AI's Next Bottleneck: 1 CEO Warns of a 6-Year Wait for Power

The artificial intelligence buildout is colliding with the physical limits of the electrical grid, shifting the primary bottleneck from silicon chips to power availability.

The AI arms race has a new chokepoint, and it’s not Nvidia’s GPUs. The primary constraint is now access to power, with the CEO of IREN warning that a new gigawatt-scale AI data center started today may not get compute power online before 2030, fundamentally altering the investment landscape.

"If you wanted to start today and build a gigawatt AI factory, you are looking 2030 before you get the first compute online,” Daniel Roberts, CEO of IREN (NASDAQ:IREN), said in a recent interview with Bloomberg Tech.

The timeline stems from the sheer physics of the electrical grid. Before construction can even begin, utilities may take 18 to 24 months just to assess whether a proposed site has available power capacity. This physical constraint comes as Goldman Sachs projects global data center power demand will surge roughly 160 percent by 2030, driven almost entirely by AI.

This reality shifts the investment calculus for the AI boom. Even with hyperscalers like Alphabet, Microsoft, and Amazon collectively holding over $800 billion in operating cash flow, capital alone cannot accelerate grid upgrades. The new dynamic puts a massive premium on companies that already possess the land, permits, and, most importantly, the secured power agreements.

The 6-Year Timeline to "Get Compute Online”

Roberts’ 2030 warning highlights a problem that has moved faster than most analysts predicted. For the past two years, the AI story was largely about the scarcity of high-end chips, sending Nvidia’s (NASDAQ:NVDA) revenue soaring from $27 billion to a projected $215.9 billion in fiscal 2026. Now, the bottleneck has moved from the digital to the physical world.

A 1-gigawatt AI facility consumes electricity on the scale of a medium-sized city, a demand that aging power grids were not designed to accommodate. Utilities cannot simply flip a switch for that level of consumption, and building new transmission lines is a multi-year process fraught with regulatory hurdles. The 18-to-24-month site assessment is just the beginning of a long and complex journey.

Infrastructure Owners Gain a "Power" Moat

This new environment creates a powerful moat for companies that secured infrastructure years in advance. IREN, once viewed primarily as a Bitcoin mining operator, spent years acquiring land and power agreements in regions with scalable energy access. Those assets, which may have seemed like a liability during crypto downturns, are now proving to be a critical advantage in the AI arms race.

The dynamic also benefits established power producers with scalable, reliable generation. Constellation Energy (NASDAQ:CEG), the largest producer of nuclear energy in the U.S., is uniquely positioned to provide the 24/7 baseload power that AI data centers demand. The company generated $4.2 billion in operating cash flow in 2025 and is signing long-term power agreements with hyperscalers to support their AI ambitions.

In short, the AI trade is expanding beyond semiconductors. As the bottleneck shifts from chips to electricity, investors are now evaluating who owns the land, power agreements, and grid connections. While IREN and CEG have rallied significantly, the multi-year timeline for new projects suggests the underlying investment thesis may have more room to run, though regulatory headwinds and execution risks remain critical factors to watch.

This article is for informational purposes only and does not constitute investment advice.