Meta has claimed billions of dollars in federal research tax credits by treating certain chips and equipment used in its artificial intelligence data centers as experimental supplies, an approach that tax specialists say could eventually draw scrutiny from the Internal Revenue Service.

By yourNEWS Media Newsroom

Meta has sharply increased its use of a federal research tax credit by classifying portions of its massive artificial intelligence data center buildout as experimental activity, cutting billions of dollars from its tax bills as the company pours unprecedented sums into AI infrastructure.

The strategy involves treating certain computer chips and other equipment purchased for AI data centers as supplies used in experimental “pilot models,” according to The New York Times, which reviewed Meta securities filings and interviewed people familiar with the company’s tax practices.

Meta began claiming the research credit for portions of its AI data center spending in 2024, according to the report.

The financial impact was substantial.

Meta reported approximately $700 million in research tax credits in 2023. That figure increased to about $2 billion in 2024 and $3.9 billion in 2025, making the Facebook and Instagram parent the largest publicly traded corporate beneficiary of the credit, according to the Times’ review of company filings.

The increase has come as Meta dramatically expands the computing infrastructure needed to train and operate increasingly powerful artificial intelligence models.

CEO Mark Zuckerberg has repeatedly characterized AI as central to Meta’s future.

“Our investments in AI are accelerating every major part of our core business,” Zuckerberg told investors. “Every sign that we’re seeing in our own work and across the industry gives us confidence in this investment.”

The tax treatment, however, raises a potentially important distinction between equipment being used in genuine experimentation and commercially available hardware being installed as part of ordinary business expansion.

Congress created the federal research tax credit in 1981 to encourage companies to undertake qualified technological research.

The credit generally applies when businesses incur certain expenses while attempting to resolve technological uncertainty through a process of experimentation.

Meta’s approach treats some of the infrastructure inside its AI data centers as part of that experimental process.

According to the Times, the company has classified some chips and other equipment as supplies incorporated into “pilot models,” allowing those costs to contribute to the research tax credit.

That interpretation has generated questions even within Meta.

Some employees in the company’s finance organization questioned the approach, according to people familiar with internal discussions cited by the Times.

The concern centers partly on the fact that many of the chips Meta purchases for its AI data centers are commercially available products rather than experimental hardware developed by Meta itself.

The tax question is not simply whether the equipment is new to Meta or used for AI development. To qualify for the research credit, the underlying activity must meet federal requirements governing technological uncertainty and experimentation.

Andre Shevchuck, a partner at advisory firm BPM who specializes in the research tax credit, described application of the credit to AI data centers as “kind of wild and out there.”

Other tax specialists interviewed by the Times expressed skepticism that Meta could successfully treat all of the relevant chips and data center equipment as qualified research expenses.

The dispute could ultimately depend on what Meta can demonstrate about how the equipment was used.

A company generally cannot transform ordinary commercial equipment into qualified research property merely by labeling a broader project experimental. The relevant inquiry includes whether the company was attempting to resolve technological uncertainty and whether the claimed expenses were sufficiently connected to a qualifying process of experimentation.

That distinction could become important if the IRS audits Meta’s claims.

The company itself acknowledges significant uncertainty surrounding portions of its research tax benefits.

Meta’s securities filings show that the amount associated with uncertain tax positions has risen sharply as its research credits have expanded.

By June 2026, the company had set aside $18.74 billion in gross unrecognized tax benefits, a 45% increase over two years.

Meta has identified “uncertainties with our research tax credits” as one of the principal reasons for those reserves, along with other tax issues.

An uncertain tax position does not mean the company believes its claim is improper.

Accounting rules require corporations to recognize that some tax benefits could be challenged, reduced or disallowed by tax authorities, even when management believes the underlying position can ultimately be defended.

Meta spokesman Andy Stone said the company is making use of incentives Congress deliberately created to encourage American research and investment.

“Over the last five years, Meta invested $200 billion in R&D, $57 billion in the last year alone, advancing frontier research, building new technology and supporting American jobs,” Stone said.

