Meta has found a way to turn one of Silicon Valley’s biggest spending sprees into one of the biggest tax breaks in corporate America, and the mechanism behind it is raising eyebrows in Washington. According to a New York Times investigation published September 30, 2026, Meta is using a decades-old federal research tax credit to save billions of dollars by labeling its sprawling AI data centers as “pilot models” and the Nvidia chips inside them as experimental materials. The classification lets a company spending hundreds of billions on compute infrastructure claim the same tax treatment normally reserved for lab experiments and prototypes.
Key takeaways
- Meta saved $3.9 billion in federal research tax credits in 2025, up from $2 billion in 2024 and $700 million in 2023, according to the New York Times.
- The company classifies massive AI data centers, including a planned 2 GW+ facility in Manhattan, as “pilot models” to qualify for the credit.
- Meta is now the biggest beneficiary of this research tax credit among all publicly traded companies.
- A 45 percent increase, bringing the company’s reserves for uncertain tax positions up to $18.74 billion, suggests that its own accountants perceive legal risk in the strategy.
- Auditor Ernst & Young approved the approach and is now pitching it to other companies buying AI chips.
- Senator Elizabeth Warren has sent questions to Meta, Google, Amazon and Microsoft about AI-related tax subsidies, CNBC reported on September 28, 2026.
Meta’s Tax Strategy on AI Data Centers and Chips
Meta is leaning on a federal research tax credit to offset the enormous cost of its AI buildout, and the trick lies in how the company describes its own infrastructure to the IRS. Facilities that Mark Zuckerberg publicly calls the backbone of Meta’s future are, on paper, treated as experiments still being tested.
Classification of AI Infrastructure as ‘Pilot Models’
In January 2025, Zuckerberg announced plans for a data center exceeding 2 gigawatts of power capacity, large enough to cover a significant part of Manhattan. For tax purposes, Meta classifies this and other similarly massive facilities as pilot models, the New York Times reported. That label sits uneasily next to what Zuckerberg told the public at the time: that these data centers would “drive our core products and business.” Calling a facility built to run Meta’s products an unproven pilot is, at the very least, a stretch that the company has had to defend.
Use of Nvidia Chips as Experimental Materials
Alongside the data center classification, Meta treats Nvidia chips deployed inside these clusters as experimental materials rather than standard production hardware. That framing, paired with the pilot-model label for the buildings themselves, forms the backbone of the company’s research tax credit claims. According to the Times, by June 2026 Meta had begun openly sharing details of its compute infrastructure with investors, covering partnerships with Nvidia, AMD, AWS, Arm and Broadcom as well as its own custom MTIA chips. None of that disclosure reads like a description of an experiment still in testing.
Financial Impact of the Research Tax Credit
The numbers behind Meta’s use of the credit have climbed sharply in just three years, turning a routine tax provision into a multibillion-dollar line item. Meta saved $3.9 billion in 2025 through the credit, up from $2 billion in 2024 and $700 million in 2023, the New York Times reported, citing Meta’s own disclosures. That trajectory makes Meta the single biggest beneficiary of this research tax credit among all publicly traded companies.
Meta defends the practice by pointing to roughly $200 billion spent on research and development over the past five years, arguing that scale alone justifies the credit. Why this matters: a tax break built for scrappy 1980s-era innovation is now subsidizing some of the largest capital expenditure projects in corporate history, and Meta’s growing reliance on it shows how far a single company can stretch a tool originally aimed at much smaller-scale experimentation.
Meta’s AI Infrastructure and Strategic Partnerships
Meta’s compute ambitions go far beyond a single building. In July 2025, Zuckerberg said Meta would “invest hundreds of billions of dollars into compute to build superintelligence,” anchoring that pledge with several multi-gigawatt clusters. The first, named Prometheus, is already partly online. A second, called Hyperion, is designed to scale up to 5 gigawatts over several years.
“We have the capital from our business to do this,” Zuckerberg wrote at the time. The company’s hardware partnerships with Nvidia, AMD, AWS, Arm and Broadcom, combined with its in-house MTIA chips, point to infrastructure built for long-term production use, not short-term testing. That gap between the language used with investors and the language used with tax authorities is at the center of the scrutiny Meta now faces.
Regulatory and Legal Risks Surrounding Meta’s Tax Credit Use
Meta’s own financial filings suggest the company knows this strategy could be challenged. The scale of its reserves tells its own story about how confident Meta’s accountants really are.
Accountants’ Warnings and Tax Reserves
In SEC filings, Meta warns investors that the tax savings from this strategy could be challenged by regulators. Uncertain tax position reserves climbed by 45 percent to reach $18.74 billion, a figure reflecting genuine internal concern about how well the company’s claims will hold up. Even if the IRS eventually claws back some of that money, Meta likely still comes out ahead, since the capital was already put to work funding data center construction and, in the process, supporting its stock price.
Role of Ernst & Young and Potential Conflict of Interest
Meta’s auditor, Ernst & Young, approved the tax strategy and reportedly helped design it in the first place. According to the New York Times, EY is now marketing this same strategy to other companies looking to offset their own AI chip purchases. That dual role, as both Meta’s auditor and the architect of a tax scheme it now sells elsewhere, adds another layer of scrutiny to a story already drawing attention from lawmakers. Separately, Senator Elizabeth Warren has sent questions to Meta, Google, Amazon and Microsoft about AI-related tax subsidies, CNBC reported on September 28, 2026, signaling that Capitol Hill is watching how the biggest tech companies are using public tax tools to fund private AI buildouts.
Origin and Intent of the Federal Research Tax Credit
The credit Meta is leaning on traces back to a law passed in 1981, decades before anyone imagined gigawatt-scale AI data centers. James Shannon, the congressman who introduced the legislation, told the New York Times it was meant to support “people power, knowledge, information.” Looking at how Meta now applies it, Shannon said the company’s use has “gone way, way beyond what anybody could have imagined.”
That gap between original intent and current application is the core tension in this story. A credit designed decades ago to reward small-scale innovation is now underwriting some of the largest infrastructure projects ever built by a private company, and Meta’s growing tax savings are a direct measure of how wide that gap has become.
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.