Meta uses experimental data center classification to reduce federal taxes

Meta is classifying its multi-billion-dollar AI data centers as experimental “pilot models” to significantly reduce its federal tax liability. By labeling these massive infrastructure projects as research and experimental expenditures under IRC Section 174A—a provision reinstated by the 2025 One Big Beautiful Bill Act—the company can claim immediate tax write-offs rather than amortizing costs over several years.

This strategy contributed to a reported drop in Meta’s federal tax expense from $9.6 billion in 2024 to $2.8 billion in 2025, despite the company investing approximately $72 billion in capital expenditures. The company is also seeking tax credits for expensive AI computer chips, such as those produced by Nvidia, by framing them as part of an experimental effort.

One major project, the Hyperion data center campus in Louisiana, involves an investment exceeding $50 billion. However, the Union of Concerned Scientists has raised concerns that if these facilities are truly experimental and risk failure, local utility customers might be forced to cover the costs of stranded infrastructure. Additionally, Meta faces ongoing disputes with the IRS regarding other tax claims, including a previous attempt to classify executive stock options as research expenses.