Stone also pushed back against interpreting the company’s large reserve for uncertain tax benefits as evidence that Meta expects the IRS to reject its position.

He called the disclosure “a mandated accounting measure of uncertainty” that encompasses “many different types of uncertainties.”

Meta declined to provide the Times with a detailed explanation of what technological uncertainties its AI data centers are intended to resolve or precisely why commercially available chips and related equipment qualify as experimental supplies.

Those details could become critical if federal tax authorities challenge the credits.

A previous court case involving a shipbuilder illustrates the issue.

In 2021, a federal tax judge considered a company’s attempt to claim research credits for expenses associated with designing and constructing new types of vessels.

The company argued that the vessels could be considered experimental pilot models.

The court rejected the claimed credit after determining that the taxpayer had not adequately demonstrated that the relevant work constituted a qualifying process of experimentation.

The decision did not establish that a commercially useful product can never qualify as a pilot model. Instead, it emphasized that businesses must substantiate specific technical uncertainties and demonstrate that qualifying experimentation was actually undertaken to resolve them.

That precedent could be relevant if the IRS examines whether Meta’s AI infrastructure meets similar requirements.

AI data centers are unusually complex facilities.

Meta and its competitors are assembling enormous clusters containing advanced graphics processing units, networking equipment, cooling systems, power infrastructure and proprietary software designed to train and operate artificial intelligence models at unprecedented scale.

The companies are experimenting with methods for connecting increasingly large numbers of chips, managing heat, reducing power consumption and improving how computing capacity is allocated.

At the same time, much of the underlying equipment is commercially manufactured and used in facilities that become permanent parts of the companies’ business operations.

That combination creates the tax question at the heart of Meta’s strategy: where qualifying research ends and routine capital investment begins.

The amount at stake is growing rapidly.

Meta’s $3.9 billion research tax credit in 2025 was more than five times the $700 million it reported just two years earlier.

The increase has coincided with one of the largest capital spending programs in corporate America as Meta races Microsoft, Alphabet, Amazon and other technology companies to build the infrastructure required for generative AI.

Zuckerberg has indicated that the company intends to continue spending aggressively.

Meta argues those expenditures are developing frontier technologies rather than merely expanding conventional computing capacity.

Tax specialists questioning the company’s treatment are not disputing that Meta conducts extensive research and development.

Instead, the potential issue is whether particular equipment purchases qualify for the specific federal credit being claimed.

The distinction matters because the tax code treats ordinary business investment differently from qualified research expenses.

Companies already receive other tax benefits for capital expenditures, including depreciation and deductions available under separate provisions of federal law.

The research credit can provide an additional benefit when expenses satisfy its more specific requirements.

Meta has not disclosed how much of its total AI data center spending it treats as qualified research expenses or what proportion of the $3.9 billion 2025 credit came specifically from data center equipment rather than other research conducted throughout the company.

Its overall research program spans artificial intelligence, virtual and augmented reality, software, hardware and other technologies.

That makes it impossible from the company’s public filings alone to determine how much tax exposure is tied specifically to the data center strategy.

The disclosures nevertheless show that research credits have become increasingly important to Meta’s tax position.

They also show the company recognizes that tax authorities could ultimately reach different conclusions about portions of its claims.

If the IRS disallows credits after an examination, Meta could face additional tax liabilities along with potential interest.

For now, no public IRS determination has found Meta’s data center claims improper.

The company continues to use the research credit while simultaneously reserving billions of dollars against uncertainty surrounding its tax positions.

As Meta accelerates its AI infrastructure buildout, the strategy could become an important test of how a four-decade-old incentive for experimental research applies to a new generation of extraordinarily expensive computing facilities.

The ultimate question is not whether Meta is conducting AI research — something the company plainly spends billions of dollars doing — but whether the chips and other equipment installed in its AI data centers satisfy the narrower requirements Congress established for expenses receiving the federal research tax credit.

Original article