Power Infrastructure Emerges as a Critical AI Constraint

The rapid expansion of artificial intelligence is confronting a physical infrastructure challenge as electricity demand grows faster than the power systems needed to support it. Goldman Sachs has identified electricity availability and grid access as major constraints on the industry’s expansion, raising questions about the timing and returns of investments in AI computing capacity.

According to the bank’s report, global data center power demand is expected to rise 170% between 2025 and 2030. More than 60% of that increase is projected to come from the United States, where technology companies are expanding data center capacity to support increasingly demanding AI workloads.

Goldman estimates that approximately $7.6 trillion in capital spending will be required across computing, data centers, and power infrastructure between 2026 and 2031.

However, the availability of financing and computing equipment does not guarantee that new facilities can begin operating on schedule. Access to electricity, grid connections, and the infrastructure required to deliver power are becoming decisive factors in determining where and how quickly AI capacity can be deployed.

Grid Connection Delays Threaten Project Timelines

The gap between proposed power projects and operational infrastructure highlights the scale of the challenge. The United States has approximately 2,600 gigawatts of proposed power generation and storage projects awaiting grid connections, more than twice the grid’s existing installed capacity. Not all these projects will ultimately be completed, but the backlog illustrates the pressure on connection processes.

The median period from submitting a grid connection application to commercial operation is approximately five years. This creates a mismatch between the speed of technological development and the time required to build and connect physical infrastructure.

Goldman also reported that 86 projects had been canceled or significantly delayed since January, including 76 directly linked to local opposition.

For AI developers and infrastructure investors, these delays could extend construction schedules, postpone revenue generation, and increase uncertainty around capital deployment. Companies investing heavily in computing capacity before securing sufficient electricity may face a greater risk of underutilized assets or slower-than-expected financial returns.

Debt Financing Increases Exposure to Delays

The scale of AI infrastructure investment is increasing the importance of construction timelines and project economics. Goldman projects global hyperscaler debt issuance of $420 billion in 2027, more than 60% above its estimate for 2026.

Debt is expected to finance approximately 35% of hyperscaler capital expenditures next year. The bank also estimates that hyperscalers account for only 40% of credit fundraising across the broader AI ecosystem, indicating that borrowing extends beyond the largest cloud and technology companies.

This financing structure raises the potential cost of delays. When projects take longer to become operational, companies may need to service debt before generating the expected returns from new infrastructure.

The risk extends across the AI ecosystem, including data center developers, power providers, and other companies financing the expansion. As capital requirements rise, access to reliable electricity may become as important to investment outcomes as demand for AI services.

Utilities Become Major Beneficiaries of AI Investment

The infrastructure bottleneck is also creating opportunities for the utility sector. Utility capital plans are expected to reach $1.3 trillion between 2026 and 2030, reflecting the growing need for electricity generation, transmission capacity, and supporting infrastructure.

Technology companies are considering arrangements to directly finance the power generation and transmission assets required by their data centers. Such arrangements could help align infrastructure development with demand from large customers, although they would not eliminate permitting, construction, and grid connection constraints.

Slower deployment could also have a countervailing benefit. If infrastructure development takes longer, actual demand for AI services may have more time to catch up with planned investment, potentially reducing the risk of overbuilding capacity ahead of demand.

What Investors Should Watch Next

Investors should monitor whether power generation and grid connection capacity can expand quickly enough to support projected data center demand. Permitting outcomes, local opposition, construction schedules, and direct investment by technology companies will help determine how effectively the industry addresses these constraints.

Financing conditions will also remain important. Rising debt issuance alongside delayed project completion could increase pressure on returns, particularly for companies that commit substantial capital before securing operational power capacity.

The central investment question is increasingly whether infrastructure can keep pace with AI ambitions. Companies that secure reliable electricity and deliver projects on schedule may be better positioned to convert infrastructure spending into operating capacity and financial returns.