Data centres and the impending deluge of e-waste - Potential 925% increase to yearly e-waste

As demand for artificial intelligence, cloud computing and digital services accelerates, Australia is experiencing a massive expansion in AI investment and approval of data centre infrastructure. This expansion has the support of the federal government as a key pillar of the Future Made in Australia Agenda¹, and state governments through expedited approval processes.²

While insufficient community consultation, and the carbon, energy and water costs of this new form of infrastructure has received significant media and political attention, the massive quantity of e-waste that will follow this expansion has been largely ignored.

Computer components within data centres contain a number of toxic heavy metals and problematic chemicals, including PFAS and other Persistent Organic Pollutants (POPs) which represent significant environmental and human health risks if not handled safely. In addition to the upstream consequences of the drive for new materials, there are a number of pathways by which these pollutants may be released during and after use as outlined below:

Figure 1: Heavy metal & chemical release.³

While individual components may contain relatively small quantities of these harmful elements (which include known carcinogens, teratogens, and mutagens), many of these substances do not naturally break-down and risk bioaccumulation in animals and progressive biomagnification up the food chain, increasing the potential for harm; Risks of this nature are compounded by increased volume.

A study in early 2024 estimated that e-waste generated from specialised servers to power generative AI could add 5 billion kgs of e-waste globally by 2030.⁴ This assessment was based on projections of AI computational demand and current server deployment, the study understandably excluded any projected contribution from Australia. The assumptions in that study were made before the rapid expansion of tech investment in data centres that has been experienced globally since 2025, and before Australia formally signalled its intent to become an AI powerhouse.⁵

To understand the potential quantity of e-waste that could be generated in Australian data centres, we need to look beyond merely the servers to power generative AI. Data centres are made of vast quantities of servers, hard disk and solid state drives, networking equipment, cooling systems, electronic security systems and back up power in the form of generators and batteries.

The drive for data centre development includes the desire for generative AIm as assessed above, but also includes the need for computational data analytics, data storage, streaming services, rendering 3d models, visual effects and computer aided design, scientific research, cloud migration and the increased need for transparent data governance and national data sovereignty. Many of these needs benefit from similar high performance Graphics Processing Units (GPU) as those required for generative AI.

It is difficult to accurately predict the volume of e-waste that will be produced by data centres in Australia. An accurate assessment would require access to the precise number and specifications of servers, GPUs, networking equipment, data storage, electrical supporting equipment, power, security and cooling systems, as well ast the intended purpose and intensity of use of the data centre in question, as as increased computational demands and the nature of the tasks performed, directly impact the lifespan of GPUs.⁶ This information is not sufficiently disclosed in public planning material.

We can however provide a rough estimate based on power requirements of the data centres themselves with the following assumptions:

Each server rack holds between 42 smaller, and 22 larger servers⁷
Servers weigh between 13kg and 31 kgs⁸
A mid-density data centre requires 1MW to support 100 standard server racks;⁹

1MW of data centre capacity with 100 server racks would include between 54,600kg and 68,200kg of servers alone, excluding the weight of racks, networking equipment, power, security or cooling infrastructure.¹⁰

The optimal lifespan of the servers could be a short as one year ¹¹due to heavy workloads and rapid obsolescence driven by depreciating peak performance windows or technological advancement. The incentive to stay ahead of the competition suggests modern data centers may replace servers every three to five years.

To illustrate this potential quantities of e-waste individual data centres could contribute, we examined three currently proposed facilities applying the above estimates:

Proposed Facility	Megawatts	Lower waste estimate every 5 years	Higher waste estimate every 3 years
NEXTDC Eastern Creek Sydney¹²	612	33,415,200 kg	41,740,000 kg
IREN Bundey Project (SA)¹³	800	43,688,000 kg	54,560,000 kg
Mamre Road Data Centre Sydney¹⁴	1000	54,600,000 kg	68,200,000 kg

This produces a range of 131,700,000 to 164,500,000kg generated by the first replacement of servers in these three data centres alone.

This volume becomes more problematic when we consider there are almost 300 data centers proposed for, or already operating in, Australia.¹⁵ While the majority of these data centres are smaller than the examples above, investment for new and expanded data centres continues to grow, with a number of proposals exceeding the capacity of those above.¹⁶

Assuming 300 facilities are, on average, half as large as the three examples relied upon above, we can assume they will produce roughly 50% of the midpoint between the above estimates.¹⁷ On that basis collectively those facilities would produce 22,215,000,000 kgs (22.215 million tonnes) of server stock every four years.

