The Market of the Dead: The Lucrative Business of AI Data

The digital salvage economy: The value of startup remains

In the current tech ecosystem, where startup mortality has reached record levels, failure is no longer an endpoint but a source of unexpected liquidity. What was considered digital trash just five years ago—forgotten servers, abandoned GitHub repositories, or archived Slack threads—has been transformed into a strategic asset: fuel for the next generation of Artificial Intelligence. Companies like SimpleClosure and Sunset have institutionalized this 'digital salvage,' acting as data brokers that connect the remains of failed startups with AI labs hungry for proprietary information.

Why is startup data worth so much?

The AI market is undergoing a high-quality data shortage crisis. After exhausting most indexable public web content, labs have begun hitting a wall of diminishing returns. Synthetic data, generated by other models, has proven insufficient to reach higher levels of reasoning. This is where the concept of 'operational exhaust' or 'data exhaust' comes in.

Unlike public data, a startup's internal information reflects human reasoning in action. According to Brendan Mahony, CEO of Sunset, this data is comparable to wood shavings in a carpentry shop: once considered waste, they are now the raw material for creating high-density boards. The value lies in the 'how': how a critical bug was resolved, how a sales team negotiated, or how a workflow was structured in Notion. This granularity is what allows current models to learn to reason, not just predict the next word.

Keelan Miskell, founder of a crypto startup that recently closed, illustrates this paradigm shift: he managed to obtain about $25,000 for his codebase and nearly $15,000 more for his workspace files (Google, Notion, Figma, and Slack). For Miskell, this return was not only a financial relief but a validation of years of work that would have otherwise disappeared into digital oblivion.

Impact and controversy in the market

The rise of this economy is intrinsically linked to the funding crisis facing the sector. Data from Carta and AngelList indicate that startup closures have increased between 26% and 56% year-over-year. Given this landscape, an entire ecosystem of 'tech undertakers' has emerged—from liquidation coaches to legal specialists—that now include data sales as a standard business line. SimpleClosure, for example, reports that 47% of its liquidation processes already include the sale of data assets. The company projects returning more than $10 million to failed startups through its Asset Hub by the end of 2026.

However, the valuation of these assets is purely speculative. There is no data futures market; the price is set based on ad hoc criteria: number of employees, years of operation, volume of data, and, above all, the intensity of collaboration tool usage. The case of Spirit Airlines, where Google acquired data assets for $10 million, while a larger corporate case, sets a dangerous precedent: the idea that a company's data is a liquid asset that must be sold to the highest bidder before closing shop.

Ethical and privacy challenges

This market operates in an ethical gray area. Although intermediaries insist on anonymization, the ability of modern models to perform reverse engineering and re-identify individuals from conversational patterns is a real risk. Users of these startups, by accepting original terms of service, hardly gave explicit consent for their private interactions to become training material for commercial AI models.

Historically, this recalls the pre-GDPR social media data mining era, but with a critical difference: the technical and specific nature of 'operational exhaust' makes the data much harder to purge once integrated into the weights of a language model. If training is irreversible, the privacy of the original users is permanently compromised.

The future: A liquidation standard?

We are facing a gold rush that, like all others, lacks clear regulation. It is highly probable that in the coming years we will see the inclusion of 'training data sale' clauses in the terms of service of any new startup, turning user privacy into a residual asset from day one. Meanwhile, the real efficacy of this data remains an unproven hypothesis: we do not know if a thousand Slack threads from a logistics startup actually improve a model's reasoning capacity, or if we are facing a data bubble where irrelevant information is bought at inflated prices.

For founders, digital salvage represents a final financial lifeline. For the AI industry, it is a desperate search for excellence in a world running out of quality data. What is undeniable is that, in the AI era, the death of a company no longer means its disappearance, but its recycling into the servers of tomorrow's tech giants.