THE AI RECKONING: WHO PAYS FOR THE FUTURE?
Updated: 6 days ago
Artificial intelligence is being positioned as the foundation of the next industrial transformation. But behind the promise of automation, superintelligence and unprecedented productivity lies an increasingly complicated set of dependencies involving electricity, water, land, capital, public consent and institutional capacity. The question is no longer simply what AI can do. It is whether the systems supporting its expansion are prepared for what it could become.
There is a rather interesting contradiction developing around artificial intelligence. The technology industry is asking governments, investors, businesses and ordinary citizens to embrace a transformation that could fundamentally alter the organisation of modern society. We are being introduced to increasingly capable systems, autonomous agents, automated workflows and the prospect of machines performing tasks that previously required human intervention. The language is ambitious, the investment is extraordinary, and the commercial expectations appear almost limitless.
Yet, behind the interfaces, demonstrations and increasingly impressive capabilities, there is a much less glamorous reality. Artificial intelligence requires physical infrastructure, enormous quantities of electricity, cooling systems, water, land, semiconductor manufacturing, specialised equipment, capital markets and, perhaps most importantly, public cooperation. It also requires something that cannot be manufactured in a semiconductor facility: trust. The expansion of AI depends on people accepting its presence in their workplaces, homes, public institutions and increasingly consequential decisions about their lives.
The recent political backlash against data-centre development in the United States should therefore be understood as something considerably more significant than a disagreement over industrial zoning.
It is an early confrontation between the ambitions of the technology industry and the practical realities of the societies expected to accommodate them. The question is not whether artificial intelligence has potential. It is whether the economic, environmental, political and institutional arrangements required to sustain that potential have been adequately considered.
THE MARKET IS BEGINNING TO PRICE IN THE POLITICS
The November 2026 United States midterm elections have brought an important development into focus. Opposition to large data-centre projects became a campaign issue in several jurisdictions, with candidates promising stronger oversight, restrictions and, in some cases, temporary pauses on development. Reuters reported on November 4 that candidates campaigning partly on data-centre moratoriums had won important state-level contests, while investors began reassessing the regulatory risks surrounding the infrastructure expansion supporting the AI industry.
The reaction extended into financial markets. Data-centre operators, infrastructure companies and associated energy businesses experienced share-price declines as investors considered the possibility of project delays and tighter regulation. Goldman Sachs estimated that temporary project halts could affect approximately 45 per cent of planned United States data-centre capacity expansion, although the duration and geographical distribution of any restrictions remain uncertain.
It would be a mistake, however, to interpret every movement in technology stocks as evidence that the AI economy is collapsing. Markets respond to multiple factors, including valuations, capital expenditure, financing conditions, projected demand and expectations of future profitability. A decline in share prices does not establish that demand for artificial intelligence has disappeared. Nor does a temporary pause in construction necessarily indicate that the underlying technology has reached its commercial limits.
What the developments demonstrate is that investors are beginning to account for something that has received less attention during the excitement surrounding AI: infrastructure expansion is also a political and regulatory undertaking.
For years, the conversation has concentrated on the companies building the models, the chips powering them and the applications reaching consumers. Increasingly, the discussion is moving towards the communities, utilities, governments and financial institutions that must accommodate the physical consequences of that expansion. The technology may be developed by private companies, but the infrastructure cannot exist independently of the societies in which it operates.
THE PHYSICAL REALITY BEHIND ARTIFICIAL INTELLIGENCE
Artificial intelligence is frequently presented as an almost weightless digital service. A person enters a prompt, receives an answer and moves on with the day. A business automates a workflow, a developer builds an application, or a government introduces an intelligent assistant. The interaction appears relatively simple, but the infrastructure supporting it is anything but.
Data centres require buildings, land, electrical substations, transmission connections, backup systems, cooling equipment, servers and extensive networking infrastructure. Those servers depend on semiconductors manufactured through complex international supply chains. Their operation requires electricity, and their thermal management requires cooling systems that can involve substantial water and energy consumption.
