AI may appear to exist in the cloud, but its infrastructure is profoundly material. Data centres, semiconductors, servers and electricity networks require enormous quantities of energy, water and minerals, including copper, lithium, cobalt, nickel, graphite and rare earth elements.

These are precisely the resources needed to expand renewable energy generation, electricity grids, battery storage and other technologies essential to the green transition. AI is therefore intensifying competition for minerals whose supply chains are already geographically concentrated and associated with significant environmental and human rights concerns.

The International Energy Agency's 2026 Critical Minerals Outlook highlights the growing importance of minerals not only for clean energy technologies but also for semiconductors, servers and AI infrastructure. Research published this year on AI data centres estimates that copper accounts for 83% of the mineral mass required, most of it in the transmission and distribution lines that power them.

Meanwhile, the environmental footprint of AI is growing. The IEA projects that electricity consumption by data centres could almost double, from 485 terawatt-hours in 2025 to around 950 terawatt-hours by 2030. Water consumption, land use, waste and pressure on local infrastructure add further costs, often borne by communities that have little say in the decisions affecting their resources.

We must ask whether this is a transition that complements the move towards a low-carbon economy or one that competes with it, potentially diverting resources, investment and political attention away from the technologies and public infrastructure needed to guarantee a just transition.

The AI economy also depends on an extensive and often invisible workforce. Behind supposedly autonomous systems are people extracting minerals, manufacturing hardware, labelling data, moderating content, training models, maintaining infrastructure and performing the human feedback work that makes these systems function.

Many of these workers face precarious conditions, low pay, inadequate protections and significant psychosocial risks. Their contributions are rarely reflected in the narratives of technological innovation or in the distribution of its financial rewards.

There is another form of extraction at work: the appropriation of knowledge. AI models have been developed using vast amounts of information, creative work, research and material produced by people, communities and institutions, often without meaningful consent, acknowledgement or compensation. The resulting products are then commercialised through increasingly expensive subscriptions and proprietary services.

Businesses are also being encouraged to spend more. Gartner found that 85% of functional leaders planned to increase AI spending in 2026, after allocating an average of 12% of their functional budgets to AI in 2025. SAP and Oxford Economics estimate that average UK business spending on AI reached £20.6 million in 2026, a 27% increase on the previous year.

Yet rising expenditure is not the same as demonstrated economic value. And while AI may improve productivity in particular tasks, its effects on employment, working conditions and the distribution of productivity gains remain deeply uncertain. The ILO has highlighted the gap between productivity improvements at the individual task level and the much less conclusive evidence of gains at firm and economy-wide levels.

The danger is that workers and knowledge producers bear the costs of this transformation while a small number of technology companies capture the value.

The scale of investment in AI is extraordinary. But the promise of transformative economic growth is running ahead of the evidence that these investments will deliver broad-based returns.

AI has real economic potential, but the scale, timing and distribution of its benefits remain uncertain.

The market reaction to the Democrats' sweeping victories in the November 2026 US midterm elections offers an interesting illustration. AI-linked and data-centre stocks initially fell amid investor concerns that stronger oversight, environmental safeguards and restrictions on data-centre expansion could affect the profitability of the sector. This volatility is a reminder that the AI investment boom depends not only on technological breakthroughs but also on political choices, public infrastructure and the regulatory environment.

There is a deeper question about who assumes the risk. Governments are being encouraged to subsidise infrastructure, provide energy and water, facilitate access to land and critical minerals, and use public finance to attract private investment. The IEA reports that public finance commitments for critical minerals investment in advanced economies reached around $65 billion in 2025, more than four times the 2023 level.

This raises a familiar concern from the political economy of development: are public resources being used to de-risk investments that might otherwise be unprofitable, while the eventual gains are captured by private investors? In other words, are we socialising the risks and privatising the returns?

