Meta has avoided billions of dollars in federal taxes by classifying its artificial intelligence data centers as pilot models and its Nvidia chips as experimental materials. According to reporting by The Decoder, this tax strategy saved the company $3.9 billion in 2025 alone. The maneuver relies on a tax credit that dates back to 1981. Even Meta’s own accountants have reportedly flagged the approach as legally risky. The classification directly contrasts with public statements made by CEO Mark Zuckerberg in January 2025, where he stated that these same data centers would drive the company’s core products and business.
The 1981 Tax Credit Context
The provision being utilized traces back to 1981, originally designed to incentivize domestic research and development in traditional manufacturing and technology sectors. Applying decades-old research and experimentation credits to modern, multi-billion-dollar hyperscale artificial intelligence infrastructure represents an aggressive interpretation of the law. As capital expenditures for AI infrastructure soar across the technology sector, companies are increasingly testing the boundaries of existing tax frameworks meant for physical goods and short-term laboratory experiments rather than permanent production hardware.
What Remains Unclear
The exact legal vulnerability and potential liabilities Meta faces from federal tax authorities remain unspecified in current reports. It is unknown whether the Internal Revenue Service plans to audit these specific filings or challenge the experimental classification of commercial-scale hardware. Furthermore, the announcements and reports did not detail whether other major technology companies are employing identical tax maneuvers for their own AI buildouts, nor what specific internal thresholds Meta uses to separate experimental infrastructure from production assets.
As federal scrutiny on corporate tax strategies intensifies, the outcome of Meta’s approach could set a major precedent for how the technology industry accounts for artificial intelligence expenditures.
Lead image: panumas nikhomkhai / Pexels
