Australia’s push to keep critical AI workloads onshore is gaining another infrastructure layer. SCX.ai and DDN have partnered to combine ASIC-accelerated AI inference infrastructure with DDN’s Infinia data platform, targeting Australian enterprises, government agencies and research organizations that need high-throughput AI without sending models, workloads or data offshore.

Australia’s sovereign AI ambitions are increasingly becoming an infrastructure question.

Building or accessing powerful AI models is only part of the equation. Organizations also need compute, high-speed data infrastructure, energy-efficient systems and enough capacity to run increasingly demanding inference workloads without compromising data residency or regulatory requirements.

SCX.ai and data infrastructure company DDN are targeting that problem with a new partnership designed to expand what the companies describe as Australia’s largest sovereign AI inference cloud.

The collaboration combines SCX.ai’s ASIC-accelerated infrastructure with DDN Infinia, a data intelligence platform designed to provide high-performance access to data for AI workloads. The resulting infrastructure is intended to function as a multi-tenant AI factory for Australian enterprises, government organizations and research institutions.

The central proposition is straightforward: keep AI workloads and data in Australia while improving the economics and performance of inference.

That is becoming more important as inference — the process of running trained AI models to produce outputs — increasingly becomes the dominant operational workload for many AI applications.

Australia’s sovereign AI infrastructure push

SCX.ai’s announcement comes days after the company listed on the Australian Securities Exchange, following a fully underwritten A$40 million IPO, according to the company.

The listing makes SCX.ai Australia’s first ASX-listed pure-play sovereign AI inference infrastructure company, the company says.

SCX.ai reported A$6.5 million in contracted annual recurring revenue at the end of July, representing a 20.9% increase from May. It also says its platform has more than 400 active users.

Those figures are company-reported rather than independently verified, but they provide an indication of the commercial traction SCX.ai is seeking to build around sovereign inference.

Its first AI node is already operational at Equinix SY5 in Sydney, while the company is targeting the end of 2026 for a second node.

The strategy is part of a wider global trend toward sovereign AI infrastructure.

Governments increasingly want sensitive government, healthcare, financial and research data processed within national or trusted jurisdictions. Europe, the Middle East and parts of Asia are pursuing similar strategies, creating demand for domestic AI compute rather than relying entirely on hyperscale cloud providers.

The data bottleneck behind AI inference

Adding AI accelerators is only one part of building an inference cloud.

Large AI models generate substantial data movement between compute, memory and storage. As context windows become longer and AI agents perform increasingly complex multi-step tasks, the data layer can become a bottleneck.

This is where DDN’s Infinia platform enters the architecture.

DDN says Infinia can deliver sub-millisecond latency and up to 27x faster KV-cache loading in applicable workloads. KV cache stores previously processed attention information so an AI model does not have to recompute it repeatedly during inference.

Faster access to that cache can matter particularly for applications involving long context windows, high concurrency and agentic AI.

The practical goal is to keep expensive AI accelerators working rather than waiting for data.

That distinction is important because inference economics are ultimately tied to utilization. A highly capable accelerator sitting idle while waiting for data can become an expensive underused asset.

Moving beyond GPU-centric infrastructure

SCX.ai is also pursuing an alternative to conventional GPU-heavy AI infrastructure.

Its platform uses SambaNova Systems’ AI processors, which are purpose-built for AI workloads. SCX.ai says testing conducted ahead of its ASX listing showed its SambaNova-based infrastructure achieving approximately 2.5x to 5.6x higher performance per watt than GPU-based systems across selected stable inference workloads.

Those figures should be treated carefully. Performance-per-watt comparisons can vary substantially depending on model architecture, workload, batch size, precision, software stack and hardware configuration. They are not equivalent to a universal performance advantage over GPUs.

The broader idea, however, is increasingly relevant.

AI infrastructure providers are experimenting with GPUs, ASICs and other specialized accelerators because the economics of inference differ from those of training. Training large models can justify extremely expensive accelerator clusters, while inference providers need to optimize the cost of processing potentially billions of requests.

Power efficiency therefore becomes a commercial metric as much as an engineering one.

