AI E-Waste Could Reach 617 Million Tonnes by 2050, BAN Warns

Artificial intelligence could drive 395 million to 617 million tonnes of cumulative electronic equipment retirements between 2025 and 2050 as data centers replace servers, power systems, cooling equipment, batteries and networking hardware, according to a new Basel Action Network analysis.

By Paul Ploumis
Published September 17, 2026

MONTREAL (Scrap Monster): The rapid expansion of artificial intelligence infrastructure could create a much larger electronic-waste stream than earlier estimates have suggested, according to a new white paper from the Basel Action Network.

BAN estimates that AI-driven electronic equipment retirements could total between 395 million and 617 million metric tonnes worldwide from 2025 through 2050.

At the upper end of that range, the discarded equipment would fill roughly 23 million 40-foot shipping containers. BAN estimates that a line of about 20 million such containers would stretch approximately 244,000 kilometres, or roughly six times around the Earth.

The scale of the estimate comes from taking a broader view of the physical infrastructure behind AI. Rather than counting only servers and graphics processors, the analysis includes the power, cooling, networking and backup systems required to operate modern data centers.

Why BAN's Estimate Is Higher Than Earlier AI E-Waste Studies

Previous AI e-waste studies have generally concentrated on servers, graphics processing units and other computing hardware.

BAN argues that this approach captures only a fraction of the equipment that ultimately becomes waste when a data center is upgraded or decommissioned.

Under its model, accelerators, servers and racks account for approximately 13% of data-center electro-mechanical equipment mass.

The remaining equipment includes power-supply and distribution infrastructure, cooling systems, backup power systems and networking equipment.

BAN estimates that the five categories together represent approximately 70,000 tonnes of installed equipment for every gigawatt of data-center capacity.

Rapid Hardware Turnover Adds to the Waste Pipeline

Equipment replacement rates are an important part of BAN's model.

The paper assumes a 2.5-year lifespan for AI accelerators, servers and racks, compared with traditional general-purpose server lifespans that have historically been longer.

Networking equipment is modeled on a 3.5-year cycle, while backup power and cooling systems are assigned five-year lifespans. Power-distribution equipment is modeled at eight years.

BAN notes that some of these figures are based on published industry data, while others are its own estimates where established lifespan data is limited.

The report argues that increasingly dense AI computing and the transition from conventional air cooling toward liquid cooling could accelerate the replacement of infrastructure that might otherwise have remained in service longer.

AI E-Waste Could Reach 31-46 Million Tonnes Annually by 2050

The cumulative 395-million-to-617-million-tonne estimate covers equipment retired over the entire 2025-2050 period.

On an annual basis, BAN projects AI-related electronic waste could reach approximately 31 million to 46 million tonnes per year by 2050.

When conventional electronic waste is included, the organization estimates total global e-waste generation could reach roughly 196 million to 211 million tonnes annually by 2050 under its modeling assumptions.

The report expects the sharpest increase in data-center waste to occur between 2027 and 2035, when equipment installed during the current infrastructure buildout begins reaching the end of its first operating cycles.

'AI Waste Contagion' Expands the Estimate Beyond Data Centers

One of the most consequential parts of BAN's methodology extends beyond equipment located inside data centers.

The organization uses the term "AI Waste Contagion" to describe electronic equipment outside data centers that could be replaced earlier as AI becomes embedded in personal computers, mobile devices, telecommunications systems and enterprise hardware.

This portion of the analysis contributes substantially to the upper end of the overall estimate.

BAN acknowledges the uncertainty involved. Some of the assumptions used to estimate AI-driven replacement of consumer and enterprise equipment are original estimates rather than observed long-term data.

The paper describes its aggressive scenario as an upper-bound case intended to show a plausible range rather than as a precise forecast.

Model Is Sensitive to Data-Center Growth

BAN includes a sensitivity analysis showing that the rate of data-center capacity growth has a larger effect on projected infrastructure waste than modest changes in individual equipment lifespans.

