The AI Infrastructure Thesis Updated January 2026

Follow the Watts

The AI Infrastructure Thesis

The Thesis in 60 Seconds

AI infrastructure requires power. Power takes years to secure. Four companies positioned early and now hold $57 billion in contracted infrastructure revenue from hyperscalers. Two adjacent players selling GPU compute add another $30 billion. The market values all six at $70-75 billion combined while speculative AI plays with zero revenue trade at similar or higher multiples. Either the contracts are worthless, or the market is still catching up to who actually benefits from AI infrastructure buildout. This thesis argues it's the latter, distinguishes infrastructure from compute, acknowledges what could prove it wrong, and provides the milestones to track.

01

Everyone's watching the AI race through the wrong lens.

The focus is on models, chips, talent. The actual constraint is power. Grid capacity took decades to build and will take decades to expand. The companies that secured allocations early own something that can't be replicated on relevant timelines.

02

The obligations have already cascaded.

Enterprises are embedding AI into production systems. Hyperscalers have sold capacity forward to those enterprises. Data center operators control the power those hyperscalers need. Each layer locked in before the next. The flywheel is already running.

03

Markets are mispricing the result.

Speculation with no revenue trades at premiums. Infrastructure generating real cash gets discounted for "execution risk." The asymmetry is structural. It's also temporary.

Read the Thesis →
The Operators
APLD

The North Dakota power moat no one can replicate.

 GLXY

Infrastructure value hidden behind a crypto label.

 CIFR

Multi-hyperscaler optionality before the market notices.

 WULF

Fastest transformation, Google as 14% equity holder.

 IREN

Same power scarcity. Different monetization layer.

 NBIS

Two hyperscalers. Revenue already scaling.
Contents 5 Parts · 15 Sections · ~70 min read

About This Thesis

What this is

An investment thesis arguing that four specific companies are undervalued based on their contracted revenue and power positions: APLD, GLXY, CIFR, and WULF. IREN and NBIS are included as adjacent positions with different risk profiles (GPU cloud rather than infrastructure). I have positions in all six names.

What this isn't

Financial advice. I'm not a financial advisor. I don't know your situation, risk tolerance, or investment goals. The thesis could be wrong. I could be missing something important. Execution risk is real. Counterparty risk is real. I've tried to present the bear cases honestly, but I'm also biased toward my own positions.

How to use it

As one input among many. Test the logic against your own understanding. Verify the sources. Every figure cites SEC filings, earnings transcripts, or official press releases. Read the bear cases as if you're trying to talk yourself out of the trade. Consider what I might be wrong about. Form your own view. Size positions for the possibility that I'm completely wrong.

Updates

Material changes to contracts, execution milestones, or thesis-relevant developments will be reflected in updates. The current version reflects information available through January 2026.

All figures sourced to SEC filings, earnings transcripts, and official releases.

The constraint is physical. The mispricing is temporary. The thesis is testable.

Begin with Section 1 →