Meta's single earnings day on October 21 carried two numbers worth reading together: a $25 billion civil penalty for deceiving users about how their data was handled, and a raised capital-expenditure floor of $135 billion for the AI infrastructure that consumes data at industrial scale3,9. The juxtaposition is suggestive but requires a caveat: the New Mexico verdict penalizes deceptive statements about historical data-handling and hate-speech moderation, not AI training specifically. No court has yet attached a dollar cost to the use of user data for model training. Still, the underlying asset is the same—user-generated content whose terms of acquisition platforms defined unilaterally—and the verdict demonstrates that courts can retroactively reprice the bargain under which that content was collected. Whether that repricing eventually reaches AI training pipelines is an open question, not a settled one, but the financial scale of the answer now has a data point.
The New Mexico case dates to the Cambridge Analytica scandal, when data firm Cambridge Analytica harvested data from more than 50 million Facebook users to craft political ads during the 2016 election6. A jury in Santa Fe found that Meta had violated the state's consumer protection laws over 43 million times, based on deceptive statements by the company and its top executives about how it handled hate speech and user data10. Judge Francis Mathew could have imposed up to $5,000 per violation under New Mexico's Unfair Practices Act; the jury's math initially yielded $219.5 billion, which Mathew reduced to $25 billion. In his filing, the judge cited the "extraordinary number of willful violations found by the jury, the duration and repetition of the conduct, the breadth of subjects on which violations occurred, the involvement of senior Facebook leadership, the size of New Mexico audiences reached, and the clear intentional and systematic nature of Facebook's repeated misstatements". Jurors saw video deposition testimony from CEO Mark Zuckerberg and former COO Sheryl Sandberg.
New Mexico was the only state that chose not to settle its claims as part of a nationwide legal battle. Meta agreed to a $17.1 billion settlement with dozens of other U.S. states earlier, and the Q3 earnings report already carried a $9.89 billion charge tied to that broader settlement5. Meta has said it expects to appeal the New Mexico decision.
ANALYSIS The verdict's per-violation mechanism is notable: the jury found 43 million violations at up to $5,000 each, producing a theoretical maximum of $219.5 billion. The judge reduced that figure to $25 billion, a sum that still dwarfs conventional regulatory fines. Meta's stated intent to appeal leaves the final amount uncertain, but the scale of the penalty even after reduction demonstrates how state consumer-protection statutes with per-violation maximums can produce outcomes far larger than federal enforcement actions typically yield.
The earnings report itself was strong. Third-quarter revenue rose 26% year over year to $64.6 billion, ahead of the $63.37 billion Wall Street had expected. Advertising impressions across Meta's family of apps climbed 10%, while the average price per ad jumped 14%8. Net income came in at $9.86 billion, or $3.83 per diluted share. Meta guided fourth-quarter revenue to $73 billion to $76 billion. Shares rose more than 4% in after-hours trading to $726 as of 4:05 p.m. Wednesday, up from a closing price of $697.52.
But operating income fell 37% to $12.99 billion, dragged down by the $9.89 billion settlement charge. ANALYSIS The gap between the reported $12.99 billion operating income and the $9.89 billion settlement charge embedded within it illustrates how profitable Meta's core advertising engine remains, and how large the legal overhang has become relative to it.
To fund what comes next, Meta announced a new Class C common stock, nonvoting shares that will begin trading in January 20274,1. The new stock will not affect Zuckerberg's voting control. "The scale of investment required to compete, sustain growth, expand infrastructure, and pursue new strategic initiatives amidst the AI transformation taking place in the industry is significant," Meta said in its press release. CFO Susan Li said on the earnings call that the new class of shares will "expand our funding tools and give us flexibility to invest behind the unprecedented momentum we're seeing across every part of our business driven by AI". The shares will also go toward potential acquisitions and employee compensation.
Meta simultaneously raised the bottom of its 2026 capex range to $135 billion from $130 billion, keeping the ceiling at $145 billion. The money is directed at AI infrastructure: chips, data centers, and development of Muse, Meta's consumer AI agent. Muse Wonder, the model powering that agent, currently scores 82.11 on FrontierMath tiers 1–3, 70.73 on FrontierMath tier 4, 1651.51 on Coding Arena Elo, and 1498.62 on Arena Text Factuality, all independently verified as of October 21 [ANNEX-B1][ANNEX-B2][ANNEX-B3][ANNEX-B4]. Its list price sits at $4 per million input tokens and $20 per million output tokens as of the same date [ANNEX-P1].
The structural tension is real but should not be overstated. Meta's AI ambitions depend on massive capital for compute and massive volumes of data for training. The new share class addresses the first. The New Mexico verdict addresses the second only indirectly: it penalizes deceptive representations about data handling, not the act of training models on user content. The penalty is a retroactive invoice for misrepresentation, not for data use per se. ANALYSIS The distinction matters. If future plaintiffs can show that representations about AI training were similarly deceptive, the per-violation math demonstrated in New Mexico could apply. That chain of reasoning is plausible but unproven; no court has yet drawn it.
Zuckerberg framed the quarter in forward-looking terms. "Our investments are paying off, the core business is strong, and we're seeing velocity across the board," he said in a statement. ◆ The market agreed on the earnings; the after-hours move suggests investors are pricing AI upside above legal downside. That could reflect a rational judgment that the penalty is a one-time cost tied to historical misconduct, not a recurring tax on AI data acquisition. It could also reflect underpricing of a tail risk whose magnitude the New Mexico verdict just made legible. Distinguishing between the two requires watching whether new data-liability claims emerge that target AI training specifically.
Three markers to watch. First, the timeline and terms of Meta's appeal of the New Mexico verdict: a significant reduction would limit the precedent, while an affirmation would invite other holdout jurisdictions to litigate rather than settle. Second, the uptake and pricing of the Class C shares when trading begins in January 2027, which will reveal whether public markets are willing to fund AI capex through dilutive, nonvoting equity. Third, whether Meta's Q4 guidance range of $73 billion to $76 billion holds, because sustained revenue acceleration is the only thing that makes $135 billion-plus in annual capex look like investment rather than overreach.