The Meta AI data center tax credit is drawing fresh scrutiny after a new New York Times report. According to the Times, Meta has used an aggressive tax strategy for two years. If the report holds up, taxpayers have indirectly helped fund the company’s AI expansion.
What the New York Times Reported
Four people familiar with Meta’s operations spoke to the Times. They said Meta describes its AI data centers to the IRS as “pilot models,” meaning experimental projects. That label matters, because experimental work can qualify for the research and experimentation tax credit.
Publicly, the company tells a different story. Mark Zuckerberg said on the latest earnings call that the company’s AI spending is already “paying off.” The tax filings, however, reportedly portray the same facilities as works in progress.
How the Meta AI Data Center Tax Credit Works
The research credit rewards companies for money spent on experimental efforts. It covers supplies used in that research, but not routine business operations.
In Meta’s case, the “supplies” are reportedly the chips inside its AI data centers. As a result, the Times says Meta has claimed billions of dollars in credits over the past two years. That makes it the largest publicly traded beneficiary of this credit.
Why Meta’s Own Accountants Are Reportedly Nervous
The classification sits in a gray area, according to the report. Meta’s accountants allegedly worry that the IRS could reject it and reverse the savings. The company even lists the uncertainty around its research credits as a risk in securities filings.
The Times also reports that Meta’s auditor, EY, has pitched the same approach to other AI companies. If so, the strategy could spread well beyond one company.
The Nvidia Connection
Those chips come from Nvidia. Earlier this year, Meta signed a massive multi-year chip deal with the chipmaker. Bloomberg estimates show Meta was already Nvidia’s second-largest customer in the last fiscal year. The February agreement adds millions more chips.
Meta also operates 28 data centers across the United States. Each one needs enormous amounts of this hardware.
AI Spending Is Squeezing Meta’s Cash
Even with the alleged credits, the spending has hurt Meta’s finances. Free cash flow last quarter fell to just $784 million. That figure is roughly $8 billion lower than the same period a year earlier.
Meta is not alone. Amazon, Microsoft, and Google also run AI data centers at massive scale. Together, these hyperscalers rank among Nvidia’s biggest customers.
Why Bubble Watchers Are Paying Attention
Some analysts see danger in this setup. If the hyperscalers weaken financially, or if demand fails to match investment, Nvidia could suffer. In that case, the effects could ripple through the whole AI industry.
Investor Michael Burry is one of those critics. This week, he said he expects an AI bubble burst to unfold over the next year. He also predicts Nvidia’s shares will fall by next September. Nvidia, for its part, says it feels very confident about its financial future.
