AI's Power Crunch
Tech giants confront AI's soaring power demand, planning massive investments as new plants lag
Anonymous
The world's five biggest AI companies plan to spend between $660 billion and $690 billion on infrastructure in 2026 alone — nearly double what they spent in 2025, according to Calgary Herald. But money cannot solve the real problem: there is not enough electricity. A single AI query uses up to 1,000 times more power than a basic web search, and new power plants take 5 to 10 years to build.
In March 2026, CEOs from Microsoft, Google, Amazon, Meta, OpenAI, Oracle, and xAI met at the White House and signed a pledge to pay for "every megawatt" of new electricity their AI projects need, Calgary Herald reported. The pledge is bold. The grid, however, is not listening.
Why the Electric Grid Cannot Keep Up With AI
AI data centers can be built in 12 to 18 months. Utility-scale power plants take 5 to 10 years, according to Clinton News Record. That gap is now the single biggest threat to AI growth. On top of that, high-voltage transformers — the machines that move electricity from plants to data centers — now take up to 5 years to order and receive, up from just 2 years before 2020.
The U.S. grid interconnection queue — the waiting list for new power projects to plug in — now holds 2,300 gigawatts of backlogged projects. That is more than the entire current power capacity of the United States. The average wait time has doubled over 15 years to about 5 years. Gartner predicts that power shortages will block 40% of all planned AI data centers by 2027.
Bitzero Secures Over a Gigawatt Before the Crunch Hit
While most companies are scrambling for power now, Bitzero (Nasdaq: AIBZ) locked in its electricity years ago. The company controls over 1 gigawatt of low-cost power across Norway, Finland, and North Dakota, according to Seaforth Huron Expositor. Its flagship site in Namsskogan, Norway runs entirely on renewable hydroelectric power. Bitzero began trading on Nasdaq in mid-June 2026.
In May 2026, Bitzero signed a binding 15-year lease with OneQode Networks worth $2.6 billion for its 110 MW Norway site, Calgary Herald reported. The site carries an estimated 85% net operating income margin — a figure that reflects just how valuable pre-secured power has become. CEO Mohammed Bakhashwain built the company by moving away from Bitcoin mining and into AI and high-performance computing.
Nordic Nations Become the New Silicon Valley
Norway and Finland are emerging as global hubs for AI infrastructure. Both countries offer stable, cheap renewable energy and cold climates that cut cooling costs. Bitzero's Finland site in Kokemäki is planned for between 600 MW and 1,000 MW of capacity. The first 80 MW phase is targeted for 2027, according to County Market.
The contrast with the United States is sharp. xAI, Elon Musk's AI company, had to use road-portable gas-fired generators in Memphis just to avoid the 5-year grid wait. Those generators cost far more than grid power. Meanwhile, 86 U.S. counties have placed moratoriums on new data center construction due to power strain, according to Seaforth Huron Expositor.
'Megawatts Over Chips' — How Investors Are Rethinking AI
The phrase "speed-to-power" has replaced "speed-to-market" as the key phrase in AI investing. Morgan Stanley analyst Stephen Byrd says the coming capex cycle will "create islands of wealth, and literal power" for those who already locked in grid-tied assets. Unpowered GPUs sitting in unlit data centers are now called "compounding waste" by industry insiders.
By 2060, AI is projected to consume 11% of all global electricity — about 6,400 terawatt-hours per year — according to energy research firm DNV. Bloom Energy's Natalie Sunderland says "access to power remains the biggest constraint to data center growth." Companies are now building their own onsite power using natural gas or small modular nuclear reactors just to bypass the grid entirely, Clinton News Record reported.
Sentiment
NeutralNEGATIVENEUTRALPOSITIVE
Publishers
4
Articles
4
Reach
4
Topics
A
AnonymousAuthor
