Sharon AI expands Australia’s sovereign compute with major GPU buildout

Sharon AI has unveiled plans to deploy a 1,000-unit NVIDIA B200 GPU cluster at NEXTDC’s Tier IV M3 facility in Melbourne, a move that sharply increases domestic capacity for training and running large-scale artificial intelligence models within Australia’s borders. The announcement places the neocloud provider among a small group globally able to operate Blackwell-generation systems at this scale and underscores intensifying competition to localise high-end compute amid tightening data-sovereignty expectations.

The buildout will be housed at NEXTDC’s M3 data centre, a Tier IV site designed for fault tolerance and high availability, and is scheduled to be commissioned in phases as power, cooling and network upgrades are completed. Sharon AI said the cluster is intended to serve enterprises, research institutions and public-sector workloads that require low-latency access, predictable pricing and controls aligned with domestic governance rules.

The deployment centres on NVIDIA’s B200 accelerators, part of the Blackwell architecture, which are engineered for mixed-precision AI workloads and tightly coupled networking. Compared with prior-generation parts, B200 systems are designed to deliver substantial gains in training throughput and energy efficiency, attributes that matter in Australia where power costs and grid constraints shape data-centre economics. Sharon AI has indicated the cluster will be integrated with high-speed interconnects to support model-parallel training at scale, alongside managed software stacks for popular AI frameworks.

Executives at Sharon AI framed the investment as a response to mounting demand from customers seeking to keep sensitive data and model weights onshore while avoiding long waits for capacity in overseas hyperscale regions. The company positions itself as a “sovereign neocloud,” a label increasingly used by providers that combine cutting-edge hardware with contractual and operational assurances around data residency, access controls and auditability.

The Melbourne installation also reflects broader shifts in Australia’s digital-infrastructure landscape. Data-centre operators have accelerated expansion across Victoria and New South Wales, driven by cloud adoption, streaming, fintech and now AI. Tier IV facilities, once rare, are gaining traction for mission-critical workloads that cannot tolerate downtime. NEXTDC’s M3 site, which Sharon AI selected for its redundancy and connectivity, sits within a network of carrier-dense campuses that can support the bandwidth demands of distributed AI training.

While Sharon AI has not disclosed the capital outlay, industry analysts estimate that a 1,000-unit Blackwell-class cluster represents an investment running into the hundreds of millions of dollars once servers, networking, power and cooling are accounted for. The economics hinge on utilisation rates and long-term contracts with anchor tenants, a model that mirrors how hyperscalers underwrite their own capacity builds.

Competition in the neocloud segment is intensifying as global demand for advanced GPUs outpaces supply. Providers in North America, Europe and parts of Asia have announced similar Blackwell-based expansions, often citing wait times for access to top-tier accelerators. By committing early to a large cluster, Sharon AI aims to secure allocation and lock in customers planning multi-year AI roadmaps, from foundation-model training to inference at scale.

Regulatory and policy considerations add another layer of momentum. Governments and regulated industries are scrutinising cross-border data flows and the concentration of AI compute in a handful of foreign regions. Local capacity offers a hedge against geopolitical risk and export controls, while also supporting national ambitions to foster home-grown AI research and commercialisation. Universities and startups, in particular, have argued that access to domestic high-performance compute is critical to retaining talent and intellectual property.