AI data centers are increasingly being built on abandoned industrial sites rather than greenfield locations, a shift that dramatically accelerates construction timelines while leveraging existing infrastructure. The trend reflects a fundamental change in how technology companies and governments approach the massive infrastructure demands of artificial intelligence, moving away from building from scratch toward retrofitting sites that already have the power, water, and zoning approvals in place.

Why Are Old Industrial Sites Suddenly Valuable for AI?

The Ettamogah mill near Albury, New South Wales, illustrates this emerging pattern. The former newsprint facility, which operated under Norske Skog before closing in 2019, sat largely idle after packaging company Visy purchased it for A$85 million in 2020 as a transport and warehouse depot. Now, Nasdaq-listed Sharon AI has acquired the site to convert it into a high-performance computing hub for AI training and inference workloads.

The appeal is straightforward: paper manufacturing is one of the most energy and water-intensive industries on the planet. A facility designed to run massive industrial machinery already has the heavy-duty infrastructure that AI data centers desperately need. Instead of spending years on environmental impact studies, zoning approvals, and infrastructure trenching, developers can skip most of those hurdles by using a brownfield site that is already zoned for heavy industry.

What Infrastructure Advantages Do Repurposed Mills Offer?

Modern AI data centers require three critical resources: electricity, water for cooling systems, and high-speed internet connectivity. The Ettamogah site provides all three. Because it previously powered massive paper manufacturing equipment, it is already connected to the high-capacity power grid needed to sustain intensive computational loads. The facility also features existing water treatment facilities and recycled water systems that can support closed-loop cooling, meaning the data center can process and recycle cooling water on site rather than constantly drawing from local potable supplies.

Location matters too. The mill sits close to major fiber optic internet infrastructure along the Hume Highway corridor, well away from residential suburbs where noise and traffic would create community friction. This combination of factors means Sharon AI can move from acquisition to operational deployment far faster than building a new facility from dirt.

How Are Governments Managing This Rapid Expansion?

The acceleration of AI data center development is not happening in a regulatory vacuum. In the United States, the Trump administration has taken an aggressive approach to streamlining federal approval processes. President Donald Trump's July 2025 executive order directed federal agencies to make federal sites available for qualifying data center projects and encouraged faster environmental reviews and permitting.

The order defines qualifying projects as facilities requiring more than 100 megawatts of new AI-related load, including training, inference, simulation, and synthetic data generation workloads. The scope is broad enough to encompass developments committing substantial capital, adding more than 100 megawatts of power demand, protecting national security, or receiving an agency designation.

In Australia, New South Wales has released specific data center guidelines to manage this growth. Those guidelines explicitly note that there are far fewer complexities and community impacts when developers choose existing brownfield sites with established energy and water capacity. However, the guidelines also impose environmental protections: data center operators cannot simply plug into the grid and drain local supply. They must contract additional renewable generation to meet their electricity demand, with at least 40 percent of that contracted generation required to be wind power.

What Are the Key Infrastructure Requirements for AI Data Centers?

• Power Supply: AI data centers require connection to high-capacity power grids capable of delivering 100+ megawatts continuously, with many facilities needing additional renewable generation capacity to meet environmental standards.
• Water and Cooling Systems: Modern data centers use closed-loop cooling systems that recycle water on site, protecting local agricultural water security while preventing server overheating during intensive computational workloads.
• Internet Connectivity: Facilities must be positioned near major fiber optic infrastructure corridors to ensure low-latency data transmission and reliable network access for AI model training and inference operations.
• Land Zoning and Permits: Brownfield sites already zoned for heavy industry eliminate years of zoning approvals and environmental impact studies that greenfield developments require.

The federal strategy in the United States extends beyond private land. The Energy Department selected four locations on federal property for further development planning: Idaho National Laboratory, Oak Ridge Reservation, the Paducah Gaseous Diffusion Plant, and Savannah River Site. These existing federal laboratories and former nuclear facilities often have industrial infrastructure, specialized workforces, security systems, and established relationships with utilities, making them attractive alternatives to public lands managed by the Bureau of Land Management.

However, the acceleration of site identification has raised procedural concerns. The Bureau of Land Management reportedly received a three-day deadline to identify public land suitable for data center development, according to a September 11 investigation. Interior leaders described the request as a top priority, and the resulting lists were produced for Interior Department leaders. The number, location, and evaluation status of identified sites remain undisclosed, creating questions about whether rapid internal requests might look like decisions before required environmental analysis begins.

A transparent process would separate early screening from formal approval and identify why a parcel advanced, which alternatives were rejected, and what evidence remains necessary. Without that separation, a rapid federal land search can create political and legal risk even when an eventual project undergoes full review.

How Can Communities Ensure Environmental Protections Are Enforced?

Justin Clancy, Member for Albury, emphasized that communities have legitimate questions about how data center development will impact the local power grid, agricultural water security, and general amenity. The key is ensuring that guardrails hold up and environmental protections are enforced.

In practice, this means scrutinizing where supporting renewable infrastructure will actually be built. A requirement for additional renewable generation to support a data center should not be regarded as an automatic green light for any random renewable project in a region. Communities need to track whether developers are meeting their commitments to source wind power and whether water recycling systems are actually protecting local agricultural supplies.

The broader context is that AI infrastructure is becoming a combined technology and energy policy issue. Interior Secretary Doug Burgum has publicly promoted data centers as "intelligence factories," connecting electricity directly to computational output and treating energy production and AI capacity as parts of the same industrial system. Federal policy now treats a data center not as an isolated building but as a network of generation, pipelines, transmission lines, substations, processors, storage, and communications equipment.

That approach can simplify high-level planning, but it also expands the area affected by each proposed development. A server campus needs more than the parcel occupied by its buildings; its supporting infrastructure can cross additional public and private land. The challenge for regulators and communities is ensuring that acceleration does not shift infrastructure costs or environmental risks onto surrounding areas.

" }