The GPU-Free AI Factory Inside Sydney's Most Connected Data Centre

SCX.ai switched on a production AI inferencing node inside Equinix's SY5 International Business Exchange at 200 Bourke Road, Alexandria, Sydney on 28 January 2026. The node uses a SambaNova ASIC-accelerated architecture and delivers AI inferencing at a claimed 10x efficiency improvement over GPU-based systems. All processing occurs within Australian borders with no data transit to offshore facilities.

The 10x efficiency figure is SCX.ai's own claim from the launch release. The company has not published its benchmark methodology or comparative dataset. Independently, hardware analysis from HowAIWorks.ai finds that purpose-built ASICs can deliver 10x to 100x better efficiency than GPUs for their target inference workload profiles, which gives the directional claim credibility.

Guy Danskine, Managing Director Australia at Equinix, confirmed SY5 is purpose-built for next-generation compute workloads, with access to 305+ cloud providers via Equinix Fabric, direct connections to AWS, Microsoft Azure, Google Compute, and Oracle Cloud, and the largest network peering platform in the Australian market. SCX.ai founder and CEO David Keane described the deployment as "enterprise-grade AI infrastructure that doesn't require massive water consumption for cooling and operates at a fraction of the energy cost of GPU-based systems."

The Water Dimension

The zero-water-cooling claim requires one important clarification. SY5 as a building uses Indirect Evaporative Cooling Air Handling Units (IDEC) across its white space and UPS switch room. The SCX.ai node eliminates water dependency at the rack and node level, the inference hardware itself does not require water-intensive cooling. This is a meaningful distinction from GPU-based deployments, which require active liquid cooling or large evaporative systems at the hardware level, but SY5 as a facility is not water-free.​

The context for why this matters is significant. A typical 100MW hyperscale data centre consumes approximately 2 million litres of water per day. Large hyperscale AI facilities can consume up to 40 million litres per day -- enough for around 80,000 Australian homes. Australia's total data centre sector currently consumes 5.5 gigalitres of water per year, forecast to reach 17 gigalitres by 2030, a more than three-fold increase in four years driven almost entirely by AI workload growth.

Sydney's data centre cooling water demand alone is projected to exceed the volume of Canberra's entire drinking water supply within the next decade. The NSW Legislative Council inquiry, with submissions closing 27 March 2026, has water resource impact listed as a formal scope item. An operational zero-water-at-node AI deployment inside Australia's most interconnected facility is a direct, working counterpoint to the assumption that AI inference is structurally water-intensive.

Why SY5 as the Location

Equinix SY5 is Sydney's most interconnected data centre campus, described in its own technical specifications as offering the "broadest range of cloud services and the largest collection of international and regional network service providers in Australia." It connects directly to the Hawaiki Cable, Southern Cross Cable, and PIPE Pacific Cable subsea systems and hosts the Equinix Internet Exchange, the largest network peering platform in the Australian market.​

For AI inferencing specifically, this matters for latency. Processing that happens locally rather than routing to offshore data centres reduces round-trip time for real-time AI applications in ways that are operationally non-negotiable for financial services, healthcare, and government workloads. SY5 also carries the DTA Certified Strategic designation.​

Why This Matters

• Zero water at node level inside a Certified Strategic facility. SY5 carries the DTA Certified Strategic designation. Running AI inference there with no water dependency at the hardware level is a direct response to the water consumption concerns driving the NSW inquiry.

• Water and energy context makes this timely. Australia's data centre water demand is forecast to reach 17 gigalitres per year by 2030. A zero-water-at-node inference deployment demonstrates that AI workloads do not require water consumption as a structural feature.

• SY5's interconnection density removes the usual colocation trade-off. 305+ cloud providers, direct hyperscaler on-ramps, and the country's largest peering platform mean the node is reachable from every major enterprise network in Australia.

Analysis written by CertifiedStrategic Editorial Team

CertifiedStrategic.com  - Australia's independent data centre index tracking capacity, certification and market news across the country's critical infrastructure providers.