AI factories need power and cooling

The infrastructure challenge extends beyond processors.

AI data centers consume significant amounts of electricity and require sophisticated cooling systems. SCX.ai says its architecture is designed to produce more AI compute within the existing power and physical footprint of commercial data centers.

The company also emphasizes air cooling, contrasting its approach with data-center designs that use substantial quantities of water for cooling.

That could become relevant in Australia, where data-center operators face a combination of power availability, grid constraints, land requirements and environmental considerations.

Efficient inference hardware could potentially allow more AI capacity to be deployed without proportionally expanding the physical footprint or power envelope.

The question for customers will be whether those theoretical infrastructure advantages translate into predictable cost-per-token improvements under production workloads.

Project MAGPiE and the case for local models

The partnership will support enterprise workloads as well as Project MAGPiE, SCX.ai’s sovereign large language model designed to be fine-tuned for Australian cultural and commercial contexts.

That reflects another dimension of sovereign AI.

Sovereignty is not simply about putting an overseas model inside an Australian data center. Organizations may also want greater control over model development, training data, deployment and customization.

A domestic inference infrastructure layer can therefore support both locally developed models and international models that organizations are permitted to run within the country.

For government and regulated industries, that distinction can be significant.

From national AI ambition to operational capacity

Australia is not alone in trying to establish sovereign AI infrastructure. Governments globally are attempting to secure access to compute, data and models as AI becomes strategically important.

The challenge is making sovereignty economically sustainable.

An AI cloud that is secure but significantly more expensive or less performant than offshore alternatives may struggle to attract commercial workloads. Conversely, infrastructure that combines data residency with competitive performance and cost could become a meaningful component of national digital infrastructure.

That is ultimately what SCX.ai and DDN are attempting to demonstrate.

The partnership combines specialized inference hardware with a high-performance data layer, while SCX.ai’s planned network of nodes provides the geographic and operational foundation for an Australian AI cloud.

Node 2 is targeted for deployment by the end of 2026, with additional locations planned.

If SCX.ai can scale that network while maintaining competitive utilization and economics, its significance could extend beyond another cloud infrastructure launch.

It would represent a test of whether sovereign AI can be operated as commercially viable infrastructure rather than simply treated as a national policy objective.

Market Landscape

The sovereign AI market is developing around four interconnected requirements: compute sovereignty, data sovereignty, model sovereignty and operational economics.

SCX.ai is primarily targeting the infrastructure layer, competing indirectly with global hyperscalers such as AWS, Microsoft Azure and Google Cloud, as well as specialized GPU-cloud providers and emerging sovereign AI infrastructure companies.

Its differentiation rests on:

• ASIC-based inference: Using SambaNova processors rather than relying exclusively on GPUs.
• High-performance data infrastructure: DDN Infinia is designed to reduce I/O bottlenecks.
• Australian data residency: Workloads and data remain within Australia.
• Multi-tenant infrastructure: Designed for enterprises, government and research organizations.
• Energy efficiency: The company emphasizes performance per watt and existing data-center footprints.
• Local AI development: Project MAGPiE provides a foundation for Australia-specific model development.

The competitive test will be cost per token at sustained production scale. Sovereignty creates value, but customers still need performance, availability and predictable economics.

Top Insights

• SCX.ai and DDN are combining ASIC-based inference infrastructure with high-performance data services to expand Australia’s sovereign AI cloud capacity.
• DDN Infinia targets AI data bottlenecks with low-latency access and faster KV-cache loading, potentially improving accelerator utilization for demanding inference workloads.
• SCX.ai’s sovereign model keeps AI workloads and data within Australia, addressing residency, regulatory and national infrastructure requirements for sensitive organizations.
• SambaNova-based infrastructure could reduce inference power requirements, although independent production testing will be needed to validate the company’s performance-per-watt claims.
• The planned expansion from Sydney’s first node toward a national network will test whether sovereign AI infrastructure can achieve commercial scale and competitive cost-per-token economics.

Power Tomorrow’s Intelligence — Build It with TechEdgeAI