Meta (META) is reportedly avoided billions in federal taxes, by claiming that its AI data centers are a "giant experiment" which could fail, NYT reports citing sources

The typical sequence has been reputational pressure, congressional or state-level scrutiny, and only later any audit, reassessment, or legislative response, so the gap between headline and fiscal consequence has historically been long. The substance here is a research and experimentation framing of data centre spending, a well-worn channel through which large technology groups have characterised capital and operating costs to access favourable treatment; the line between legitimate R&D classification and abusive positioning is precisely what tax authorities contest, and outcomes in such disputes have turned on documentation rather than on press accounts. The distinction worth drawing is between a retrospective tax liability, which would be a one-off earnings matter, and a forward narrowing of the deduction, which bears on the economics of the AI build-out at the margin. Follow-ons are any company response, comment from tax authorities, and whether peers with comparable data centre programmes draw the same scrutiny. As a single-source report without an official action attached, the signal is directional on regulatory risk rather than on numbers.