Assuming the server turnover is spread evenly across those four years, and no other e-waste is generated from these facilities, the yearly volume of e-waste produced in Australia would increase by more than 925% from 600m kgs, to more than 6,150m kgs. ¹⁸

Optimistically, assuming the average size of the currently operating and proposed data centres was as low as 150MW, and assuming no more than a total of 300 data centres are developed in Australia (which appears unlikely) total e-waste generated from end-of-use servers after four years would be 2,763,000,000kg (>2.75m million tonnes).¹⁹ If this was spread evenly over a four year period, this would add more than 690m kgs to our yearly e-waste, from data centre servers alone, representing an increase of 115% from current levels.²⁰ In the absence of a firm commitment to refuse additional data centres beyond 300 centres, this figure is likely to represent a significant underestimate of future quantities of e-waste that will be generated by data centres in Australia.

These estimates are necessarily broad, noting publicly disclosed documents rarely state the precise components, the intended use, or the anticipated lifespan of the relevant hardware proposed to be used. These indicative scenario estimates highlight the importance of assessing and addressing the cumulative impact of e-waste generated by the rapid expansion of data centres in Australia.

In government’s rush to facilitate the expansion of data centres, previous recognition of the importance of circularity and sustainability in regards to e-waste, has been largely forgotten.

Australian Government, Department of Industry, Science and Resources (2025) Australia launches National AI Plan to capture opportunities, share benefits and keep Australians safe. Web Article. ↩︎
NSW state significant development (SSD) and VIC Development Facilitation Program both facilitate approval within 75 days. ↩︎
Reproduced with permission from I. Chowdhury, N. Aich (2026) Digital Infrastructure and the Overlooked Chemical Footprint: Toward a Holistic Environmental Framework for Data Centers. Environ. Sci. Technol. 18 August 2026; 60 (32): 22065–22070. ↩︎
For the ten year period 2020 to 2030 – see: Wang, P., Zhang, LY., Tzachor, A. et al. (2024) E-waste challenges of generative artificial intelligence. Nat Comput Sci 4, 818–823 (2024). ↩︎
Australian Government Department of Industry, Science and Resources (2025) National AI Plan. ↩︎
Which can be as short as one year. See: Falk, S., Kluge Corrêa, N., Luccioni, S. et al. From computation to environmental cost the resource burden of artificial intelligence. Commun Earth Environ 7, 397 (2026). https://doi.org/10.1038/s43247-026-03537 ↩︎
Based on specifications for the smaller and larger servers below. ↩︎
The Gigabyte G152-Z12 has a net weight of 13.2 kg based on online specifications ; whereas a System76 Ibex Pro 2U GPU Server has a base weight of 30.84kg based on online specifications. ↩︎
6-9kw per rack as per Datacenters.com (2025) kW per Rack Explained: Optimize Your Data Center. Plus a small allowance for support infrastructure, including cooling = 10kw. 10 x 100 = 1,000kw or 1MW. ↩︎
This is perhaps a conservative as there are reportedly some generative AI racks that weigh more than 1800kgs each. See: Andy Patrizio (2025) Heavy Compute: AI Data Centers Have a Weight Problem; Webarticle on Datacenterknowledge ↩︎
See: Falk, S., Kluge Corrêa, N., Luccioni, S. et al. From computation to environmental cost the resource burden of artificial intelligence. Commun Earth Environ 7, 397 (2026). https://doi.org/10.1038/s43247-026-03537 ↩︎
State Significant Development Application profile on NSW Planning portal (Available here) Low range: 42 x 13 x 612 x 100 = 33,415,200kgs High Range: 22 x 31 x 612 x 100 = 41,738,400kgs ↩︎
Madden, A Waters (2026) Locals fear one of the largest data centres proposed for Asia Pacific will use water from ‘stressed’ River Murray. ABC News online 16 June 2026. ↩︎
State Significant Development Application profile on NSW Planning portal (Available here) ↩︎
datacentermap lists 296 data centers in Australia ↩︎
For example, Zerra DC’s proposed Western Downs Digital Park in Kogan QLD for a 1.44GW facility, and NextDC’s proposed M5 facility in Melbourne at 1.2GW. ↩︎
i.e 50% of 148,100,000kgs = 74,050,000 per facility times 300 = 22,215,000,000kgs ↩︎
22,215,000 kgs/4 = 5,553,750,000kgs on top of 600m kgs already reaching EoL = 6,153.75million kg. 600m x 925.6% = 5,553.6m kgs (or 5.5536 billion kgs) ↩︎
Maintaining a range of 54,600kg and 68,200kg per 1MV, times 150 = 8,190,000 to 10,230,000kg. Midpoint waste would be 9,210,000kg. To account for 300 data centers 9,210,000kg x 300 = 2,763,000,000kg of e-waste produced from servers alone. . ↩︎
2,763,000,000kg/4 = 690,750,000kg. Together with the current 600m kgs currently produced, would be 1,290,750,000kgs in total. Increase 600m kgs x 115.125% ↩︎