Big Tech’s $635b AI capex faces energy cost risk

Massive AI spending plans by Microsoft, Amazon, Alphabet, and Meta could face pressure as the Iran war adds to uncertainty around global growth and energy costs, S&P Global Visible Alpha research head Melissa Otto said.

S&P Global estimated the four firms planned about $635 billion of AI-related capital spending in 2026 on data centers, chips, and other infrastructure, up from $383 billion the prior year.

Persistently high oil prices could force capex revisions and trigger a broader equity market correction.

She added that data centers depend heavily on electricity prices and grid capacity, and higher energy costs could hit consumers and corporate earnings.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

AI spending is running into power limits

Combined spending tied to AI could top $630 billion in 2026, near the annual gross domestic product (GDP) of Sweden 1.
Money is moving from a GPU (graphics processing unit, chips used to train and run AI models) shortage toward limits in data center buildouts and available electricity 1.
Locking in power capacity for a new data center often takes 12 to 36 months 2.
Large tech firms are lining up their own supply, since Meta has signed deals for more than 6 gigawatts of nuclear power for AI data centers 3.

AI power demand is changing markets and policy

The infrastructure push is shifting investor interest toward chip designers like Nvidia, while it has also fed worries about software pricing after a nearly $1 trillion selloff in software and services stocks 1.
Higher data center load can strain local grids, which can push up electricity bills in some areas 4.
In the PJM grid (a regional U.S. power operator serving about 65 million people), capacity prices reached $333.44 per megawatt-day, about 12 times higher than two years earlier 4.
Oregon passed the POWER Act (HB 3546) that sets a separate customer class for users above 20 megawatts (MW) plus long-term supply contracts of 10 years or more 5.
Some tech companies now treat electricity like a scarce input, using tools to handle price swings instead of leaning only on fixed long-term power purchase agreements (PPAs) 6.