The International Energy Agency’s October 2026 report, Powering Artificial Intelligence, provides an important assessment of the scale of this transformation. It examines 34 policy jurisdictions representing approximately 90 per cent of global data-centre capacity. The report estimates that the world’s largest data-centre operators are set to spend more than $700 billion in capital expenditure during 2026, predominantly on new facilities. That is an extraordinary investment programme. It also illustrates how much of the AI economy depends on the ability to construct and operate physical infrastructure at a pace that electricity systems, supply chains and public authorities may struggle to accommodate.
The IEA identifies growing pressure on local electricity networks, lengthy connection queues, shortages of electrical equipment and uncertainty about the timing and scale of future demand. In some jurisdictions, proposed data-centre connections collectively represent electricity capacity exceeding existing peak demand. Not every project in those queues will necessarily proceed, and some proposals may be speculative or duplicative. Nevertheless, utilities and regulators must plan around a pipeline that could materially change the geographical distribution of electricity consumption.
This creates a difficult planning problem. Underestimating future demand can produce infrastructure bottlenecks, delays and insufficient generation capacity. Overestimating it can lead to expensive grid investments that remain underutilised if the anticipated projects never materialise. The challenge is particularly significant because AI-focused facilities are not simply another category of commercial building. Their electricity requirements can be substantial, concentrated and geographically unusual. A region that historically planned its electricity network around residential consumption, manufacturing and established commercial centres may suddenly face a proposed facility requiring an entirely different level of capacity.
The technology industry is effectively asking electricity systems to reorganise themselves around the requirements of a new industrial customer. The question is whether those costs and risks are being allocated fairly.
WHO PAYS FOR THE ELECTRICITY?
Electricity is perhaps the most immediate point of tension between the AI industry and the public. A data centre may be financed by a private corporation, but its electricity requirements interact with a shared infrastructure system. New transmission lines, substations, generation capacity and network upgrades must be financed, constructed and maintained. The critical issue is determining which costs should be borne by the data-centre operator and which, if any, can ultimately be passed to other electricity consumers.
The IEA makes an important distinction here. The effect of data centres on electricity bills is not automatically negative in every circumstance. Where a new facility makes productive use of existing capacity and pays an appropriate share of the associated costs, it can help distribute fixed infrastructure expenses across a larger customer base. However, where a new large load requires additional generation and network investment, and those costs are recovered from other consumers, household and business electricity bills can face additional pressure.
The allocation of these costs is therefore a matter of public policy, not simply an engineering decision. Regulators must consider minimum-payment commitments, long-term electricity contracts, collateral, exit fees and charges reflecting the capacity reserved by large customers. These arrangements can reduce the risk that households and existing businesses end up financing infrastructure built around commercial projections that subsequently fail to materialise.
There is also the question of energy security. Some data-centre operators are exploring arrangements involving dedicated generation, renewable-energy procurement, battery storage, advanced geothermal technologies and proposed small modular nuclear reactors. Others are considering on-site natural-gas generation. These developments may improve supply resilience in certain circumstances, but they also introduce additional questions about emissions, grid integration, fuel availability and the actual environmental performance of the facilities.
The phrase frequently used in this debate is bring-your-own-power. The underlying principle is straightforward: a large new electricity customer should contribute meaningfully to the generation and system capacity required to support its operations rather than simply arriving at the connection point and expecting the surrounding community to absorb the consequences. That principle deserves serious consideration. An industry cannot indefinitely describe its infrastructure as a private commercial investment while treating the electricity network supporting it as somebody else’s responsibility.
THE WATER QUESTION
Electricity is not the only resource requiring attention. Water presents another dimension of the infrastructure challenge, particularly in regions already experiencing water stress. Data centres generate heat. Depending on their cooling architecture, facilities may use water-intensive evaporative cooling, air-based systems, liquid cooling or combinations of different technologies. The choice affects water consumption, electricity requirements, operational efficiency and the surrounding environment.
It is important to distinguish between direct water consumption at the facility and the wider water footprint associated with electricity generation and the production of equipment. A data centre may report relatively modest on-site consumption while relying on an electricity supply whose generation requires significant quantities of water elsewhere in the system.
The IEA’s October assessment identifies an important geographical consideration: nearly 40 per cent of identified AI-dedicated data-centre capacity is located in areas experiencing high or extremely high baseline water stress. This is not a minor planning consideration. Water is a strategic resource shared by households, agriculture, industry, ecosystems and public infrastructure. Its availability varies geographically and seasonally, and the consequences of industrial consumption depend heavily on local conditions.
The appropriate question is therefore not simply how much water the global data-centre industry consumes. It is where that water is being consumed, what sources are being used, what competing demands exist and whether the facilities have been assessed against the long-term sustainability of the relevant watershed. A national average can conceal a local problem. A company may operate within an acceptable national water-consumption profile while placing substantial pressure on a particular municipality or watershed.
This is why disclosure matters. Communities should have access to credible information about projected and actual water consumption, cooling methods, seasonal requirements and the environmental conditions under which facilities operate. The same principle should apply to electricity, emissions and land use. If the industry expects communities to accommodate infrastructure of this scale, transparency cannot be treated as an optional public-relations exercise.
THE SOCIAL LICENCE TO BUILD
The political dimension of the data-centre expansion is becoming increasingly difficult to ignore. A September 2026 Pew Research Center survey of 10,548 American adults found that public perceptions of data centres had become more negative since January. Fifty-four per cent of respondents considered data centres mostly bad for the environment, compared with 39 per cent in January. Half considered them mostly bad for household energy costs, while 49 per cent believed they had a negative effect on the quality of life of nearby residents.
These figures do not establish that every data centre causes the same environmental or economic consequences. They measure public perceptions, and those perceptions should be distinguished from independently verified facility-level impacts. Nevertheless, they indicate a substantial shift in public sentiment. The issue is not necessarily opposition to technology itself. People can appreciate the convenience of digital services, use AI applications and recognise the potential benefits of automation while questioning whether a particular industrial development is appropriate for their community.
Residents may ask why electricity prices should increase to accommodate a facility that employs relatively few permanent workers. They may question whether tax incentives are proportionate to the economic benefits being delivered. They may want to know how much water will be consumed, whether local infrastructure can accommodate the development and what happens if the company eventually abandons the project. These are legitimate questions.
The technology industry must recognise that the ability to finance and construct a facility does not automatically confer public permission to impose its consequences on the surrounding population. This is what can be understood as a social licence to operate. It is not simply a permit issued by a planning authority. It is the broader acceptance that a development is sufficiently transparent, economically justified and responsibly managed to coexist with the interests of the people living around it.
Governments have an important role in protecting that relationship. They should establish clear rules for grid connections, environmental assessments, water disclosure, community consultation, tax incentives and infrastructure contributions. They should also ensure that public subsidies and commercial concessions are evaluated against measurable public benefits. The objective should not be to prevent technological development. It should be to ensure that the public is not expected to subsidise private expansion without adequate consideration of the costs.
THE GOVERNMENT IS ENTERING THE MACHINE
The infrastructure debate is only one part of the larger transformation. The more consequential development may be the increasing integration of artificial intelligence into the machinery of government itself. Governments are not simply regulating AI from a distance. They are becoming users of it. The OECD’s Digital Government Outlook 2026 reports that 35 of 36 OECD countries use AI in at least one area of government. Adoption is strongest in internal administrative processes and public services, while use in policymaking and oversight remains more limited.
That distinction is important because the consequences of deploying AI vary considerably depending on the function being automated. Using AI to classify documents, summarise records or assist with routine administrative work presents a different risk profile from using it to influence immigration decisions, public-benefit eligibility, law enforcement, healthcare prioritisation or the administration of justice.
