Why the future of AI is physical – and what Governments must do about it

AI is often discussed as an invisible force in the “cloud,” but its real-world impact is anything but virtual. From energy-hungry datacentres and critical raw materials to mounting e-waste and infrastructure pressures, AI is increasingly a physical system with tangible environmental and social consequences.

In this Q&A, Diego Bermudez, PhD, DICE Network+ Impact and Research Fellow at the University of Exeter Business School, unpacks what the rise of “physical AI” means for governments.

We’re hearing the term, “the future of AI is physical.” How does it change the way governments should think about digital transformation?

To say “the future of AI is physical” is to acknowledge that the abstract digital world of code, algorithms, and computing is tethered to a massive, resource-intensive infrastructure of steel, silicon, and energy. This shifts the conversation from the cloud as an ethereal concept to the datacentre as a factory-scale physical asset that consumes land, water, and electricity while generating significant waste.

For governments, this reality necessitates a radical change in how digital transformation is approached; moving from a stand-alone digital policy to an integrated industrial and environmental strategy that secures and optimises availability and capacity of the global resources. It starts with critical raw materials and optimising resources, including digital sufficiency – which advocates for balance when using digital technology, advocating for “enough tech” and avoiding excessive usage – and extending the lifetime of the devices and their components, and maintaining value in decommissioned and end-of-life hardware.

How can physical AI help governments meet sustainability and net-zero goals in practical ways? 

At the DICE Network, we focus on two goals: using digital technologies like AI to drive sustainability and circularity, and ensuring those technologies are circular throughout their own lifecycles. We refer to this as the physicality of AI. This involves making AI infrastructure circular while using AI to optimise physical systems like datacentres and sensors for better resource management.

The potential for government impact is significant. AI-enabled smart grids and buildings can drastically reduce energy waste and recirculate heat. Furthermore, AI can tackle the global e-waste crisis, which saw $91 billion in minerals lost in 2022, through intelligent sorting, predictive logistics that cut transport emissions by 35 percent, and the discovery of new ways to use waste as industrial feedstock.

What risks are there that physical AI could increase environmental impact rather than reduce it?

While physical AI offers significant potential to optimise the real world, its rapid expansion presents several critical risks that could exacerbate, rather than reduce, the global environmental footprint. The primary danger is a rebound effect, also called Jevons Paradox, where efficiency gains in individual AI tasks are overwhelmed by an exponential surge in aggregate demand.

Some of the risks includes AI growth outpacing the energy grid decarbonisation, skyrocketing energy use and creating infrastructure strains that could trigger a tragedy of the commons across industries. Furthermore, the ‘physical body’ of AI relies on finite, often unrecovered raw materials. The absence of high-value retention strategies, such as repair, refurbishment, and remanufacturing, threatens a significant deficit at both material and component level (i.e. memory shortages). This risk is further compounded by a supply chain heavily concentrated in East Asia, leaving the global digital economy precariously vulnerable to geopolitical shifts and logistical volatility.

What policy frameworks or procurement models are needed to ensure physical AI systems are designed sustainably from the outset?

To ensure physical AI systems are sustainable from the outset, we must move beyond treating sustainability as a nice-to-have engineering fix and embed it into the legal and financial foundations of technology.

A first step would be mandating circularity into regulatory frameworks, EU Ecodesign for Sustainable Products Regulation (ESPR), Digital Product Passports (DPP) and Right-to-Repair laws are good examples. Second, shifting procurement models to target the ‘sustainability sandwich’ problem, where sustainable requirements are squeezed and price drives most of the decisions. The UK’s Government Buying Standards are a good effort to push OEMs and service providers towards more sustainably and circular digital systems. Refurb by Default policies for enterprise hardware should be the first choice for non-critical compute tasks, and for those that are critical, Hardware-as-a service (HaaS), where vendors retain ownership are a good example to maintain high SLA (Service Level Agreements) and incentivise sustainability through profitable circular business models.

How should governments balance innovation with safety, ethics, and public trust when AI systems operate in physical spaces and public infrastructure?