The organization's baseline infrastructure model assumes annual capacity growth of 8.8% and equipment intensity of about 70,000 tonnes per gigawatt.

When BAN reduces the capacity-growth assumption to 6%, modeled annual data-center infrastructure waste falls substantially. Increasing growth assumptions produces the opposite result.

The range underscores why the 2050 numbers should be read as modeled scenarios rather than a fixed prediction of how much equipment will ultimately be discarded.

Global E-Waste Collection Already Lags Generation

The projected AI waste stream would arrive on top of an existing global e-waste challenge.

The International Telecommunication Union and United Nations Institute for Training and Research estimated that the world generated 62 million tonnes of electronic waste in 2022.

Only 22.3% was documented as formally collected and recycled in an environmentally sound manner.

The UN's Global E-waste Monitor projects total e-waste generation could reach 82 million tonnes by 2030 if current trends continue.

That gap between generation and documented recycling capacity is one of the reasons BAN argues that AI hardware retirement needs to be considered during infrastructure planning rather than after equipment reaches end of life.

What the AI Hardware Wave Means for Electronics Recyclers

The expansion of AI data centers is also changing the type of material likely to enter electronics-recycling and IT asset disposition channels.

AI-focused servers contain dense concentrations of GPUs, memory and other high-value components. That creates opportunities for testing, refurbishment, component harvesting and resale before equipment is sent for commodity recovery.

Industry participants expect retirements to occur as a rolling series of upgrades rather than as a single wave, as data-center operators replace hardware in stages.

That could increase demand for secure data destruction, serialized equipment tracking, component testing and specialized recovery services.

For recyclers, the economic value of retired AI systems may therefore depend as much on reuse and component recovery as on the underlying copper, aluminum, precious metals and other commodities contained in the equipment.

Power and Cooling Equipment Broaden the Recycling Challenge

The waste stream will also extend beyond conventional IT equipment.

BAN's model includes large volumes of copper-intensive power-distribution hardware, battery backup systems, cooling equipment and network infrastructure.

Different categories require different end-of-life processes. Batteries require specialized handling, while cooling systems and power equipment can contain materials or components that fall under electronic-waste controls.

The broader equipment mix means data-center decommissioning increasingly sits at the intersection of electronics recycling, battery management, metal recovery and industrial asset disposition.

International E-Waste Rules Have Tightened

The regulatory framework governing cross-border electronic waste has also become more restrictive.

Amendments to the Basel Convention that took effect January 1, 2025 expanded controls so that transboundary movements of both hazardous and non-hazardous e-waste are generally subject to prior informed consent requirements among participating countries.

The rules cover discarded electrical and electronic equipment, components and certain residues generated during processing.

That makes traceability and final destination increasingly important considerations as larger volumes of data-center equipment enter international reuse and recycling markets.

Reuse Could Reduce the Amount Entering Recycling

The BAN estimate is based on equipment retirement, which does not necessarily mean every unit will immediately be shredded, smelted or disposed of.

Secondary markets could absorb some servers, components and related equipment that remain useful for less demanding computing workloads.

Reuse, refurbishment and parts harvesting can extend equipment life and delay entry into the recycling stream.

BAN recognizes this as a potentially important moderating factor, although the organization argues that secondary markets are unlikely to eliminate the larger waste challenge if AI infrastructure continues expanding at the rates assumed in its model.

What to Watch

The size of the eventual AI-related e-waste stream will depend heavily on how quickly data-center capacity continues to expand, how often equipment is replaced and how much retired hardware finds a viable second use.

For recyclers and IT asset disposition providers, the more immediate issue is the composition of the incoming material.

The current AI buildout is creating a future recovery stream that extends well beyond servers. It includes high-value computing hardware, large quantities of copper-bearing electrical infrastructure, batteries, cooling systems and networking equipment—all with different reuse, processing and compliance requirements.

BAN's projection remains a scenario rather than a certainty, but its broader accounting highlights an issue that previous server-only estimates largely missed: the physical footprint of AI infrastructure is much larger than the processors at the center of it.