hyperscale AI factory order in the country. Firmus has said the first 150MW stage across its two lead sites, about 54,000 GB300 GPUs in total, is targeted for delivery by mid-2026, with the greenfield Tasmanian campus set to host roughly 36,800 GB300 GPUs by late 2026. The programme is backed by a US$10 billion debt facility led by Blackstone and Coatue. Firmus had signalled an ASX listing during 2026, but reporting in May 2026 indicated the float was pushed beyond September after its pre-IPO pitch landed poorly with institutions, and no firm listing date has been set.
Sharon AI
Sharon AI (Nasdaq: SHAZ) is the first Nasdaq-listed neocloud operator with live GPU infrastructure inside an Australian Certified Strategic facility. The company deployed 1,024 NVIDIA B200 GPUs at NEXTDC’s M3 data centre in Melbourne, and its disclosed Australian AI factory capacity now stands at 132MW, of which 102MW is contracted, with more than 55,000 GPUs due by mid-2027. In February 2026, Sharon AI partnered with Cisco and NVIDIA to launch Australia’s first Cisco Secure AI Factory at NEXTDC’s S3 data centre in Sydney, running 1,024 Blackwell Ultra GPUs. In June 2026 it announced an oversubscribed US$1.6 billion private placement anchored by Situational Awareness and Oaktree, and it is pursuing an ASX dual listing via CHESS Depositary Interests.
IREN
IREN (Nasdaq: IREN) is the Sydney-founded operator that built its AI cloud business in North America and is now committing capacity at home. On 3 June 2026 the company signed a transmission connection agreement for an 800MW campus at Bundey in regional South Australia, its first announced Australian project, with energisation targeted from 2028. The grid deal secures four 330kV feeder exits sized for the full 800MW load without network upgrades, anchored to ElectraNet’s proposed Northern Transmission Project. IREN holds a multi-billion-dollar AI cloud contract with Microsoft and puts its total development pipeline at around 5GW. Bundey broadens the national pipeline beyond the Sydney basin and ranks among the largest single-site loads disclosed in the country.
Polaris Data Centre
Polaris Data Centre, a Brisbane operator at Springfield Central that describes itself as Queensland’s largest privately owned data centre, has branded its own NeoCloud platform offering AI-ready, GPU-optimised infrastructure. The Australia Data Centre Index lists Polaris as a Certified Strategic enclave under the Hosting Certification Framework. Polaris positions NeoCloud as a scalable platform with hybrid and multi-cloud integration, built-in compliance and disaster recovery, and on-demand GPU resources for AI and machine learning workloads.
ResetData
ResetData is the operating tier of an integrated PropCo and OpCo neocloud structure built with Centuria Capital Group (ASX:CNI), which acquired 50% of ResetData in August 2024 for up to A$21 million, with a call option for the remaining 50% from year five. ResetData runs AI-F1, which it describes as Australia’s first sovereign public AI factory, inside Centuria Office REIT’s 818 Bourke Street tower in Melbourne’s Docklands. The deployment is built around roughly 1,024 NVIDIA H200 GPUs with liquid immersion cooling, and ResetData is replicating the model across Centuria’s managed commercial property portfolio nationally.
Operator | Type | Australian capacity (disclosed) | Host facility | Backing |
Firmus Technologies | Neocloud / AI factory | 1.6GW planned (Project Southgate); first 150MW stage under construction | CDC | Blackstone, Coatue, NVIDIA |
Sharon AI (Nasdaq: SHAZ) | Neocloud | 132MW disclosed, 102MW contracted; B200 and Blackwell Ultra live | NEXTDC M3, S3 | Oaktree, Situational Awareness; NVIDIA Cloud Partner |
IREN (Nasdaq: IREN) | AI cloud / campus | 800MW Bundey campus, energising from 2028 | Self-developed (Bundey, SA) | Nasdaq-listed; NVIDIA, Microsoft cloud contract |
ResetData | Neocloud / AI factory | AI-F1 live, around 1,024 H200 GPUs; national rollout planned | Centuria Office REIT (818 Bourke St) | Centuria Capital Group (ASX: CNI) |
Polaris Data Centre | Neocloud platform | Brisbane (Springfield Central) | Self-hosted | Springfield City Group |
Macquarie Data Centres | AI factory host (enabling) | IC3 Super West, 47MW (opens Sept 2026) | Macquarie Park, Sydney | Macquarie Technology Group (ASX: MAQ) |
Source: Certified Strategic Editorial, primary company disclosures, June 2026. Figures reflect announced or planned capacity and are subject to staged delivery.
Also in the market
Other operators serve the Australian neocloud tier without the same disclosed local footprint. DUG Technology (ASX: DUG), headquartered in Perth, runs the immersion-cooled DUG McCloud high-performance computing platform and modular DUG Nomad units that support NVIDIA H200 GPUs. International on-demand GPU clouds, including Vultr, offer NVIDIA A100 and H100 instances from Australian regions for developers who want pay-as-you-go compute without long-term commitments or a dedicated local facility.
SCX (SouthernCrossAI)
Beyond the GPU-first five sits SCX, which operates what it describes as Australia’s first sovereign AI cloud, built on SambaNova’s reconfigurable dataflow ASICs rather than GPUs. SCX brought its first sovereign inference node online in Sydney in January 2026 and is rolling out nationally across Equinix facilities through 2026 and 2027, incorporating SambaNova’s SN50 chip from the second half of 2026. SCX processes all models, data pipelines and system logs within Australian jurisdiction, and by its own account targets up to 10x greater energy efficiency than conventional GPU clusters for inference workloads. It is the clearest example of the neocloud tier specialising in inference rather than training.
Macquarie Data Centres (enabling infrastructure)
Macquarie Data Centres is not a neocloud operator, but it has explicitly identified neocloud as a target use case for its IC3 Super West facility in Sydney, a 47MW data centre hosting the Dell AI Factory with NVIDIA. Macquarie describes the facility as supporting enterprise AI, private AI and neocloud workloads, and positions it as sovereign, government-grade infrastructure. Macquarie operates several Certified Strategic facilities, though IC3 Super West is not yet certified on the Australia Data Centre Index. It topped out in late 2025 and is on track to open in September 2026, the third site on Macquarie’s Macquarie Park campus and part of a roughly 200MW Sydney development plan.
Why this matters for Australian infrastructure
Neocloud operators are signing multi-megawatt, long-term leases that change the economics of Australian colocation. Firmus alone has committed to 150MW at its Melbourne facility with a single hyperscale customer, backed by multi-billion-dollar contracts and institutional-grade debt facilities rather than speculative options. For colocation providers, neocloud tenants represent a demand tier sitting between traditional enterprise and hyperscaler, requiring higher power density, liquid cooling and direct connectivity to GPU fabric.
The demand that has converted into disclosed contracts is commercial: Sharon AI’s US$1.25 billion ESDS agreement and its Canva lighthouse deal, and Firmus’s single hyperscale customer at Melbourne, all won on price, GPU availability and speed.
The connectivity dependency
AI workloads at scale require resilient, high-capacity connectivity between clusters, not just compute. SUBCO’s SMAP hypercable, delivering more than 400Tb of inter-capital capacity by May 2026, serves this need directly, and SUBCO has confirmed neocloud providers are among its anchor customers. The diverse routing between Perth, Adelaide, Melbourne and Sydney reduces a single-point-of-failure risk for operators running distributed GPU clusters across multiple Australian cities. Firmus has gone further, underwriting a new SUBCO subsea cable to Tasmania to wire Project Southgate’s flagship campus to the mainland.