Hassan, Colleagues Question Meta’s Use of Billions in Research and Development Tax Credits to Subsidize Its AI Data Center Construction Boom

WASHINGTON - U.S. Senator Maggie Hassan joined colleagues from the Senate Finance Committee in calling on Meta to justify its use of billions in federal research and development tax credits to fund its rapid buildout of AI data centers, in a letter to Meta CEO Mark Zuckerberg today.

According to recent reporting, Meta claimed its AI data centers are experimental "pilot models", and therefore eligible for federal research and development tax credits. That is a potential misuse of that credit, which is intended to incentivize experimental research, not to scale up established computing infrastructure. As Meta spends tens of billions of dollars on AI data centers, its research and development tax credits have grown from $700 million in 2023, to $4 billion in 2025.

"The federal research credit can be a useful and necessary tool to support our nation's innovation economy and cutting-edge research," Senator Hassan and her colleagues wrote in their request to Meta. "Subsidizing your A.I. data center build is not close to meeting this standard."

If the reporting is accurate, Meta's federal research and development tax credit claims would represent more than 10 percent of all research tax credits claimed in the U.S., based on Joint Committee on Taxation figures.

"Any effort by Meta to use A.I. data center expansion as eligible for research credits is potentially a significant misapplication of federal tax law that merits further investigation," the senators continued.

Read the full text of Senator Hassan and her colleagues' request to Meta here or below.

We write regarding reports that Meta Platforms, Inc. ("Meta" or "the company") may be abusing the federal research and development tax credit ("research credit") to avoid billions of dollars in taxes. According to public reports, Meta has claimed to the Internal Revenue Service ("IRS") on federal tax filings that the company's artificial intelligence data centers are experimental "pilot models" that qualify for the research credit. This strategy - in violation of the spirit and, potentially, the letter of the tax law - supposedly saved Meta billions of dollars on its A.I. data center buildout, including on the purchase of advanced microchips and computing equipment.

As you are aware, Meta has indicated on federal regulatory filings that it has used the federal research credit to reduce its federal income tax liability by $6.6 billion between 2023 - 2025. Meta claimed $700 million in research credits in 2023, more than doubled it in 2024 to $2 billion, and doubled it again in 2025 to nearly $4 billion. Meta is a prolific user of the federal research credit and represents a disproportionately large share of the federal tax revenue loss that is attributable to the credit. According to the estimates of the Joint Committee on Taxation, the federal research credit cost the government $32 billion in 2025. By this measure, Meta alone would be responsible for more than 10% of all research credits claimed in the U.S.

While it remains unclear exactly how Meta was able to reduce its federal tax liability by over $6.6 billion in the last three years using federal research credits, it appears from published reports that a substantial share is by claiming the research credit related to its massive A.I. data center build out. Under section 41 of the Internal Revenue Code, companies may claim the research credit for costs related to domestic research or experimental expenditures.5 The research credit is intended to help offset the cost of experimental efforts with a high risk of technical uncertainty; it is not intended for standard business operations or commercially available technologies already in use. For these reasons, acquisition of hardware used in data centers, which is widespread and relies on proven technology, would not normally be considered eligible for the federal research credit.

It appears that Meta was aware of the risks associated with claiming billions in tax breaks on A.I. data centers but decided to move forward with this questionable strategy anyway. According to public reports, Meta is treating its A.I. data centers and standard data centers differently for tax purposes, claiming that its A.I. data centers are part of a giant research experiment. Meta apparently employed this questionable theory in order to use research credits to cut billions of dollars of the costs incurred from purchasing costly hardware needed to operate its data centers, such as advanced microchips acquired from companies like Nvidia.

Meta in its disclosures to investors indicated that this tax strategy may not withstand scrutiny from the IRS. In its most recent annual report, Meta indicated that its "uncertain tax positions" were related to "uncertainties with our research and development tax credits that are based on an assessment of whether our available documentation corroborating the nature of our activities supporting the tax credits will be sufficient." Any effort by Meta to use A.I. data center expansion as eligible for research credits is potentially a significant misapplication of federal tax law that merits further investigation.

This is unfortunately not the first time Meta has claimed federal research credits for questionable purposes. In 2013, Meta claimed that $4.1 billion you received when you exercised stock options should be eligible for research credits. The IRS is challenging this position that billions of dollars in CEO compensation you received is eligible for the research credit and seeking $355 million in back taxes from Meta. The federal research credit can be a useful and necessary tool to support our nation's innovation economy and cutting-edge research. Subsidizing your A.I. data center build is not close to meeting this standard. In order to ensure the research credit is administered responsibly to protect the American taxpayer, and to better understand Meta's use of federal research credits to subsidize its A.I. data center build out, please provide answers to the following questions no later than October 31, 2026: