100% overnight gains? (From The Oxford Club)

Artificial intelligence (AI) infrastructure stocks just learned how quickly spending fears can outrun demand. CoreWeave (NASDAQ: CRWV) fell about 25% in a matter of days during the recent sell-off, while Applied Digital (NASDAQ: APLD) and IREN (NASDAQ: IREN) slid roughly 50% from their summer highs.

The worry is familiar. Data center builders are pouring enormous capital expenditure (CapEx) into new campuses, much of it borrowed, and investors want to know whether that capacity ever pays for itself.

But the market may be counting the wrong thing. The bigger threat to the buildout isn't too much supply. It's how many announced projects quietly never get finished.

If permits, water and power are the real constraints, the sites that clear those hurdles could become far scarcer and far more valuable than the sell-off implies.

2 Growth Verticals Driving the AI Infrastructure Trade

Kevin O'Leary, the businessman known as Mr. Wonderful from Shark Tank, invests in Bitzero Holdings (NASDAQ: AIBZ) and is developing private data center projects in Utah and Alberta. The theme is his largest, at well over 20% of his book.

He splits the opportunity into two verticals. Power is the first, since the economy needs far more grid capacity regardless of AI. Compute and storage capacity is the second, and it runs well beyond AI.

Much of what Amazon.com (NASDAQ: AMZN), Alphabet (NASDAQ: GOOGL), and Microsoft (NASDAQ: MSFT) run through their data centers has nothing to do with AI, per O'Leary. In his view, compute demand sits at roughly 3x available capacity, with no slowdown in sight for three years.

Permits and Power Separate Real Projects From Press Releases

The price tags explain the market's nerves. Tenants that once asked for 250 to 500 megawatts now want at least 1.4 gigawatts, roughly $16 billion from scratch to operations, split about evenly between equity and debt.

Before a tenant rolls in a single rack, the developer needs land, Environmental Protection Agency (EPA) air permits, water rights, dark fiber, and power. Planning alone can run $20 million to $30 million per gigawatt.

Power is the hardest piece. Former crypto miners pivoted to AI because they already had grid connections, but most states now require new projects to bring their own power, usually natural gas run through turbines.

Time is the hidden cost. Internal rate of return (IRR) depends on speed, and a site facing a three-year permit wait cannot produce a return that anyone will finance. Tenants want capacity online by the fourth quarter of 2027, a demand O'Leary calls insatiable.

Here's where it gets interesting. Once the power contract, water permit, and air approval are all signed, a project's value can jump sharply, per O'Leary.

Scarcity, Not Oversupply, Could Define the Next 24 Months

The popular bear case says too much capacity is coming, and pricing power could fade once it's built. O'Leary rejects it.

By his estimate, about 65% of data center projects announced across North America in the past two years never get built, sunk by missing water, fiber or power, or by a political moratorium. He expects 60% to 65% could be shut down within 24 months.

Politics is the sharpest risk. O'Leary cites polling showing more than 70% of people oppose a data center in their state, though answers shift when the site sits three miles away and offers jobs paying $150,000. With midterm elections approaching, blocking data centers can win votes.

The oversupply debate assumes announced capacity equals built capacity. O'Leary's math assumes most of it never shows up.

He has not heard of a new project announced in roughly seven months, and his Box Elder County, Utah, site could fund about 30% of the county's budget within 24 months, he estimates. A digital asset bill known as the CLARITY Act could add further compute demand if it passes.

Public Liquidity and Private Projects Can Share 1 Portfolio

O'Leary splits his exposure on purpose. Bitzero, with land, power, and permits in Norway and Finland, gives him liquidity. His Utah and Alberta projects are private.

Location drives cost. Saudi Arabia offers power near one cent per kilowatt-hour, though capital is hard to raise there right now. Norway, Finland, and Alberta come in under six cents, backed by hydro, nuclear, and stranded natural gas.

Long term, he treats data centers like commercial real estate with utility-style cash flows. In climate-controlled storage, he targets 17% to 22% returns and prefers no more than 30% leverage. Investors seeking lower volatility could stick with large caps like Microsoft or Anthropic, if it ever goes public, per O'Leary.

The upside is scarcity. If roughly two-thirds of announced projects fail, permitted and powered capacity becomes the prized asset, and operators already online gain leverage with a short list of about 11 major tenants.

The risk is political and financial. A moratorium can stop any project midway through permitting, and heavy debt leaves little room for delay. Spreading sites across countries lowers that risk without erasing it.

Path to Power Matters More Than Headline Spending

CoreWeave, Applied Digital, IREN, and Bitzero share the same fault line. The market is pricing their spending. What may matter more is whether their sites have power, permits, and tenants in place.

In O'Leary's view, Bitzero's public listing allows it to issue debt and equity, and its operations show it can run a facility profitably—an edge that private projects still chasing permits lack.

The scarcity argument doesn't need every builder to win or AI enthusiasm to hold every quarter. It needs compute demand to stay strong and permitting to stay hard. A sudden wave of newly permitted capacity or a hyperscaler pullback in chip spending would weaken it.

Stay focused on who has signed power and cleared permits, because that's what moves data center stocks once the spending headlines fade.

Read this article online ›