Consider the ordinary relationship between a citizen and a government department. A person applies for a benefit, submits documentation, provides information and expects an administrative decision. If an AI system assists with the assessment, the government must still be able to explain how the decision was reached, identify errors and provide an effective mechanism for appeal.
The citizen should not be left in a position where an institution cannot explain a consequential decision because the relevant process has been delegated to a system supplied by a private technology company. The same principle applies to immigration, taxation, policing, healthcare and national security. Governments may use automated systems to improve efficiency, identify patterns and process information, but the responsibility for public decisions cannot simply disappear into a technical system. A government cannot outsource its constitutional and administrative responsibilities to an algorithm.
This becomes particularly important when the technology is supplied by a private company. Public institutions must understand the systems they procure, establish contractual protections, retain appropriate access to records, assess cybersecurity risks and maintain the capacity to challenge vendor claims.
The OECD’s findings also identify gaps in the practical governance arrangements supporting public-sector AI. While many countries have strategies and institutional frameworks, fewer require systematic pre-deployment risk assessments, independent review and post-deployment audits. Only a minority report formal algorithm registers or established mechanisms for measuring the impact of AI deployments. The distinction between announcing an AI strategy and having the institutional capacity to govern its implementation is therefore significant. Governments may be adopting the technology faster than they are developing the mechanisms needed to supervise it.
THE QUESTION OF DIGITAL SOVEREIGNTY
There is another issue that deserves considerably more attention: digital sovereignty. A country’s ability to govern itself increasingly depends on infrastructure that may be owned, operated or controlled by private companies located within different jurisdictions. Cloud computing, telecommunications, cybersecurity, digital identity, financial infrastructure and AI services are becoming integral to the operation of modern economies. As governments become more dependent on these services, the relationship between public authority and private technological infrastructure becomes increasingly complicated.
Consider the implications if essential public services rely on a small number of external technology providers. What happens if a provider changes its pricing, discontinues a service, modifies its terms or experiences a major operational failure? What happens if geopolitical tensions affect access to computing infrastructure, semiconductors or critical software? What happens if a government cannot independently audit the systems on which its administrative operations depend?
These are questions of national resilience. Digital sovereignty does not necessarily mean that every country must manufacture its own semiconductors, operate its own hyperscale data centres or develop every AI model domestically. That would be economically unrealistic for many jurisdictions. It does, however, mean that governments should understand their dependencies, preserve meaningful control over essential public functions and avoid becoming so reliant on individual providers that they lose practical alternatives.
This is particularly relevant for smaller economies. Countries such as Belize must consider how to participate in the emerging AI economy without simply becoming consumers of infrastructure, software and services developed elsewhere. The opportunity is substantial, but participation requires more than purchasing subscriptions or introducing automated customer-service systems. It requires digital infrastructure, technical education, cybersecurity, data governance, local enterprise development and the capacity to negotiate commercial arrangements that create genuine domestic value.
A country that adopts technology without developing the institutional and economic capabilities to manage it risks becoming dependent on the very systems it hopes will modernise its economy.
THE SOCIAL CONTRACT IS ALSO AT RISK
The discussion about AI and employment is frequently reduced to a prediction about which occupations will disappear. That is an important issue, but it is not the only economic question. Modern governments operate within economic arrangements built around employment, taxation, consumption, business ownership and public expenditure. Human labour generates income, supports household consumption and contributes to tax revenues. Businesses employ people, invest capital, purchase goods and services and contribute to the wider economic system. Artificial intelligence could alter these relationships substantially.
If organisations can produce more with fewer employees, productivity may increase while the distribution of income changes. If ownership of the most productive AI systems remains concentrated, a disproportionate share of the resulting economic gains could accrue to a relatively small group of companies and investors. This raises questions about the future of taxation, consumer demand, social protection and economic mobility.
If automation reduces the demand for certain categories of labour, how will affected workers maintain purchasing power? If productivity gains accrue primarily to capital owners, how will governments ensure that economic growth translates into broader prosperity? If businesses become increasingly automated, will traditional employment-based tax systems remain sufficient to finance public services?
