SHARON AI & VAST Data launch sovereign AI supercluster in Melbourne

SHARON AI has partnered with VAST Data to provide enterprise and government organisations in Australia with scalable AI inference capabilities.

The collaboration has resulted in the launch of a sovereign supercluster at NEXTDC's Tier IV M3 data centre in Melbourne. This infrastructure integrates VAST Data's InsightEngine, aiming to deliver secure and production-ready AI solutions for industries such as financial services, public safety, and the public sector.

InsightEngine details

The VAST InsightEngine is designed as an end-to-end system for ingesting, embedding, indexing and retrieving data. It enables organisations to continuously process structured, unstructured, and streaming data in real time, supporting inference systems by delivering low-latency, parallel vector and hybrid searches. These capabilities are suited for Retrieval-Augmented Generation (RAG) and agentic workflows at scale.

Integrated with the VAST AI Operating System, InsightEngine enforces unified governance, security, and data lineage. The system incorporates policy-based access controls, encryption, and auditability, ensuring every query is secured and meets compliance requirements.

"As AI systems grow more capable, the ability to reason securely over large datasets in real time will define the next generation of enterprise intelligence," said Ofir Zan, AI Solutions & Enterprise Lead, VAST Data. "Together, SHARON AI and the VAST InsightEngine orchestrate event triggers and functions connected to data pipelines that scale complex multistep retrieval and reasoning workflows - all within a sovereign environment."

Industry applications

Bringing SHARON AI's cloud infrastructure together with InsightEngine is intended to move organisations from experimentation phases to production environments, supporting repeatable and enterprise-grade workflows. In the financial services sector, where throughput and latency are critical factors, the combined offering supports RAG at large scale. It utilises native vector indexing technology to search billions of embedded records, with the enforcement of fine-grained access permissions.

Within public safety and smart cities, the technology facilitates the ingestion and analysis of massive volumes of video and metadata in real time. This approach is designed to help organisations lower operational costs, enhance situational awareness, and improve incident response, all while ensuring sensitive data remains within national borders.

"By combining SHARON AI's sovereign GPU cloud with the VAST InsightEngine we're creating the foundation for enterprises and government institutions to run cutting-edge AI workloads locally, securely, and without compromise," said Wolf Schubert, CEO of SHARON AI. "With our supercluster now live in NEXTDC's Tier IV M3 data centre in Melbourne, this milestone demonstrates our commitment to delivering sovereign, high-performance AI infrastructure for Australia."

Academic research and collaboration

The first workloads on the new cluster have begun in partnership with the University of New South Wales (UNSW). Researchers and PhD students are leveraging SHARON AI's cloud to conduct AI research spanning several domains. Areas of focus include improving the reasoning abilities of small language models through structured reasoning and auto-formalisation, and developing novel post-tuning methods for Mixture-of-Experts architectures.

Other research efforts involve the fine-tuning and evaluation of state-of-the-art large language models-such as Falcon, Llama, Qwen, and Deepseek-in parallel, targeting specific tasks like question-answering with applications in mathematics and spatio-temporal reasoning. Additionally, UNSW researchers are working on accelerating global weather forecasting by training high-resolution, data-driven models with large-scale ERA5 datasets to achieve faster and more accurate predictions.

Collectively, these projects are investigating how specialised post-tuning, fine-tuning, and GPU-accelerated model architectures can drive improvements in reasoning performance, scalability, and broader domain applications of AI. The research undertaken on this infrastructure aims to enable the development of smaller, more efficient reasoning models with implications for science, forecasting, and future AI evaluations.