Could Electricity Become AI’s Biggest Constraint? Elon Musk’s G20 Warning

Artificial intelligence continues to advance at an extraordinary pace. But could its greatest near-term constraint be something considerably less sophisticated than the technology itself — electricity?

Speaking virtually at the G20 Innovation Ministerial in Chapel Hill, North Carolina, on 1 September 2026, Elon Musk discussed AI-driven economic growth, humanoid robotics and technological transformation. Yet one of his most relevant observations concerned the physical infrastructure needed to support that transformation.

A 15-gigawatt electricity shortfall in 2027?

Musk warned that AI computing could face a power shortfall of at least 15 gigawatts in 2027, citing what he described as estimates from analysts following the sector. He suggested that AI-chip production was rising by around 40–50% annually while electricity capacity outside China was growing by 10–20%.

These figures are Musk’s characterisation of forecasts, not independently established projections. His underlying point, however, is important: computing equipment can be manufactured and deployed faster than new power stations, grid connections and data centres can necessarily be built.

Physical and geopolitical constraints

Musk highlighted another complication: China has extensive electricity-generation capacity, but export restrictions on advanced AI chips constrain deployment of the latest hardware there. Elsewhere, developers must secure power, suitable sites, cooling and data-centre infrastructure.

He argued that countries able to provide reliable electricity and supporting infrastructure could attract future AI investment. More broadly, the location and resilience of AI services may increasingly reflect energy markets, supply chains and government policy as much as model capability.

Energy security, Net Zero and the risk of widespread disruption

Musk’s warning raises a difficult policy question: how can governments balance carbon reduction with affordable, reliable electricity and security of supply? The UK and EU continue to pursue ambitious decarbonisation policies, while decisions about North Sea developments such as Rosebank reflect wider debates over domestic production and import dependence. Rosebank is principally an oilfield, however, and is not itself a direct answer to electricity-generation capacity.

Wind and solar can provide substantial low-carbon electricity, but their variable output requires an electricity system with sufficient firm capacity, storage, interconnection, demand flexibility and effective grid controls. Whether particular policies increase or reduce reliability depends on how the entire system is designed and operated — not on the presence of renewables alone.

The April 2025 Spain–Portugal blackout illustrates the systemic consequences of electricity failure. The European investigation identified interacting issues involving voltage and reactive-power control, oscillations and generation disconnections. It would be misleading to attribute that event solely to renewable energy policy, but it is an important lesson in the complexity of changing electricity systems.

Geopolitical shocks add another dimension. Russia’s war in Ukraine and conflict affecting the Middle East have exposed the vulnerability of energy markets to supply disruption and price volatility. Electricity and energy security are therefore not merely environmental or economic concerns: they are dependencies underpinning almost every important business service.

Innovation, nuclear power and the difficulty of predicting the future

Reducing carbon emissions while expanding reliable electricity supply need not be a permanent choice between environmental protection and prosperity. Technological innovation — including advanced nuclear fission, potentially fusion, better energy storage, more flexible grids and more efficient computing — could change what is possible. Yet firms cannot assume that emerging technologies will arrive in time to solve today’s capacity constraints.

Physicist and quantum-computing pioneer David Deutsch offered a provocative historical parallel in a recent conversation with Chris Williamson on Modern Wisdom. Deutsch argued that opposition to nuclear power, influenced partly by environmental fears, curtailed a major source of low-carbon electricity and thereby contributed to today’s climate challenge.

This is Deutsch’s counterfactual argument, not an established historical conclusion. Nuclear development was also shaped by economics, construction costs, safety considerations and government decisions. His wider insight is nevertheless valuable: attempts to prevent one uncertain future harm can produce unintended consequences, while progress depends on creating knowledge, correcting errors and allowing new technologies to emerge.

For energy policy, that suggests pursuing decarbonisation and innovation while preserving reliable, affordable supply throughout the transition. For financial institutions, it means planning for today’s infrastructure risks without assuming tomorrow’s technology will remove them.

What would a prolonged power failure mean for operational resilience?

A widespread electricity disruption could affect banks, payment infrastructure, telecommunications, transport, cloud services, critical suppliers and customers at the same time. Backup generators and alternative sites may be less effective if fuel resupply, network connectivity and supposedly independent providers are also constrained.

Scenario testing should therefore consider a prolonged regional outage with correlated failures across data centres, telecommunications, payments, fuel supply and customer access — not merely a short interruption at one office or server room. This matters even where AI itself is not yet supporting an important business service.

Why should financial institutions care?

Financial institutions are integrating AI into customer-facing processes, risk management and internal operations. As agentic AI assumes more operational responsibilities, dependence on underlying compute and energy infrastructure could become more consequential.

• Capacity and performance: constraints could limit expansion or degrade service at peak demand.
• Costs: competition for computing and power could influence pricing and service commitments.
• Concentration: providers with secure access to power and chips may gain greater market share.
• Resilience: common infrastructure dependencies may affect multiple apparently separate AI services.

A projected shortfall in electricity for AI expansion does not mean existing financial-services AI systems will necessarily suffer outages. But it is a credible scenario for risk assessments, supplier oversight and continuity planning.