There is no single predetermined answer. AI could create entirely new industries, increase productivity, lower the cost of certain services and generate new categories of employment. Its effects will vary across sectors, occupations and countries. However, it would be unwise to assume that technological productivity automatically produces equitable economic outcomes. Productivity and distribution are different questions. A society can become substantially more productive while experiencing greater economic concentration.
The social contract must therefore be part of the AI conversation. Governments should be examining how taxation, education, labour protections, competition policy and social insurance might need to evolve as automated systems become more capable. Waiting until employment disruption becomes widespread before considering these questions would be a particularly expensive approach to public policy.
THE FINANCIAL BET BEHIND THE AI ECONOMY
The physical infrastructure supporting artificial intelligence is also one of the largest capital-allocation exercises currently taking place. The more than $700 billion in projected 2026 capital expenditure by the largest data-centre operators illustrates the scale of the financial commitment. Investors are financing facilities, semiconductor procurement, power infrastructure, networking equipment and the wider supply chain required to support anticipated demand. But capital expenditure is not the same as a guaranteed return.
The commercial case depends on several assumptions. AI adoption must continue expanding, customers must be willing to pay for increasingly capable services, enterprise productivity gains must translate into sustainable revenue and infrastructure must remain sufficiently utilised to justify its cost. There are also technological uncertainties.
Semiconductor performance continues to evolve, model architectures change, efficiency improvements alter computing requirements and the economics of inference may differ substantially from the economics of training increasingly large models. A facility constructed around today’s requirements may face different commercial conditions several years from now. The IEA identifies the risk of speculative or duplicative projects within data-centre connection queues. It warns that overestimating demand can create stranded or underutilised grid investments, while underestimating demand can produce infrastructure bottlenecks.
This is an important distinction because the infrastructure may have a longer economic life than the technology or commercial assumptions supporting its construction. A power station, transmission line or specialised facility represents a substantial commitment of capital. If the anticipated demand does not materialise, the consequences do not necessarily disappear with the original investor. There may be implications for lenders, utilities, local governments, taxpayers and electricity consumers.
The AI industry is therefore not simply betting on technological progress. It is making a series of interconnected financial, regulatory and infrastructure bets. The question is whether the risks are being properly measured and whether the parties benefiting from the expansion are also carrying an appropriate share of the downside.
EFFICIENCY DOES NOT AUTOMATICALLY MEAN LESS CONSUMPTION
One of the strongest arguments in favour of AI infrastructure expansion is that technological efficiency will improve. Models become more capable. Hardware becomes more efficient. Algorithms require fewer resources for certain tasks. New cooling systems and energy-management technologies can reduce operating costs.
These improvements are real and important. However, efficiency per task does not necessarily translate into a reduction in total resource consumption. When a technology becomes cheaper and more efficient, it can become economically attractive for applications that previously would have been too expensive. More businesses adopt it, more consumers use it and increasingly complex tasks become commercially viable.
The result can be an expansion in total demand even while the resources required for an individual task decline. The IEA highlights a related tension in the AI industry. While market pressures encourage more efficient computing, companies also face strong incentives to increase customer numbers, usage and the availability of high-performing models. The commercial objective of expanding consumption can therefore coexist with the engineering objective of reducing consumption per task.
This is not unique to AI. It is a familiar challenge in energy economics and industrial development. The appropriate measure of sustainability cannot be limited to the efficiency of an individual model or application. It must also consider the total amount of computation being performed, the energy sources supporting it, the water consumed throughout the value chain and the environmental consequences of expanding the infrastructure. The question should be whether AI is becoming more resource-efficient at the level of the entire economy, not simply whether a particular model is more efficient than its predecessor.
CAN GOVERNANCE KEEP PACE WITH THE TECHNOLOGY?
This is ultimately where the discussion becomes larger than the technology industry itself. Governments operate through institutions, laws, administrative procedures and political processes. These systems were developed to manage human societies, allocate public resources, regulate commercial activity and establish accountability.
