Everyone is focused on Meta's Capex going up and treating it like a problem but it's not.
Milk Road AI Twitter · Milk Road AI (@MilkRoadAI) · 2026-06-30
Milk Road AI argues that Meta's $125-145 billion 2026 capital expenditure is building a 21.2 GW AI compute empire that could generate $20 billion in annual revenue by reselling just 1 GW of excess capacity, a business the market is not yet pricing in.
Appears in
Extraction
Topics: metaai-infrastructurecapextech-stockscompute-capacity
Claims
- Meta is building 21.2 gigawatts of total compute capacity by 2028, with 13.2 GW classified as AI-focused capacity that exceeds its own operational needs.
- Reselling just 1 GW of that excess compute to external customers could generate $20 billion in annual revenue at an 85% operating margin.
- Meta's AI-driven ad ranking already produces 4x the revenue impact compared to simply increasing ad load, with measurable click and conversion lifts on Facebook and Instagram.
- Meta's incremental ROIC from recent AI investments runs above 20%, with cash-based ROIC above 52%.
- Meta trades at just 15x forward earnings despite 33% annual revenue growth and $26.8 billion in quarterly net income, suggesting significant market undervaluation.
Key quotes
Meta isn't just an advertising company spending recklessly on infrastructure but rather becoming one of the most powerful compute landlords on the planet, and the market is not pricing that in at all.
Analysts are pointing out that Meta could resell just 1 GW of that excess compute to outside customers, and that single gigawatt would generate $20 billion in revenue at an 85% operating margin, producing $17 billion in operating income.
The stock trades at 15x forward earnings. That is a company growing revenue 33% annually, with 3.3 billion daily active users, printing $26 billion in quarterly profit, building a compute empire that could become a standalone business priced like a mature, slow-growth utility.