Research Wing | Republic Policy
The global race to build artificial intelligence infrastructure is confronting a growing mismatch between projected demand, available electricity, semiconductor production and the economics of data centres. What once appeared to be an unstoppable investment boom is increasingly exposed to the risks of overcapacity, supply-chain constraints and uncertain returns.
The artificial intelligence revolution has triggered one of the largest investment races in modern economic history. Since the public release of ChatGPT in late 2022, the world’s largest technology companies have committed hundreds of billions of dollars to building the infrastructure required to sustain the expansion of AI. Microsoft, Google, Amazon, Meta and Oracle have announced enormous investment programmes involving data centres, specialised semiconductors, cooling systems and electricity infrastructure. However, beneath the excitement surrounding artificial intelligence, a fundamental economic question is beginning to emerge: are technology companies building infrastructure faster than the market can absorb it?
Recent developments suggest that the answer may be more complicated than investors initially anticipated. Several data centre projects around the world are facing delays, restructuring or cancellation. Reports of the postponement of expansion plans at Oracle and OpenAI’s major Texas data centre project have added to concerns about the sustainability of the investment boom. At the same time, estimates suggest that a substantial proportion of planned American data centre projects for 2026 could be delayed or abandoned. These developments indicate that the AI infrastructure market may be approaching a period of adjustment.
The scale of investment explains why these developments matter. By 2023, spending on AI-related infrastructure had already exceeded $200 billion. In 2024, investment expanded further, while in 2025, major technology companies reportedly allocated close to $400 billion in capital expenditure, much of it directed towards data centres, advanced chips and supporting infrastructure. This level of spending was remarkable even by the standards of the technology industry. Indeed, estimates suggest that investment in data centre infrastructure during 2025 exceeded spending on single-family housing construction in the United States. An emerging technology had therefore become one of the largest destinations for global capital.
However, the underlying business assumptions behind this expansion deserve closer examination. Technology companies have been making investment decisions based on projections of future AI demand rather than the revenue generated by their existing operations. Investors increasingly expect these companies to demonstrate how enormous infrastructure commitments will translate into sustainable profits. Building data centres, purchasing advanced GPUs and securing electricity supplies do not automatically guarantee commercial success. The central challenge is whether the demand for AI computing will grow sufficiently to justify the enormous capital being committed.
The Electricity Constraint: AI’s Most Serious Infrastructure Problem
The first major constraint is not necessarily the availability of advanced chips but the electricity required to operate them. Data centres are extraordinarily energy-intensive facilities. As their numbers increase, electricity demand is placing additional pressure on power grids that were not designed to accommodate such rapid growth in consumption.
According to industry estimates, data centres already account for a significant share of global electricity consumption, and their demand is expected to rise substantially as AI applications expand. In the United States, concerns are growing that the expansion of data centres could intensify competition for electricity, increase energy prices and complicate the provision of reliable power to households and businesses.
This is not simply a technical problem. It is an economic and public-policy challenge. Electricity grids require substantial investment, modernisation and years of planning. Data centres, however, are being developed at a pace that frequently exceeds the ability of electricity providers to deliver the necessary infrastructure. Consequently, projects that appear financially viable on paper may encounter serious obstacles when developers attempt to secure electricity connections.
The shortage of electrical equipment makes matters worse. Transformers, generators, power supplies and transmission components are essential for operating large computing facilities. Yet their production and delivery are constrained by manufacturing capacity, supply-chain dependencies and international trade uncertainties. The prices of certain electrical components have increased considerably, raising construction costs and extending project completion schedules.
The importance of this bottleneck is often underestimated. A data centre may have land, financing, buildings and advanced processors, but without a reliable electricity supply, its computing infrastructure cannot operate at full capacity. In other words, the digital economy remains dependent on physical infrastructure that cannot be expanded overnight.
Nvidia and the Risk of Overestimating AI Demand
Another important dimension of the emerging challenge concerns Nvidia. The company has become the central supplier of specialised graphics processing units (GPUs) used to train and operate advanced AI systems. Its extraordinary financial performance reflects the unprecedented demand for computing power.
However, Nvidia’s position also exposes it to the risks associated with excessive investment across the AI industry. Technology companies have been purchasing enormous quantities of processors based on expectations of future demand. These purchases are partly motivated by competition: companies fear that delaying investment today could leave them without sufficient computing capacity tomorrow.
This has created a cycle in which expectations of future AI growth encourage infrastructure investment, while infrastructure investment itself reinforces expectations of continued demand.
The difficulty is determining whether projected demand will materialise at the scale anticipated. If AI services generate insufficient revenue to justify the cost of computing infrastructure, companies may reconsider their expansion plans. This could leave developers with expensive facilities, unused capacity and specialised equipment that depreciates rapidly.
