Google Paid $10 Million for a Bankrupt Airline's Data: Data Is the Moat in the AI Era
The most valuable thing an enterprise owns in the AI era is its data — Google paid $10 million for a bankrupt airline's; AI agents holding legitimate credentials are the new top threat; and data infrastructure has to be rebuilt.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Models and compute have zero switching cost; only data is uniquely yours
Eon's founders argue that models and compute carry almost no switching cost, so the only asset an enterprise truly owns alone is its data. Two days ago Google bought data from the bankrupt Spirit Airlines for ten million dollars; the other bidder was reportedly Mercor. This is no longer an isolated case: CEOs are regularly asked whether they would sell their data, and AI labs have even gone to Wall Street to buy data from hedge funds. Real enterprise data has become the scarce fuel for training agents — even a bankrupt company's data can end up on the auction block.
Real enterprise data is scarcer than people think; what's public is mostly synthetic
The host pushes back that "data is the new oil" didn't hold up in the slogan era — but post-training and reinforcement learning have turned it into real demand. Companies like Applied Comp offer fine-tuning services against specific datasets, yet Eon's view is that genuinely usable real enterprise datasets are extremely scarce, and most public data is synthetic. Spirit Airlines' data can train an airline customer-service agent, but it also works as a full-scope sample of how a large enterprise operates — hierarchy, and how middle management and workers coordinate — which is exactly the material agent training lacks most.
Data stays locked not because tools are missing but because incentives don't line up
There are actually plenty of data tools; the problem is that data is locked inside business units. Data teams used to pick their own projects and tools (Fivetran, DBT, Monte Carlo), and now CEOs, educated by ChatGPT, demand that the data be activated — but it is scattered across systems twenty years old, including servers nobody dares turn off. Getting the data out means pulling in engineers, at the cost of security, compliance and production disruption; the incentives of business owners and platform teams simply don't align. Eon's approach is to first automatically map, classify and build a semantic layer, then keep supplying data without affecting production.
Security models were built to stop humans; what's breaking in now is agents
Existing security models are designed for human threats. At AWS, Eon lived through a customer whose environment was 60% exposed to ransomware because resources hadn't been correctly mapped, classified and tagged. Now the same threat comes from non-human actors — an AI agent walks into the database with legitimate credentials and can drop a table in an instant. The detection and recovery methodology is similar, but it happens extremely fast. A few months ago nearly every enterprise customer was either afraid of agent attacks or had already been hit; organizations have to design their defenses assuming they are already compromised.
The more agents you have, the more indispensable dashboards become
The enterprise data stack was built around humans asking questions through dashboards; agents will reason dynamically over larger datasets, reach into SaaS and historical data, and act directly. The guests believe dashboards won't disappear — they'll become more common, because humans need them to understand what agents have been doing in the environment. Coding agents will write most of the code, agents will activate other agents, and tracking non-human identities (NHI) becomes a core problem, with NHI security companies proliferating. When non-technical employees build agents with Lovable, they themselves don't know where the data compliance boundaries are, so a new kind of actor appears inside the organization that the rules don't constrain.
Old pipelines don't connect to each other because full context was never needed
Take buying coffee with a card: the transaction is written into some database, extracted, sent somewhere else and processed on its own, and the pipelines have no connections between them — because full context was never needed. You only handled the questions you had defined in advance. Today, if you can collect data intelligently, store it efficiently and activate it, you let teams produce combinatorial insight: merge the list of "people in New York who like burgers" with "people in New York who like pizza" and you get a new segment. Data ingestion volumes are now growing insanely, old ETL tools can't handle noise from that many sources, and vendors like Databricks are reinventing themselves too.
The AI transition is faster than the cloud's, yet customers have less control
Eon's two founders ran large-scale cloud migrations at AWS, where customers routinely had anywhere from thousands to hundreds of thousands of servers, and the transition demanded investment in both technology and people. The AI-era transition runs at cloud speed amplified — but customers are losing control: afraid the system breaks, afraid data leaks, afraid IP walks out the door, so control itself becomes the inhibiting factor. Cloud was abstract; AI is something anyone can grasp (the ChatGPT moment), so boards are pressing companies to use AI from both ends, opportunity and fear. For a traditional enterprise that needs to deploy fast, the only way to shorten the sales cycle is to bring in Silicon Valley engineers who arrive with a mature solution and a playbook.
The fastest AI transition for a large company is to just buy a startup
The way enterprises consume software is changing. Big traditional companies want AI, but their internal processes take a year or two, so they bring in top-tier engineers and off-the-shelf solutions to accelerate; product-led growth (PLG) works especially well in AI infrastructure, and Cognition is a case of going PLG first and then into large accounts. The more radical path is to acquire a startup outright and convert the company into an AI company, lifting gross margin and efficiency faster. The guests believe "the revolution is only beginning" — most companies are still at the start of their AI journey, but all of them will end up being pushed forward.
In their own words · checked verbatim
The most valuable thing that you have Is actually your data.
They bought data for $10 million, because they think it's very important in a dark perspective.
We have a lot of data in the organization. We now realized data's new oil.
I actually think we'll see more dashboards because this will be the only way to kind of let us figure out what they have going on in the wall.
Figures
| Share of a customer's environment exposed to ransomware | 60% | 15:00 |
| Server counts involved in cloud migrations | thousands to hundreds of thousands | 28:13 |
Glossary
- NHI / Non-Human Identity
- The identity credentials used in systems by machine accounts, service accounts and AI agents — the key concept in agent security.
- PLG / Product-Led Growth
- A go-to-market strategy where the product itself attracts and converts customers, with enterprise sales following to expand the account.
- Data Foundation
- Eon's proposed unified data layer, bringing storage, protection, classification, the semantic layer and AI access together in one place.
How to listen
Enterprise CTOs, data platform leads and AI engineers who care about data assets and security in the AI era, plus investors valuing data companies.
The opening cold open teaser is skippable; start at 1:27 for the substance.