Artificial intelligence is frequently described as a digital revolution, something intangible and infinitely scalable. Yet the systems underpinning AI are unmistakably physical. Every advance in model performance depends on hardware, from GPUs and servers to cooling systems, transformers, and reinforced grids. Embedded across that infrastructure are copper, nickel, cobalt, gold, and a range of specialty metals that enable modern computing.
As AI adoption accelerates, demand for these materials is rising at a pace that existing recovery and processing systems were not designed to match. The International Energy Agency projects that global data centre electricity demand could approach 945 TWh by 2030, as a result of AI and digitalisation. Electricity demand at that scale reflects a corresponding expansion in physical infrastructure, and therefore in the metals embedded within it.
The digital economy may feel abstract, but its foundations are industrial.
The scaling mismatch beneath AI growth
In the UK, the debate around AI centres on competitiveness, productivity, and national capability. What receives concerningly less attention is whether the physical systems beneath that ambition are capable of scaling in parallel.
The National Grid’s Future Energy Scenarios anticipate substantial growth in electricity demand from data centres over the coming decade. At the same time, the UK remains heavily reliant on imported raw and refined materials, and its Critical Minerals Strategy explicitly acknowledges the vulnerability created by concentrated global processing capacity.
This creates a structural mismatch. Digital infrastructure can be financed and deployed in relatively short timeframes, and software evolves even more rapidly. Metals processing infrastructure, by contrast, typically requires long development cycles and significant capital investment. Capacity expands slowly and is often geographically concentrated. AI can be deployed anywhere, but the physical reality is vastly different.
The question is therefore not simply one of procurement. It is whether material systems can scale in step with digital systems. When one layer expands incrementally and quickly, and the other expands slowly and in large steps, fragility increases.
Centralisation in a distributed economy
The prevailing metals recovery model remains highly centralised with processing assets optimised for bulk commodity flows and predictable industrial growth. These systems were not designed for a world defined by compressed innovation cycles, distributed compute clusters, and geopolitical fragmentation. Furthermore, such systems are unable to economically or practically capture new growth in metal processing capacity.
AI infrastructure scales unevenly and geographically, responding to energy availability, regulation, and competitive pressure. Centralised processing systems, dependent on long logistics chains and heavy capital commitments, struggle to respond with the same agility to the global AI race.
This is not an argument for dismantling existing infrastructure. Rather, it is recognition that centralised systems aren’t the solution to scale to capture metals growth driven by an electrified and deglobalised world.
The domestic resource in plain sight
Another dimension of the discussion concerns the materials already above ground. Globally, electronic waste reached 62 million tonnes, $100bn worth of metals, in 2022, according to the UN Global E-waste Monitor, with 1.7 million tonnes in the UK, yet less than a quarter was formally collected and recycled, leaving a staggering amount of natural resources uncaptured. Within the legacy equipment lies a concentration of metals essential to next-generation compute infrastructure. What is left underutilised above ground should be concerning us far more than it does. Simply put, it is the largest waste of resources on the planet.
In an economy characterised by accelerating hardware refresh cycles, the strategic relevance of these waste streams becomes clear. Circularity is often framed in environmental terms. Under sustained digital demand, it also represents a pragmatic response to supply concentration and long logistics chains.
Recovering metals from domestic waste streams can shorten supply chains, reduce exposure to geopolitical risk, and create traceable sovereign supplies of the materials needed for the next generation of hardware and industry. Building scalable domestic recovery capacity is, therefore, not simply an environmental initiative but a strategic infrastructure decision, essential for economic and national security.
For economies such as the UK, which lack both significant primary mineral extraction and midstream processing capabilities but aspire to AI leadership, this is not merely a sustainability consideration. It is a matter of industrial and sovereign resilience.
Distributed processing as infrastructure
Just as digital workflows have become distributed, with data storage moving from localised hardware to the cloud, we believe metals processing infrastructure needs to shift from a centralised to a distributed model.
The UK government has made supply chain resilience and critical minerals security major policy priorities, with an explicit goal to have 20% of metals supply generated by in-country recycling by 2035. The only way to achieve this is to invest in next-generation metals processing infrastructure, both to deploy production capacity at speed and to face the uncomfortable truth that the UK has no mid-stream metals recycling capability, as UK energy costs deem it economically unviable.
Distributed and modular processing models allow capacity to expand incrementally, closer to where materials are generated. Distributed infrastructure can be rolled out at speed, breaking down commercial barriers to entry and diversifying investment risk. However, deploying at speed can only be achieved with next-generation metals processing infrastructure, which, unlike its forbears, is not restricted by high-capex, major energy demands, and permitting headwinds.
Globally, complementing existing centralised infrastructure with systems designed for faster deployment and capital efficiency would better reflect the realities of AI-era growth.
From digital ambition to material strategy
AI strategy conversations, once centred on talent, regulation, and compute capacity, are now focused on energy and infrastructure. The bottleneck to the growth of AI is its dependence on the physical world.
The next phase of AI expansion will be shaped not only by advances in model architecture but also by grid capacity, cooling systems, and advanced metals throughput. Aligning digital ambition with material capability requires integrating recovery and processing considerations into broader industrial planning.
This does not imply the abrupt replacement of established systems. It does suggest that modularity, repeatability, and resilience, principles long associated with scalable software, should increasingly inform physical supply strategies.
The physical future of digital growth
AI may be driven by algorithms, but its future rests on a physical foundation. The durability of AI-led economic growth will depend in part on whether material supply chains are resilient, responsive, and capable of scaling alongside digital infrastructure.
The countries and organisations that succeed in the AI era will not simply be those that build more powerful models. They will be those who align digital ambition with physical systems capable of sustaining it.
Recognising the material dimension of AI does not diminish its transformative potential. It grounds it in industrial reality and ensures that its potential will be unlocked.