Nvidia’s financial position deserves attention in this context. Its reported inventory value increased substantially, reaching approximately $10.08 billion in fiscal 2025, compared with $5.28 billion in fiscal 2024. This represents a significant increase in the value of inventory, although it should not automatically be interpreted as evidence that the company cannot sell its products. Inventory also reflects production schedules, supply commitments and the value of increasingly sophisticated components. Nevertheless, the expansion warrants attention alongside the industry’s enormous investment commitments. MacroTrends
The wider concern is the possibility of an imbalance between supply and demand. If semiconductor manufacturers continue expanding production while data centre projects are delayed or cancelled, the market could eventually experience excess capacity. Such an adjustment would affect not only Nvidia but also semiconductor manufacturers, cloud-computing providers, equipment suppliers and investors exposed to the AI industry.
The Financial Risks Behind the AI Race
The cancellation and postponement of individual projects do not necessarily indicate the collapse of the AI revolution. The Texas development involving Oracle and OpenAI illustrates the distinction. Bloomberg reported in March 2026 that the companies abandoned negotiations over a planned 600-megawatt expansion near their flagship Stargate facility in Abilene. Financing difficulties and changing infrastructure requirements were among the reported factors. The proposed capacity was expected to be accommodated elsewhere, while the broader infrastructure initiative remained in progress. Bloomberg+1
Nevertheless, the episode demonstrates that even the largest technology companies must reconsider their investment strategies when financial requirements, computing needs and infrastructure constraints change.
A February 2026 report by Sightline Climate, reported by Axios, estimated that between 30 and 50 per cent of the data centre projects scheduled for completion during 2026 could be delayed until at least 2027. The principal obstacles included electricity constraints, shortages of grid equipment and community opposition. This suggests that the challenge extends beyond individual companies or isolated developments. Axios+1
There is also a broader question concerning the relationship between capital expenditure and economic returns. The technology industry is investing in infrastructure whose profitability depends on the future commercial adoption of artificial intelligence. Although AI applications are already transforming businesses, education, research and industrial production, their long-term revenue models remain uncertain in several sectors.
Investors therefore face two simultaneous risks. The first is that demand for AI computing may grow more slowly than anticipated. The second is that the infrastructure required to support this demand may become considerably more expensive than originally projected. Either development could reduce expected returns and force companies to reconsider their expansion strategies.
Geopolitical Competition and Supply-Chain Vulnerabilities
The AI infrastructure race is also becoming increasingly dependent on international supply chains. China, South Korea and Mexico are important participants in the production of electrical equipment and other industrial components. Trade tensions, tariffs and restrictions on technology transfers introduce additional uncertainty into an already complicated investment environment.
The United States and China are competing for technological leadership, particularly in advanced semiconductors, computing infrastructure and artificial intelligence. However, this competition also exposes the vulnerabilities of an industry that depends on internationally connected manufacturing networks.
Technology companies have incentives to secure supplies of essential components before prices rise or shortages become more severe. This can encourage advance purchasing and stockpiling, even when the facilities required to install those components are not yet operational. Such behaviour may protect individual companies against future disruptions, but it can also contribute to inefficient capital allocation across the industry.
The result is an unusual economic situation: companies are investing aggressively to avoid being left behind, even though the commercial requirements and physical infrastructure necessary to justify these investments remain uncertain.
Beyond the AI Investment Bubble
The emerging difficulties should not be interpreted as evidence that artificial intelligence is a temporary technological phenomenon. AI has genuine economic potential and will continue to transform industries, public administration, scientific research and international commerce. The real question is whether the current pace and structure of infrastructure investment are economically sustainable.
The experience of previous technological revolutions offers an important lesson. The development of railways, telecommunications and the internet involved enormous capital investment, periods of excessive speculation and eventual market corrections. These corrections did not eliminate the technologies themselves. Instead, they exposed unsustainable business models and helped establish more realistic investment structures.
The AI industry may be approaching a similar phase. Companies will increasingly need to distinguish between infrastructure that serves demonstrable commercial demand and projects driven primarily by competitive pressure and optimistic forecasts. Investors, meanwhile, will demand greater transparency regarding capital expenditure, electricity requirements, utilisation rates and actual revenue generation.
Governments also have an important responsibility. The expansion of data centres must be aligned with national electricity planning, environmental considerations and the interests of ordinary consumers. Public infrastructure should not become an indirect subsidy for private technology companies without transparent arrangements concerning costs, benefits and accountability.
Ultimately, the future of artificial intelligence will not be determined by the number of data centres announced, the volume of GPUs purchased or the billions of dollars committed to construction. It will depend on whether these investments generate sustainable economic value.
The technology industry is discovering that the digital revolution has very physical foundations. Artificial intelligence requires electricity, industrial equipment, skilled labour, investment and functioning supply chains. Capital alone cannot overcome these constraints. The companies that recognise this reality will be better positioned to build sustainable AI businesses, while those that confuse ambitious projections with actual demand may face substantial financial corrections.
The AI revolution is far from over. However, its next phase may be less about building infrastructure at any cost and more about demonstrating that the infrastructure already being built can deliver genuine economic returns.
Research Wing | Republic Policy
