Putting AI and data on the balance sheet
Firms spend heavily on data, software and AI models, but much of that spending is expensed and barely shows on balance sheets. Intangible assets are a steady finance topic, yet only one of 57 recent papers on them carries an AI, data or software keyword. How large is the missing AI capital, and does it distort book-based measures?
What prompted this
Since October 2025, 57 papers carry an intangible-assets, intangible-capital or intangible-investment keyword (30 in finance journals, 11 in economics, 10 in management). Only one of them also carries an AI, data-asset, software or digital-asset keyword, while AI keeps rising in almost every category on the trends page.
Questions
- How much do listed firms spend on AI and data each year, and how much would a capitalized "AI capital" add to their assets?
- Does adding it change book-to-market sorts, the value premium or measured returns on capital?
- Where the rules allow capitalization, are firms that capitalize more of it valued differently?
Data that could help
- R&D, software and overhead expenses (Compustat), and AI job postings as a proxy for AI labour spending.
- China's accounting rules for enterprise data resources, in force since 2024, which let some firms report data as assets.
Next step
Estimate AI labour spending of listed firms from job postings and salaries, capitalize it with a perpetual-inventory method, and recompute book-to-market.