GreenSquare Opens 1995 IBM Data Center for 8,200 NVIDIA GPUs

The GreenSquare SYD1 data center in Sydney, Australia, has entered operation after an eight-month conversion of a facility originally opened in 1995. The first 15MW stage of the former IBM data center is now ready for high-density AI computing, with approximately 8,200 NVIDIA Blackwell Ultra GPUs planned for deployment by Australian cloud provider Sharon AI.

Announced on October 19, 2026, the project demonstrates how older data-center infrastructure can be adapted for modern artificial intelligence workloads. GreenSquare increased IT capacity within the original building footprint by 275%, introduced liquid-to-chip cooling and retained much of the existing structure instead of replacing the building.

GreenSquare SYD1 Data Center: What Has Opened?

GreenSquare confirmed that the first stage of its SYD1 campus in Norwest Business Park, Western Sydney, is operational.

The building previously served IBM and later Kyndryl. Its conversion involved extensive changes to electrical distribution, mechanical infrastructure and cooling, preparing the site for substantially higher computing densities than its original configuration supported.

Sources: GreenSquare’s October 19 announcement, company infrastructure disclosures and independent reporting by Data Center Dynamics. Planned GPU capacity and projected efficiency should not be interpreted as measured operating results.

How a 1995 Data Center Was Rebuilt for Modern AI

The most important engineering decision was to retain the original structural shell rather than demolish the building and start again.

GreenSquare preserved the existing slab and much of the structure while relocating its data halls to the ground floor. According to the company, the revised configuration accommodates high-density computing equipment without additional structural reinforcement.

Earlier reporting identified the original facility as a 4MW data center. The new 15MW first stage represents approximately 3.75 times that capacity. This comparison illustrates the scale of the conversion, although capacity alone does not establish how many GPUs are already running.

Construction began in February 2026. GreenSquare worked with Erilyan on the building and Fredon on electrical infrastructure and commissioning. The company reported more than 350,000 working hours without a lost-time injury.

Why Liquid Cooling Matters for 8,200 NVIDIA GPUs

GreenSquare designed the facility to support NVIDIA HGX B300 systems using Blackwell Ultra accelerators. These high-density systems require cooling infrastructure capable of transferring heat away from computing components during demanding AI workloads.

Liquid-to-chip cooling addresses this challenge by moving heat through a liquid circuit connected to the computing hardware. The approach also requires appropriate coolant distribution, monitoring and maintenance infrastructure.

SYD1 replaces the older building’s cooling towers with a closed-loop design. GreenSquare expects the system to operate without evaporative cooling for approximately 96% of the year.

Rainwater harvesting is expected to offset about 10% of annual cooling water requirements. These figures are company projections rather than independently verified annual operating results.

Operational Facility Does Not Mean All GPUs Are Installed

This distinction matters when evaluating new AI infrastructure. A commissioned facility provides the physical environment needed to deploy servers, but actual computing availability depends on hardware installation, networking, testing and customer activation.

The development also adds context to NVIDIA’s broader Australian AI infrastructure expansion plans , which include Sharon AI among the participating companies.

Unlike a gigawatt-scale development target, the SYD1 announcement identifies a specific completed facility stage. Its actual GPU utilization and production workload capacity remain separate questions.

Energy Efficiency and the Environmental Impact

GreenSquare reports a projected power usage effectiveness, or PUE, of 1.20. PUE measures total facility electricity consumption relative to the electricity delivered to IT equipment.

For illustration, a facility operating at a sustained 15MW IT load with a PUE of 1.20 would consume approximately 18MW in total, including 3MW for cooling and other overhead. This is a theoretical calculation, not a measurement of SYD1’s current power consumption.

An assessment by engineering consultants TTW, described in GreenSquare’s structural carbon analysis , found that the redevelopment retained 92% of the existing building’s embodied structural carbon.

Retaining existing concrete and structural materials can avoid some of the emissions associated with replacement construction. However, the environmental performance of the operating facility will also depend on electricity consumption, utilization and the energy supply.

What This Means for Data Center Operators

SYD1 offers a practical example of adaptive reuse as an alternative to building every AI facility from the ground up. Its delivery suggests that suitable existing structures can provide a foundation for substantial infrastructure upgrades.

The approach is not automatically transferable to every older data center. Structural capacity, electrical connections, cooling design and available expansion space must still be evaluated individually.

The distinction between available infrastructure and installed computing hardware is equally important. As previously examined in Servers.News coverage of Applied Digital’s operational capacity expansion , completed data-center space does not by itself reveal the utilization of customer computing equipment.

What Happens Next for GreenSquare SYD1?

The next important milestone is the installation and commissioning of Sharon AI’s planned NVIDIA Blackwell Ultra infrastructure.

GreenSquare also intends to expand the campus. The New South Wales Government’s Investment Delivery Authority has endorsed support for a proposed 80MW second stage. That expansion should not be confused with operational capacity already delivered.

The immediate achievement is narrower but significant: an existing data center from 1995 has been transformed into an operational 15MW facility designed for contemporary AI computing.

For the wider industry, SYD1 demonstrates one potential route to expanding AI infrastructure through the redevelopment of suitable existing assets rather than relying exclusively on new construction.