To balance innovation with safety, ethics, and trust, governments must transition from passive regulators to active Systemic Architects. This shift requires a focus on three critical pillars.

First, Equitable Resource Allocation & Whole-Systems Planning; massive AI infrastructure, such as datacentres, often competes with public services for limited resources. In regions like West London, AI demand has caused decade-long delays for housing and hospitals due to grid constraints. To prevent digital innovation from compromising social equity, governments must implement Whole-Systems Planning, ensuring infrastructure is not approved in a vacuum but balanced against community needs.

Second, move datacentres from ‘Black Boxes’ to Community Partners, involving all the value chain stakeholders, from investors to Big Tech companies; public trust erodes when infrastructure is viewed as a drain on local energy and water. For example, governments can bridge this gap by mandating or incentivising Heat Reuse. By adopting standards like Germany’s 10 percent Energy Reuse Factor (ERF), datacentres are transformed into public utilities that provide low-cost heating to homes, pools, and greenhouses.

Third, push for Transparency & Resilience through Open Standards; trust requires visibility into both AI code and hardware. Digital Product Passports (DPPs) can provide a transparent record of a device’s material origin and carbon footprint, ensuring ethical sourcing. Furthermore, relying on proprietary black box vendors poses a significant security risk. By mandating Open Standards, modelled after frameworks like the Open Compute Project or Open Banking rely, governments ensure hardware remains modular and interoperable. This shift grants users sovereign control over repairs and upgrades removing dependency on proprietary corporate lifecycles. Ultimately, public procurement must pioritise long-term resilience and operational sufficiency to break the cycle of forced obsolescence and excessive technology replacement.

What skills and institutional capabilities will public sector organisations need to manage and govern physical AI systems responsibly?

To manage the ‘physical body’ of AI responsibly, public sector organisations must move beyond digital literacy (i.e., software and data) and develop industrial and circular literacy (i.e., hardware, energy, and materials).

Four essential institutional capabilities and skills sets required:

1. Strategic Procurement & Financial Modelling, public sector teams need the capability to shift from short-term ‘buy-and-replace’ cycles to long-term ‘value-retention’ models.
2. Whole-Systems Planning & Infrastructure Integration, governments must stop viewing datacentres as isolated black boxes and start seeing them as integral parts of local infrastructure.
3. Circular Supply Chain Governance, with 70 percent of critical raw materials (CRMs) concentrated in East Asia, public institutions need the capability to secure and recover resources domestically.
4. Technical Regulatory Oversight, as regulations like the EU’s Ecodesign for Sustainable Products Regulation (ESPR) become standard, public bodies must be able to enforce and verify hardware standards.

Looking ahead, what would “sustainable physical AI” look like in practice for a city or national government in 10-15 years?

Thinking in 10-15 years’ time, the successful implementation of a circular infrastructure roadmap would transform AI and other digital technologies from a resource-draining ‘bottleneck’ into a resilient ‘competitive’ market.

In this future, the cloud is no longer an abstraction but a cornerstone of integrated urban and national resource management. The datacentre becomes a Resource Hub, rather than being isolated facilities, datacentres act as active System Contributors. For example, datacentres are co-located with ‘heat off-takers’ such as district heating networks, greenhouses, and public pools. These facilities act as stabilising assets for the energy grid, helping manage decentralised energy assets through models like Virtual Power Plants. Every datacentre serves as an urban mining site (or feeds a centralised site), where critical raw materials (CRMs) like those found in chips are recovered and fed back into the domestic supply chain.

To create a systemic circular ecosystem, circular principles and KPIs are mandatory in all investment criteria, public infrastructure planning permissions, and procurement decisions. Moving beyond Total Cost of Ownership to a holistic Total Value of Sustainability that integrates the whole value chain. A mature, liquid market for refurbished hardware is the primary source for ‘non-leading-edge’ compute tasks, and last, the state no longer owns disposable hardware. The Black Box era of proprietary hardware has been replaced by open-standard, modular designs that is differentiated by service level satisfaction. This systemic change requires coordination and collaboration from everyone to make it happen.