This question becomes particularly urgent when set against extreme wealth concentration and the extraordinary political influence of some of the individuals and corporations driving the AI boom. Oxfam reports that global billionaire wealth reached $18.3 trillion in 2025, while its 2026 analysis estimates that $3.55 trillion in wealth was held offshore and untaxed in 2024. Research by Gabriel Zucman and colleagues finds that the 400 wealthiest US households paid an effective tax rate of just 24% in 2018–2020, compared with 30% for the population as a whole.

When those accumulating extraordinary wealth also exercise growing influence over public debate, political decisions and the rules governing technology, the challenge is not simply economic concentration. It is democratic accountability.

Moreover, should a speculative bubble ultimately lead to an AI crash, as many fear it will, it is unclear who will bear the cost, and recent history does not bode well. The IMF warned just last month that an AI investment bubble could burst, triggering a global financial shock, while others are warning that it could be as severe as 2008. As CESR has demonstrated, austerity programs implemented in the wake of the last global financial crisis drove historic retrogressions in human rights all over the world.

The pressure on governments to invest in AI also ignores the radically unequal financial circumstances in which countries are competing.

The World Bank's latest International Debt Report, published on 5 November 2026, reveals that developing countries' external debt reached a record $9.4 trillion in 2025. Low- and middle-income countries paid $1.4 trillion in principal and interest that year. For countries eligible for concessional financing, the average interest rate on new private-creditor commitments reached 8.2%, the highest since 1989; new bonds carried an average rate of 9.4%.

These are resources that could otherwise support healthcare, education, social protection, climate adaptation, renewable energy and decent employment.

Telling countries already struggling with debt and the rising cost of capital that they must compete in an expensive AI race risks deepening existing inequalities. Wealthier economies and corporations have the financial resources, technological capabilities and infrastructure to capture the opportunities. Others risk taking on additional debt, subsidising infrastructure and supplying raw materials without capturing a fair share of the value generated.

The question is not simply whether countries should invest in AI, but what they should invest in, under what conditions, and in whose interests.

AI is advancing faster than the institutions and international rules designed to protect people from abuses of economic and political power. Voluntary corporate commitments and fragmented national regulations are not enough to address the transnational nature of AI supply chains, ownership, data extraction and profits.

International human rights law provides an essential foundation. Existing obligations relating to economic and social rights, labor rights, equality, privacy, participation, access to information and environmental protection must be interpreted and applied to these new technological frontiers.

States must cooperate across borders, businesses must answer for their value chains, and affected communities need a say and access to remedy.

This requires more than regulating algorithms. It means governing the entire political economy of AI: from mineral extraction and manufacturing to data collection, model development, deployment, ownership and the allocation of profits.

There is also an urgent fiscal dimension. If AI generates enormous profits through the use of knowledge, data, infrastructure and markets across multiple jurisdictions, how should those profits be taxed? Where should taxing rights lie? How can countries, particularly those whose resources, labor and markets contribute to value creation, secure a fair share of the revenues?

These questions cannot be left to the same international tax arrangements that have historically disadvantaged many developing countries and enabled multinational corporations to shift profits across borders.

The negotiations for the United Nations Framework Convention on International Tax Cooperation offer a particularly important opportunity. The Protocol on the Taxation of Cross-Border Services (Protocol 1) must address the challenges posed by the digital economy and emerging AI business models, including how taxing rights are allocated when services, data and value cross borders without a conventional physical presence.

Fairer taxation of AI-related profits should be part of a broader effort to ensure that the wealth generated by technological innovation contributes to financing public services, social protection, education, climate action and the infrastructure needed for a genuinely inclusive transition. 

Promising proposals on how to approach the taxation of AI already exist. These include enabling countries to tax AI-generated profits even where companies have no physical presence, recognising user data and algorithm training as sources of taxable economic value, and reforming approaches to profit allocation so as to reflect AI's fragmented global value chains.

Tax justice is not simply about raising revenue. It is about ensuring that those who benefit most from the global economy contribute fairly to the societies and public resources on which their wealth depends.