New AI Infrastructure Enhances Real-Time Decision-Making
As artificial intelligence (AI) transitions from digital platforms to the physical realm—boosting robots, sensors, and instant decision-making—the need for quick and dependable computing close to data sources is increasing. Traditional cloud-based solutions prove insufficient for applications that require immediate responses and high safety standards, prompting a shift in how AI infrastructure is developed and positioned.
In response, Australia’s national science agency, CSIRO, has introduced a groundbreaking infrastructure located at their Queensland Centre for Advanced Technologies (QCAT) in Pullenvale. This innovative setup, named Vetra, focuses on bringing powerful AI processing right next to the machines and sensors that need it the most.
Vetra stands out by offering high-performance AI computing in a compact, modular, and environmentally friendly design. Unlike conventional cloud services that operate from afar, Vetra enables rapid on-site processing, allowing robots and sensing systems to react swiftly, learn continuously, and navigate complex environments more safely.
“Edge AI” refers to the practice of running AI systems close to the data source rather than sending that data to distant servers. Liming Zhu, Director of CSIRO’s Data61, noted that Vetra provides a reliable and local option for AI computing. “AI is quickly moving into real-world applications such as robotics and critical safety systems,” he explained. “Vetra’s proximity to data allows for immediate responses, learning, and operations that remote systems simply cannot achieve.”
The facility is positioned next to Australia’s largest robotics research center, enabling AI systems to gather insights from real-world experiments rather than relying solely on simulations. Dr. Peyman Moghadam, head of CSIRO’s Embodied AI Cluster, emphasized that Vetra complements larger supercomputing efforts in Canberra by managing immediate, local processing before passing data on for extensive analysis.
“Robots must learn from the real world, not just online datasets,” Dr. Moghadam added. “Vetra fills the crucial gap in our workflow, transforming real-world robotics data into more effective and safer AI systems.”
Cooling remains a significant concern for advanced AI systems due to the heat they generate. Vetra’s design aims to lessen environmental impact by employing carbon dioxide-based cooling methods and closed-loop liquid cooling systems, greatly reducing reliance on water. During regular operation, the infrastructure is expected to save about 225 tonnes of carbon dioxide emissions annually, equivalent to removing 50 cars from Queensland’s roads.
CSIRO’s Chief Technology Officer, Angus Macoustra, shared that sustainability was a core consideration from the beginning. “Vetra demonstrates how high-tech solutions can minimize water usage and emissions,” he commented.
The launch of Vetra was backed by support from Australian small and medium-sized businesses, including Oper8 Global and XENON, as well as global technology partners. With 48 high-performance graphics processing units (GPUs), Vetra can conduct thousands of complex calculations simultaneously on large datasets. The infrastructure is also designed to grow with future demands in research and industry.
This initiative reinforces CSIRO’s commitment to developing effective research systems that support safe and responsible AI in Australia, particularly across sectors such as energy, infrastructure, environment, and advanced manufacturing.
