- The wave of massive AI infrastructure construction is creating a major transparency issue in accounting, as chip and data center costs are often lumped together.
- Tech companies announce total long-term AI project costs but rarely decouple infrastructure construction costs from chip expenses.
- This is a serious issue due to vastly different depreciation lifecycles: data centers can last 20–40 years, while AI chips may become obsolete in less than 3 years.
- Short-term chip costs are thus “buried” with long-term assets, making it difficult for investors to assess real risks.
- The “Construction-in-Progress” (CIP) line item is seen as a “gaping loophole” for hyperscalers to hide costs, as it is not yet subject to depreciation.
- Tech giants are spending hundreds of billions of dollars on data centers, chips, and networking.
- Spending on data centers is expected to exceed office construction costs as early as next year.
- AI projects risk delays due to power constraints and local politics, increasing financial risks.
- Alphabet recorded $50.6 billion in assets not yet in service in 2024, a 44% increase year-over-year.
- Amazon recorded $46.4 billion (up 62%); Meta reached $26.8 billion (up 10%).
- These amounts represent 30%, 18%, and 22% of the net total PPE (Property, Plant, and Equipment) of the three companies, respectively.
- Total Construction-in-Progress for public companies with market caps over $2 billion related to AI reached approximately $214.5 billion.
- In 2024, public companies changed asset lifecycle estimates 79 times, mainly extending the lifespan of servers and network equipment.
- This helps reduce depreciation and increases reported profits by hundreds of millions of dollars.
- Investors complain about the lack of information regarding chip lifecycle assumptions and the level of obsolescence risk.
- FASB currently has no plans to require the decoupling of Construction-in-Progress costs.
📌 The wave of massive AI infrastructure construction is creating a major transparency issue in accounting, as chip and data center costs are often lumped together. Over $214.5 billion in construction-in-progress is combined between short-term chips and long-term assets. When AI chips may become obsolete in under 3 years but are hidden within 20–40 year projects, investors struggle to assess true risks. This lack of transparency could obscure capital waste and distort profits in the global AI race.
