If your organisation is running AI workloads, there is a hardware problem coming that most IT teams have not planned for.
Enterprise AI infrastructure is now cycling out every 12 to 18 months. That is not a forecast. It is what is happening at data centres and enterprise deployments globally right now, as organisations chase faster GPUs, higher memory bandwidth, and the latest accelerator architectures. Compare that to the traditional server refresh cycle of three to five years, and you start to see the issue.
More hardware in, more hardware out, faster than ever. And most organisations have no end-of-life plan for it.
The Data Risk Most Teams Are Not Thinking About
A decommissioned GPU is not just metal and silicon. It is a device that has held model weights, training datasets, inference logs, and in many cases, the queries and outputs of every user who interacted with a production AI system. That data does not disappear when you power the card down.
NIST 800-88 Rev. 1 applies to every storage medium that has held sensitive data, including the NVRAM, HBM2 memory, and onboard flash in modern AI accelerators. The same standard you apply to an old hard drive applies here. If you are not following a documented, verified data sanitisation process, you are carrying risk you cannot see.
Victoria’s Privacy Act obligations and the Commonwealth Australian Privacy Principles do not provide an exemption for hardware you have declared end-of-life. The data was there. It needs to be destroyed.
The E-Waste Angle Is Not Going Away
Victoria’s e-waste landfill ban covers all electronic equipment, including enterprise server and GPU hardware. Putting decommissioned AI accelerators in a skip, handing them to an unaccredited recycler, or letting them stack up in a storeroom is not a strategy. It is a compliance gap.
AS/NZS 5377 sets the standard for how e-waste is collected, stored, handled, and recycled in Australia. For hardware of this complexity and value, working with a provider who operates to this standard is not optional if you want to demonstrate compliance in an audit.
The Residual Value Opportunity
Here is the other side of the equation. Enterprise AI hardware that is 12 to 18 months old still carries substantial value in secondary markets. A structured ITAD process, one that includes grading, testing, and resale of reusable components before anything goes to material recovery, recovers far more than sending hardware directly to recycling.
Organisations with an ITAD plan for their AI hardware can offset part of the cost of the next refresh. Those without one leave that value on the table and pick up compliance exposure at the same time.
What a Compliant AI Hardware Decommission Looks Like
The process is not complicated, but it needs to be documented and defensible.
Data destruction first: every storage medium on every device, verified to NIST 800-88 Rev. 1, with a certificate of destruction that names the specific devices and the sanitisation method applied. Then asset assessment: what can be refurbished and reused, what goes to material recovery, and what is the residual value. Then disposal under AS/NZS 5377, with chain-of-custody documentation from collection through to final outcome.
At Electronic Waste Victoria, we handle this for organisations across Victoria, from single decommissions to full data centre refreshes. Our processes align with ISO 14001 for environmental management and ISO 27001 for information security, so the documentation holds up under audit.
The Question to Ask Now
If your organisation is running AI infrastructure today, ask your IT team one question: when the next hardware refresh happens, what is the plan for the outgoing equipment?
If the answer is vague, now is the time to sort it out. Contact EWV to discuss a decommission plan for your AI hardware before the refresh cycle forces the decision.