Everyone’s talking about who builds the best AI model. But the real bottleneck in artificial intelligence right now isn’t software — it’s concrete, copper wire, and electricity.

KKR just put $10 billion behind that thesis.

Helix Digital Infrastructure: The Pitch

The private equity giant launched Helix Digital Infrastructure this week — a standalone company that will design, build, own, and operate the physical backbone AI depends on. Data centers. Power generation. Transmission lines. Cooling systems. The unsexy stuff that makes everything else possible.

To run it, they tapped Adam Selipsky, the former AWS CEO who oversaw the cloud division’s growth past $100 billion in annual revenue. The man who spent years watching hyperscalers scramble for compute capacity is now positioning himself as their infrastructure partner instead of their rival.

The pitch is elegant: instead of owning every data center on your balance sheet, offload the buildout to Helix. Lock in capacity. Reduce risk. Move faster.

The $700 Billion Infrastructure Crunch

The timing isn’t accidental. Big Tech’s latest earnings calls revealed staggering numbers: Alphabet, Amazon, Meta, and Microsoft collectively plan to spend roughly $700 billion on AI infrastructure in 2026. Meta alone bumped its forecast to $125–145 billion, up from $115 billion.

And it’s still not enough.

Models keep getting larger. Inference workloads — the computing power needed to actually run these models for millions of users — are growing faster than training costs. The gap between what the industry needs and what physically exists widens every quarter.

The bottleneck isn’t GPUs anymore. It’s everything around the GPUs: the buildings, the electricity, the cooling, the network pipes. New data center projects across the U.S. are stalled for years waiting on power grid connections and permits.

AI Infrastructure as Utility

KKR’s move signals a fundamental shift in how Wall Street thinks about AI investment.

Wave one chased model builders — OpenAI, Anthropic, Mistral. Wave two targeted chipmakers like Nvidia. Wave three is here: AI infrastructure as a utility play.

Data centers and power generation produce steady, predictable returns over decades. They look like pipelines and electric grids, not software companies. For institutional investors — pension funds, sovereign wealth funds — that’s the sweet spot. AI exposure without the volatility of betting on which model wins.

KKR already owns CyrusOne, one of the world’s largest data center operators, and recently scored a massive exit from CoolIT Systems (acquired by Ecolab for $4.75 billion). Helix brings it all under one AI-focused roof.

The $10 billion is just the opener. A sovereign wealth fund has already expressed interest.

The Power Problem Getting Worse

Here’s the part that should unsettle anyone watching AI’s trajectory: the electricity crisis is accelerating.

Bloomberg NEF data shows the cost of building new combined-cycle gas turbine power plants has surged 66 percent since 2023. Construction timelines are 23 percent longer. The supply chain for power generation equipment is being crushed by data center demand.

The U.S. needs tens of gigawatts of new power capacity for projected AI workloads through decade’s end. One gigawatt powers roughly 750,000 homes. We’re talking about powering several major cities — just for AI.

This is why Helix isn’t just building data centers. It’s building power generation and transmission too. That vertical integration is the differentiator. Most operators lease power from utilities. Helix wants to generate its own, potentially bypassing the grid bottlenecks strangling projects nationwide.

A Crowded Gold Rush

Helix isn’t alone at the claim. SoftBank announced Roze AI the same week — a robotics company designed to build data centers autonomously, targeting a $100 billion IPO valuation. Blackstone and Brookfield are raising multi-billion-dollar infrastructure funds. CoreWeave expanded from GPU cloud into physical infrastructure. Even traditional utilities like NextEra Energy are pitching dedicated AI power packages.

The competition validates the thesis. But it raises the overbuilding question: if everyone pours billions into construction simultaneously, do we end up with stranded assets?

Most analysts say no — not before 2030. The demand projections are enormous and the supply gap significant enough that aggressive buildout won’t overshoot near-term. But these investments extend 20–30 years, and predicting AI demand on that timeline is anyone’s guess.

The Bottom Line

KKR’s Helix launch is the clearest signal yet that AI is transitioning from a software story to an infrastructure story. The models matter, but they’re useless without the physical systems to run them. And those systems — data centers, power plants, cooling infrastructure, fiber networks — are becoming the binding constraint on progress.

With $700 billion in hyperscaler spending planned for this year and total AI infrastructure investment projected to exceed $1 trillion by decade’s end, the opportunity is massive. The risk is execution: can anyone actually build fast enough?

Adam Selipsky and KKR are betting $10 billion they can. In an industry full of wild bets, this one might be among the most rational.