A new report from the Basel Action Network argues that AI datacentre e-waste has been drastically underestimated once power, cooling and networking gear are counted alongside GPUs. For UK operators already navigating WEEE e-waste compliance, the maths on refresh cycles and disposal liability looks different.
View the data behind this chart
| GPUs/Accelerators | Networking | Cooling | Backup Power | Servers | Power Distribution | |
|---|---|---|---|---|---|---|
| Years to replacement | years2.5 | years3.5 | years5 | years5 | years6 | years8 |
What the Basel Action Network actually found
The nonprofit's report, How Big Is the AI Waste Wave?, models e-waste from the industry's own stated build-out plans rather than relying on server-and-GPU counts alone. It finds that accelerators and servers make up just 13 percent of a datacentre's equipment by weight, meaning previous academic projections omitted the bulk of the problem: networking, power distribution, backup power, and cooling systems.
Once those five categories are added, BAN puts total e-waste from AI infrastructure between 2025 and 2050 at 395 million to 617 million tonnes — 40 to 60 times higher than the most widely cited prior estimates. Lined up end to end, that volume of shipping containers would stretch roughly 244,000 km, or about six times around the Earth. A single reference 100 MW AI facility alone is estimated to generate around 7,000 metric tons of equipment mass over its life, much of it torn out during upgrades rather than at natural end of life.
Why this lands harder on UK operators
UK data centres were formally designated critical national infrastructure in 2024, which raises the bar on chain-of-custody controls and continuity planning around any hardware retirement programme, not just headline outages. New rules also let data centres apply for nationally significant infrastructure project status, a route that speeds up planning approval but makes large-scale build or refresh decisions more politically visible.
Grid capacity adds another pressure point: Ofgem opened a consultation on overhauling grid connections in July 2026, underlining that power access remains a binding constraint on expansion. Political scrutiny is rising too — commentary from the Green Party in August 2026 pushed for slowing UK datacentre construction until water and energy use are addressed, and e-waste is likely to become part of that same conversation.
GPU replacement cycles versus the rest of the estate
BAN's model assumes GPUs and other AI accelerators are swapped out every two to three years, against five to seven years for general-purpose servers. Because OEMs typically ship hyperscalers complete, preconfigured systems, the report assumes whole servers are replaced alongside the chips, not just the accelerator cards. Other categories run on different clocks: three to four years for networking, five years for backup power and cooling, and eight years for power distribution — the copper busbars and cabling that, in a single 100 MW facility, can amount to roughly 2,700 tonnes of copper.
That mismatch matters for buyers. A facility upgrading from 5-15 kW racks to the 50-140 kW racks AI workloads demand may have to rip out power infrastructure long before it was due for replacement on its own schedule, dragging forward capital spend and waste volume at the same time.

Disposal liability doesn't disappear when hardware leaves the building
Equipment that still works and is genuinely destined for reuse, repair or donation can remain classified as electrical and electronic equipment rather than waste under UK rules — but donating hardware is not a substitute for proper waste handling if it ends up discarded anyway. The Environment Agency's position is that businesses shouldn't be discouraged from donating usable kit, provided they can show it's genuinely going back into service.
Where equipment is retired outright, operators need due diligence on recyclers and carriers covering WEEE obligations, hazardous-waste handling, controlled-waste transport and relevant permits. Non-compliant downstream handling creates both legal exposure and reputational risk, and industry guidance is explicit that operators should keep an inventory of end-of-life assets and verify what recyclers actually do with equipment, not just assume it's been dealt with correctly.
- •Keep documented proof that donated or resold kit is still working and intended for continued use
- •Audit recycler and carrier compliance rather than taking resale claims at face value
- •Track reuse, recycling and landfill-diversion rates as ongoing operational metrics
Lifecycle strategies that cut both cost and risk
The report's own framing — that reuse and refurbishment could meaningfully reduce future e-waste volumes — points to where UK buyers have room to act now. Sector guidance sets out a clear order of preference: reuse and refurbish equipment inside your own estate first, cannibalise retired units for spares second, and send only what's left to certified recycling.
Refresh cycles commonly run three to five years across the industry, though schedules stretch from one to eight years depending on workload and reliability needs, and setting refresh rates by workload class rather than a blanket calendar can cut both environmental impact and spend. Buyers weighing whether to hold onto ageing capacity or refresh early can lean on tools like an IT hardware refresh planner or a server end-of-life checker to model the trade-off properly instead of guessing.
On the supply side, refurbished servers and circular IT and server remanufacturing extend usable life for workloads that don't need the newest silicon, while third-party maintenance support keeps existing estate running well past OEM-recommended refresh dates. Consumption-based procurement — extending usage, going month-to-month, or returning kit for refresh later — can also defer large capital replacements and smooth the volume of equipment hitting the waste chain at any one time; IT finance options built around that model are worth reviewing before committing to a fixed refresh calendar.
View the data behind this chart
| Layer | Detail |
|---|---|
| Reuse in-house | Redeploy racks and nodes elsewhere in your own estate first |
| Refurbish & resell | Strip for spares or move through certified refurb channels |
| Recycle or dispose | Certified recycling only once reuse options are exhausted |
The bottom line for infrastructure buyers
No hyperscaler, government or international body has yet published a plan for handling AI-driven e-waste at the scale BAN projects, and the report is blunt that current recycling infrastructure can't safely process even today's volumes. As BAN founder Jim Puckett put it, "If companies and governments do not begin planning for this new waste tsunami, today's AI buildout could become an even more cataclysmic toxic waste crisis than we are already experiencing."
For UK buyers, the practical response isn't waiting for a national framework — it's tightening reuse-first lifecycle policy, documenting disposal chains, and treating hardware refresh as a planned cost rather than a reactive one.
- 01The Register — AI boom could leave an e-waste trail that wraps 6 times around Earth · 19 September 2026
- 02Data Center Dynamics — UK energy regulator Ofgem launches grid connection overhaul consultation · 1 July 2026
- 03Computer Weekly — Avoiding security issues when recycling hardware · 1 January 2026
- 04The Register — Green Party wants to slam the brakes on UK datacenter construction · 28 August 2026
- 05Data Center Dynamics — UK data centers can be considered nationally significant infrastructure projects · 1 January 2026
- 06Computer Weekly — Storage explained: Consumption models of storage procurement · 1 January 2026
- 07computerweekly.com
- 08theregister.com
