Artificial intelligence companies have spent years racing to secure more chips, build larger data centers, and train increasingly powerful models. But another resource is becoming just as important: new data.
And it’s making for some unlikely partners.
Google deciding to pay $10 million for a massive collection of internal business data from the defunct Spirit Airlines illustrates justhow valuable fresh, real-world information is becoming as AI developers search for material beyond the public internet.
The Spirit data reportedly includes employee emails, Microsoft Teams conversations, documents, spreadsheets, calendars, internal workflows, software, and operational information. Google says customer and credit card information is excluded and the data will be deidentified. But the bigger story is not Spirit Airlines. It is the growing competition for data that shows how people actually work.
AI Has Already Learned Much of the Public Internet
Large language models have traditionally been trained on enormous collections of books, websites, articles, code, and other publicly available material.That strategy has limitations. The web is finite, contains substantial duplication and low-quality information, and increasingly includes content that was itself generated by AI.
Developers therefore need new sources of high-quality information if they want AI systems to perform increasingly sophisticated tasks. Internal corporate records are particularly attractive because they capture something the public web often does not: the processes behind real work. Emails show how employees communicate and resolve problems. Documents capture decision-making. Spreadsheets contain operational structures. Internal messages reveal how teams collaborate, while software and workflows demonstrate how organizations actually complete tasks.
That kind of information could help train AI agents intended to perform business functions rather than simply answer questions.
RELATED: What Happens to AI Learning When Human Knowledge Runs Out?
Corporate Data Is Becoming an AI Asset
The competitive bidding for Spirit’s records offers a glimpse of how valuable such information may become.
AI training company Mercor reportedly bid as much as $7.5 million for the data before Google ultimately prevailed with its $10 million offer.
“Companies are sitting on decades of records that show how real work gets done, and that data is now some of the most valuable material for training and evaluating AI,” a Mercor spokesperson told Business Insider.
Google offered a similar explanation, saying Spirit’s internal information can be helpful in improving products and AI models.
Spirit is unlikely to be an isolated case. AI training startup Micro1, for example, has offered midsize companies between $100,000 and $2 million for access to anonymized corporate data.