Meta has reportedly used an aggressive interpretation of a US federal research tax credit to cut billions of dollars from its tax bill by classifying some of its artificial intelligence data centres as experimental facilities, according to an investigation by the New York Times.
The publication reported that Meta has described its AI data centres as "pilot models" for tac purposes, allowing the company to claim the research and experimentation tax credit on certain equipment used at the facilities. The strategy reportedly includes expensive AI chips purchased from companies such as Nvidia.
The newspaper stated that its investigation was based on a review of Meta's securities filings and interviews with four people familiar with the company's operations.
ALSO READ: Meta Ray-Ban Smart Glasses Face Sales Halt In Netherlands Over Privacy Concerns
According to the New York Times, Meta began using this approach around 2024, distinguishing chips intended for AI data centres from those sent to its conventional data centres.
Meta's Tax Savings
The tax strategy has produced substantial savings.
According to Meta's filings cited by the NYT, the company's research tax credit reduced its tax bill by about $700 million in 2023, before the AI data-centre strategy began.
Meta has been exploiting the loophole for two years now. In 2024, the company saved $2 billion. In its 2025 fiscal year filing, it saved almost $4 billion, making it the largest known beneficiary of the credit to date among public companies.
The nearly $6 billion claimed in 2024 and 2025 makes Meta the largest beneficiary of the research tax credit among publicly traded companies according to the New York Times.
The newspaper reported that the tax credit was created in order to encourage companies to invest in research and experimentation rather than routine commercial operations.
Why Is The Strategy Controversial?
The NYT reported that some Meta employees in the company's finance department questioned whether the approach would satisfy IRS requirements.
The central issue is whether buying commercially available computer chips for large-scale AI infrastructure constitutes a research expense.
Tax experts cited by the New York Times, said the distinction is important. Andre Shevchuck, a partner at advisory firm BPM, described characterising Meta's AI data centres as experimental as “kind of wild and out there”. Another tax expert, Shawn Marchant of Tanner, told the newspaper that he would be skeptical of applying the credit to all the computer chips used across Meta's data centres.
The newspaper also reported that the Internal Revenue Service (IRS) has previously challenged attempts by companies to use the research credit for proven and commercially available equipment and technology. Some employees within Meta's finance department also reportedly questioned whether the company's approach would withstand scrutiny from the IRS.
Meta's Own Tax Warning
Meta has acknowledged potential risks associated with its tax position in securities filings. The company has reportedly warned investors about "uncertainties within our research tax credits" with its disclosure indicating that billions of dollars in tax benefits could potentially be challenged by the IRS.
The New York Times reported that Meta's reserve for potential tax disputes increased by about 45 percent, from $12.9 billion to $18.74 billion, during the period in which the AI data centre tax strategy was introduced.
Meta Defends Its Position
Meta has defended its use of the tax credit, stating that its enormous investment in research and development makes it eligible for incentives established by the Congress.
Company spokesperson Andy Stone assured the New York Times that Meta was spending copious amounts of money on research and development to support “American jobs.” He stated that the company had invested around $200 billion in research and development in the US over five years, including about $57 billion in the most recent year.
Meta's Massive AI Buildout
The exploit is part of a much broader and troubling trend. The AI industry has already garnered a reputation for posting inflated “vibe revenue” and “creative accounting,” while operating in a regulatory vacuum, adding to concerns over a possible AI bubble, which experts fear is on the verge of popping.
Meta has expanded its Hyperion data centre project in Louisiana towards approximately five gigawatts of computing capacity, with the project estimated to cost more than $50 billion, according to reports based on the New York Times' investigation.
The controversy therefore centres not on whether Meta is investing in AI research, but on whether its large scale commercial infrastructure can legitimately qualify for a tax incentive designed to encourage experimentation.
In short, Meta is trying to fudge the numbers to save itself from having to pay taxes — which could throw even more fuel on the fiery debate surrounding the AI industry's attempts to justify its enormous data centre buildouts.
ALSO READ: Meta Launches Forum: New App Turns Facebook Groups Into A Reddit-Like Platform
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.