Urgent.News

What's breaking now, across thousands of outlets.

AI

Zuckerberg says Meta’s AI bet is paying off, so why does Meta tell IRS it could fail

Mark Zuckerberg tells investors Meta’s AI push is accelerating every part of the business, but Meta tells the IRS its AI data centers are an experiment that could fail. The New York Times reports Meta classifies its AI data centers as ‘pilot models’ to claim research tax credits on Nvidia chips, saving $3.9 billion in 2025. Its own filings warn the strategy could be overturned, with unrecognized…

Zuckerberg says Meta’s AI bet is paying off, so why does Meta tell IRS it could fail

Mark Zuckerberg declared two years ago that Meta's investments in artificial intelligence were proving successful across its platforms. However, when it comes to the US Internal Revenue Service, a contrasting narrative emerges. According to a report in The New York Times, Meta classifies its AI data centers as a high-risk experiment that may not pan out.

This framing enables the company to claim the Research and Experimentation Tax Credit, designed by Congress in 1981 to incentivize innovative risk-taking. The credit has yielded significant benefits for Meta, reducing its tax bill by $2 billion in 2024 and a staggering $3.9 billion in 2025, compared to $700 million in 2023, the year prior to launching the AI data centers.

Meta's utilization of Nvidia AI chips for tax purposes has made it the largest recipient of this credit among publicly traded companies. In its tax filings, Meta labels its multi-billion-dollar AI data centers as 'pilot models,' qualifying them for the credit only when they are part of an experimental effort, as opposed to routine business operations.

This classification hinges on a 2021 IRS ruling that denied the credit to an Indiana shipbuilder for merely building new products. After deliberation, Meta began to categorize chips intended for AI data centers differently from those destined for standard purposes.

A tax specialist at BPM, Andre Shevchuck, described Meta's experimental label as "kind of wild and out there." Despite this, Meta's accountants acknowledge the uncertainty surrounding this strategy, with their unrecognized tax benefits growing by 45% to $18.74 billion from $12.9 billion. The strategy is not without risk; Meta has allocated an additional $18.74 billion as a buffer for potential challenges from the IRS.

While Apple, Amazon, Alphabet, and Microsoft also benefit from over $1 billion annually in research credits, Meta's approach has raised eyebrows, as it diverged from what the former sponsor of the credit, James Shannon, intended — to support researchers rather than manufacturing.

Written by urgent.news from Times of India's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at timesofindia.indiatimes.com →

More in AI

How to implement context decay so an AI agent forgets stale or incorrect instructions over time?

A year into a project, one agent's memory holds three instructions that should no longer win. It says to install packages with npm, and since June it also says to use pnpm; both are still there, and…

  • Closing outdated instructions marks them with a date they stopped applying and their successor.
  • Decay lowers an instruction's ranking over time if it's not used, confirmed, or contradicted.
  • Pushing down gives a negative mark to misleading instructions that caused the agent to act.

Scoop: AI companies plot "day after" scenarios for public revolt

Top executives at Anthropic, OpenAI and other AI companies are privately gaming out scenarios for a public and political revolt after a catastrophic AI event . Why it matters : These officials anticipate a large-scale event, most likely a cyberattack, that shuts down access to financial services, internet connectivity, or even power and…

More from Friday 9 October →