The buildout of AI infrastructure is progressing so quickly that demand for computing capacity is growing faster than the power systems needed to support it.

That’s according to McKinsey’s Global Energy Perspective 2026 report, which says electricity consumption by data centers could grow at a stunning compound annual rate of 24% through 2030.

By the end of the decade, data centers could consume 10% to 15% of global electricity, making them the fastest-growing source of new load across developed markets.

Yet the amount of new data center capacity that actually comes online will depend heavily on the availability of electricity. McKinsey’s near-term scenarios indicate that power supply will fall short of demand for IT compute through 2030.

This imbalance is shifting the data center industry’s focus from simply building more computing infrastructure to securing reliable energy. New generation capacity is only part of the solution. Transmission lines, transformers, grid connection approvals, storage systems and construction resources all play a role in determining how quickly power can reach a new facility.

Meeting that demand calls for massive investment in the grid. Global investment in transmission and distribution is projected to rise from $443 billion annually in 2025 to $970 billion in 2050.

Worldwide energy consumption is already moving higher. Total energy demand rose 2.7% in 2025, with oil, natural gas, coal and renewable energy all recording growth. Annual demand growth by 2035 could be twice the rate seen during the prior decade, with emerging economies accounting for much of the increase.

Seeking Alternatives to Long Timelines

Data center operators are looking for ways around long utility timelines. Natural gas turbines and engines, fuel cells and battery systems can provide power at the facility, allowing developers to move forward before a full grid connection is available.

But that strategy creates its own challenges. Natural gas generation needs adequate pipeline capacity, and projects built for high reliability may require redundant connections. Pipeline construction can face right-of-way issues that impact the short schedules desired by data center builders.

On-site generation is not necessarily a replacement for utility power. More than 60% of data center operators now plan on combining local generation with a grid connection. So a facility could initially depend heavily on its own generation and then keep that equipment for backup or grid support after utility service arrives.

The arc of AI’s growth creates open questions for data center development. Advances in models, computing hardware and data center design could reduce the amount of electricity required for a given workload. Operators could also shift some computing tasks to periods when power is more readily available. These improvements make demand after 2030 considerably harder to forecast.

Geopolitics, as always, adds complications. About 30% of global oil and gas production currently lacks full access to international markets. Under one McKinsey scenario, efforts by the U.S. and European Union to move supply chains away from China could raise solar capital costs by 20% to 25% and battery storage costs by 40% to 50%.

The result of all these unknowns is a data center market in which access to computing hardware is only part of the capacity equation. Developers must now plan around power generation, grid connections, fuel supplies, and equipment availability while attempting to forecast the future effect of AI on data center design.