Artificial intelligence introduces new capabilities into almost every one of those areas. It can assist with legislative research, analyse economic data, identify patterns in public expenditure, support infrastructure planning, improve administrative processes and help government departments process information at a scale previously difficult to achieve.
Used appropriately, these capabilities could materially improve public administration. But there is a difference between using AI as an instrument of government and allowing government to become dependent on systems it cannot independently understand or supervise. The more consequential the decisions, the more important the safeguards become. Public institutions must preserve human responsibility, establish transparent procurement arrangements, conduct independent assessments, maintain accessible appeal mechanisms and ensure that citizens can challenge decisions affecting their rights.
They must also develop the technical capacity to understand the systems being introduced. A regulator cannot effectively supervise an industry if it lacks the expertise, resources and access required to examine the technology. The challenge is not simply that legislation moves more slowly than software development. It is that governments must build entirely new forms of institutional competence. They must understand data infrastructure, model behaviour, cybersecurity, automated decision-making, intellectual property, algorithmic accountability and the commercial relationships between technology providers and public authorities.
International coordination will also matter. AI systems and the companies supplying them operate across borders, while governments remain responsible for their own populations and legal jurisdictions. A regulatory framework developed in one country may influence products and commercial arrangements elsewhere. Yet the resources available to supervise those systems vary considerably between jurisdictions. There is a genuine risk that the countries with the greatest technological power will have the greatest influence over the rules, while smaller economies will inherit systems and standards they had limited opportunity to shape. That is not simply a technology-policy issue. It is a question of international economic and political power.
THE FUTURE IS NOT JUST A TECHNOLOGY QUESTION
There is a temptation to approach artificial intelligence as though the primary objective is to develop the most capable model, construct the largest computing facility or achieve the greatest possible level of automation. Those are important measures of technological advancement, but they are not sufficient measures of societal progress. A technology can be extraordinarily capable and still be deployed without adequate consideration of its economic consequences. An infrastructure project can generate substantial investment while placing disproportionate costs on surrounding communities. A government can improve administrative efficiency while weakening the ability of citizens to understand or challenge consequential decisions.
The question is not whether artificial intelligence should advance. It is whether the institutions, markets and public safeguards surrounding that advancement are developing with comparable seriousness. The recent data-centre backlash demonstrates that the public is not necessarily prepared to accept every consequence of the AI expansion simply because the technology is described as revolutionary. Communities are asking about electricity bills, water, land, employment and local accountability. Investors are examining regulatory risk. Governments are being forced to consider whether existing infrastructure and administrative arrangements can accommodate the scale of the proposed development.
These developments should be treated as part of the same conversation. AI depends on electricity systems, water resources, supply chains, financial markets, public institutions and social cooperation. It is therefore exposed to the economic, environmental and political conditions governing all of them. The industry may be developing increasingly autonomous systems, but those systems still depend on human institutions and physical resources.
And this is the central contradiction.
We are discussing the possibility of artificial intelligence transforming the way societies function while the governance systems responsible for managing that transformation are still working through some of the most fundamental questions about accountability, public consent, economic distribution and institutional control. The future of AI cannot be determined exclusively by the companies developing it or the investors financing its expansion. Governments have responsibilities. Regulators have responsibilities. Infrastructure providers have responsibilities. And citizens have a legitimate interest in understanding the consequences of the systems being introduced into their economies and communities.
The next phase of technological development will require more than faster processors, larger data centres and increasingly sophisticated models. It will require institutions capable of governing the infrastructure, markets capable of allocating its costs fairly and public policies capable of ensuring that the benefits are not confined to those who own the technology.
Artificial intelligence may become one of the most consequential technologies in human history. But technological capability is not the same thing as institutional readiness, and commercial investment is not the same thing as public benefit. The industry is building the infrastructure of the future. The question is whether society has been given a meaningful opportunity to determine the terms.
Comments