A $1.5 Billion Wake-Up Call for Every AI Company on Wall Street

Key Points

• A federal judge granted final approval to Anthropic's $1.5 billion settlement with authors, putting a real market price on AI training data for the first time.
• Data licensing deals, including Reddit's reported $60 million-a-year agreement with Google, show large platforms already treat user data as a recurring revenue line.
• Investors seeking exposure can weigh small-cap pure-plays like Datavault AI alongside hyperscalers such as NVIDIA, Meta Platforms, and Alphabet that depend on a steady supply of licensed data.

A federal judge just approved a $1.5 billion settlement that puts a substantial legal cost on one source of artificial intelligence training material: pirated books

The judge separately ruled that training Claude on lawfully acquired books qualified as fair use. The settlement covers 482,460 works and is expected to pay roughly $3,000 per work, a much narrower group than the more than 7 million pirated books identified in Anthropic’s broader library.

Dan Novaes, CEO of Mode Mobile, calls this a turning point for how markets value human data. His firm operates a rewards platform built on consented data-sharing, and he argues the settlement is the clearest signal yet that the human data market could grow into a market worth $1 trillion or more by 2033.

A Market Price Emerges for Training Data

For most of the last two and a half decades, Novaes says, the deal was simple. Platforms collected user activity, trained smarter systems on it, and users got a free product in return. That arrangement is now under legal pressure not just in the United States but in Germany and other jurisdictions applying stricter consent standards, echoing how GDPR reshaped data privacy years earlier.

The practical effect shows up in licensing deals already being signed. Reddit Inc NYSE: RDDT has reportedly licensed its user-generated content to Google for roughly $60 million a year — high-margin revenue built entirely on comments and posts the platform already owned.

Models Still Need Millions of Human Hours

Large language models still hallucinate, and Novaes says closing that gap increasingly requires specialized human input rather than more raw internet text. Companies now pay subject-matter experts, from physicists to physicians, to refine model responses in a process known as reinforcement learning from human feedback. Mercor, a marketplace connecting specialized workers to AI labs, reportedly grew from $1 billion to $2 billion in annualized revenue in four months this year, according to Novaes.

Robotics adds another layer entirely. Training a robot to fold laundry or change a tire requires millions of hours of what Novaes calls egocentric video: first-person footage of humans performing that exact task. Meta Platforms Inc NASDAQ: META took a 49% non-voting stake in Scale AI at a $29 billion valuation last year to help meet that demand, a deal that also reportedly cost Scale some of its other frontier-lab clients over data confidentiality concerns.

Publishers Start Getting Paid to Play

Cloudflare Inc NYSE: NET, which sits in front of a meaningful share of the web, will begin blocking mixed-use AI crawlers from ad-supported pages by default starting Sept. 15. That forces AI labs to separate search indexing from model training and agent traffic, giving publishers new leverage to charge for content that bots previously scraped for free.

Novaes sees this as part of a broader pattern: courts, regulators, and infrastructure providers are converging on the same idea. Consent and compensation for data are becoming the norm, not the exception. A similar suit brought by a coalition representing nearly 400 newspapers sued OpenAI and Microsoft in June 2026, alleging that their content was copied without permission or compensation. Novaes doesn’t expect the pace of these cases to slow.

Where the Investable Opportunity Sits

For investors, Novaes frames the data supply chain as the last piece of the picks-and-shovels trade that began with NVIDIA Corporation NASDAQ: NVDA chips and extended into energy and rare earth names. The data layer, he argues, is the piece the market hasn't fully priced yet.

Novaes points to SpaceX NASDAQ: SPCX as a cautionary example. By the time most retail investors got a chance to buy in, the company was valued at $1.75 trillion, and the biggest gains had already gone to early venture backers. He argues that Anthropic and OpenAI are tracking the same path, with valuations reportedly jumping from roughly $300 billion to $1 trillion or more within months. Mode Mobile itself is not publicly traded. The company has raised capital through Regulation CF crowdfunding and has more than 64,000 shareholders, positioning it, in Novaes's view, as a way for everyday investors to get in before a public listing rather than after.

His caution: this is a concentrated market. Roughly seven frontier labs make up most of the demand, so companies dependent on just one or two of those clients carry real customer-concentration risk. He points to diversified revenue, not just data licensing, as a sign of a sturdier business model.

Datavault AI Inc NASDAQ: DVLT trades as one of the more direct public plays on data monetization, though its small size makes it considerably more volatile than the hyperscalers building or buying their way into the space. The upside case rests on continued regulatory and legal pressure forcing more licensing deals into the open; the risk is that a handful of dominant labs strike direct deals with the largest data holders, leaving smaller data-supply companies to compete for scraps.

Either way, the settlement made one thing clear: data has a price now, and someone has to pay it.