AI data gold rush hits snags: pricing, privacy collide

The gist
The AI data gold rush has hit a wall as soaring demand for corporate datasets collides with wild-west pricing and a tidal wave of privacy crackdowns.
What to know
- With public web data drying up, AI developers are racing to license proprietary corporate records—fueling a market expected to hit $40.35 billion by 2035.
- Data is being priced 'by the terabyte' and sometimes padded with thousands of Frozen 2 copies, as buyers and sellers scramble for value in a chaotic marketplace.
- California’s Delete Act took effect August 1, 2026, triggering over 500,000 deletion requests in a month and legal warnings that AI data scraping is 'the largest theft of labor in human history.'
Corporate Data Becomes Currency
As public web signals vanish, proprietary company records are now the essential—and tradable—fuel for AI, with contracts and exclusive partnerships reshaping what data is worth.
The engine of this market is the exhaustion of easy, observable web signals and the resulting turn toward data that companies own, measure and can charge for. As the Explainer noted, AI’s effects often appear as absences — “something that no longer happens” — which means they vanish from conventional public datasets, so buyers increasingly need proprietary corporate records to see what changed and to train or evaluate models on behavior that free web data cannot capture.
Once that information became necessary, it also became tradable: the Explainer said, “Every contract will settle against PYMNTS Intelligence data, built over nearly 30 months of continuous measurement across consumers, enterprises and public companies, with more added every month.” That same piece underscored how pricing mechanisms are forming around access to such datasets — “Starting in September, people will be able to trade on what’s otherwise been so very hard to put a price tag on,” under “a two-year exclusive partnership” — while PR Newswire captured the broader commercial shift in its headline: “Alternative Data Market to Reach USD 40.35 Billion by 2035 as AI Developers and Corporations Expand Data Buying, Credence Research Finds.”
Valuations Skewed by Guesswork
With no industry standard, datasets are priced by crude proxies like terabytes or padded with irrelevant files, exposing deep confusion about what AI-ready data is truly worth.
The clearest sign of an immature market is that participants still price datasets with blunt stand-ins rather than any shared measure of usefulness. Token Dispatch reported that one broker valued a dataset “by the terabyte, the current standard because nobody in the industry has figured out how to calculate the data’s actual worth yet,” a description that matters because it presents terabyte pricing not as an odd shortcut but as the prevailing rule in a market still missing stable fundamentals for what corporate data is actually worth to a model buyer.
Once pricing is anchored to rough proxies, mismatches become almost inevitable: Token Dispatch described a lab opening purchased files and finding “about 4,000 copies of Frozen 2 buried at the bottom of the folder structure because someone padded the whole thing with Disney movies to make it look bigger.” Marketing School’s examples show the same fuzziness at the offer stage, with brokers saying a company with “a couple hundred people” might fetch “a million dollars… up to like five million,” while one with “50 people” might get only “a hundred grand,” underscoring how inconsistent heuristics leave buyers and sellers far apart on value.
Revenues Rocket Amid Data Frenzy
A late-2026 buying spree sent company revenues soaring up to 18-fold, signaling an aggressive new phase as AI buyers locked in bigger deals and million-dollar data access.
By late 2026, the AI training data market was no longer merely expanding; it was accelerating fast enough to show up in company revenue lines and in the tone of market coverage. BigGo Finance captured that inflection directly in its headline, “AI Training Data Market Booms: Snorkel AI Revenue Surges 18-Fold in One Year,” using Snorkel AI’s extraordinary one-year jump as concrete evidence that buyers were committing materially more money and that deal flow had moved into a new, more aggressive phase.
That matters because revenue growth at that scale is not a vague sentiment indicator but a hard signal that spending and contract size both stepped up sharply in the closing months of 2026. As BigGo Finance framed it, Snorkel AI’s 18-fold surge served as a measurable proxy for broader market momentum, confirming that the period was defined by rapidly rising budgets and larger transactions, including reported broker payments reaching as high as $2 million for a mid-sized company’s data access.
Privacy Laws Hit Data Deals
Regulators and mass deletion requests are turning privacy compliance into a costly obstacle, while legal and ethical battles over consent and confidentiality threaten the entire AI data trade.
Legal resistance is no longer theoretical: on August 10, CalPrivacy announced an order against Iowa data broker LocateSmarter, LLC alleging it violated both California’s data broker law and the CCPA after requiring consumers to submit extra sensitive information to opt out, while a separate order said Cybba “failed to register as required under California’s data broker law” even as it sold personal information including “geolocation data, identifiers, internet activity data, and inferences.” Those cases show that once corporate data contains personal information, AI-adjacent brokerage can collide directly with privacy rights and basic compliance duties.
Deletion rights are also becoming a practical constraint on the market: California’s “new data broker deletion requirements (known as the Delete Act) went into effect on August 1, 2026,” and CalPrivacy orders required brokers to access “CalPrivacy’s Delete Request and Opt-Out Platform (DROP)” to process future requests, while by August 25 more than 500,000 California residents had filed such delete requests from data brokers. That volume turns deletion from an abstract consumer right into an operational burden for any company trying to license, retain, or repurpose corporate datasets for AI use.
The ethical challenge runs beyond privacy into consent and confidentiality: according to unsealed filings, Microsoft’s director of Applied Science, Brent Hecht, called AI scraping “the largest theft of labor in human history” and warned of “an astonishing theft of unprecedented proportions,” while Microsoft data showed click-through to The New York Times falling by as much as 93%, threatening “the economic foundations of its essential suppliers.” In professional services, the concern is sharper still: ABA Formal Opinion 512 says lawyers generally need informed client consent before using self-learning AI, and Swidler & Berlin v. United States — the “Vince Foster notes case” — confirms privilege survives “even the client’s death,” meaning a matter “closed in 2014 is every bit as privileged today” and “historical is not a loophole.”



