Google Buys Spirit Airlines' Data for AI
When a company goes bankrupt, the things that are usually sold are obvious: airplanes, offices, equipment, branding rights, licenses or real estate. However, in the age of artificial intelligence, a much more abstract but extremely valuable asset is now added to this list: corporate data accumulated over the years.
This is exactly why Google's $10 million bid for the corporate data archive of Spirit Airlines, which is in bankruptcy process, is important. Because the agreement is not just a technology company purchasing data from an airline company. A sign of a larger transformation: Companies' past operations are now becoming independent entities with economic value for training AI models.
The scope of the sale is quite wide. Millions of records Spirit has produced over the years, including employee emails, Microsoft Teams conversations, calendars, spreadsheets, operations, marketing and productivity data, are at the center of the agreement. According to Reuters, Google wants to use this data for product development and training of artificial intelligence models. However, the transaction is not finalized yet; The hearing to approve the sale was postponed to September 9, 2026, following the confidentiality objection of the union representing former Spirit cabin employees.
This detail is important. Therefore, rather than saying "Google bought all of Spirit Airlines' data", it is more accurate at this stage to say that Google won the bankruptcy auction for its data archive and the transaction is awaiting court approval.
Why Does Google Want Old Emails from an Airline Company?
At first glance, the question seems a bit strange. Google already has access to some of the world's largest data sources. Why would a company that processes massive amounts of digital information through its search engine, YouTube, Android, Workspace, and countless other products pay $10 million for a bankrupt airline's old emails and Excel spreadsheets?
The answer lies in the ever-changing data needs of modern artificial intelligence models.
The Internet has provided an incredibly large source of text for models. Wikipedia, news, blogs, forums, books, code repositories and public websites played an important role in the development of major models. However, an important part of the internet is public communication.
Corporate life is completely different.
In a company's real-world day-to-day operations, information such as this occurs: How does one employee assign tasks to another employee? How do teams coordinate when an operational problem arises? How are meetings planned? Which departments does a customer problem go through? How does a company organize thousands of flights? How are financial statements related to operational decisions? How do teams communicate with each other during a crisis?
These are much more difficult types of data to find on the internet.
Spirit's archive is therefore not just “a lot of text”. A digital trace of how a real company works over decades.
And in the age of agentic AI, such data can be extremely valuable.
The New Need for AI Models: Learning How People Work
In the early days of ChatGPT and similar systems, the most important skill was answering questions. Much more is expected from the new generation models.
It is not enough for an AI agent to just write well.
He needs to understand a task.
It needs to find the files.
He needs to evaluate the calendars.
He needs to interpret emails.
He needs to understand at what stage the workflow is.
It needs to track the relationship between multiple employees or departments.
So AI needs to learn the behavior of the business world, not the language of the business world.
The corporate data of a large company like Spirit, spanning many years, contains exactly such examples.
An airline business is also an extremely complex organization. Flight planning, crew, maintenance, airport operations, pricing, finance, marketing, customer service and regulation work together.
Such a data set could therefore be of value not only for developing an “aviation-related model,” but more generally for developing AI agents that understand corporate tasks.
Google's statement states that the data will be used to "improve products and AI models," but it is not explained exactly in which models or with which training methods it will be used.
Therefore, there is not yet enough information to make more specific inferences, such as whether the dataset will train a specific version of Gemini.
10 Million Dollars is Actually an Interesting Number That Shows the Value of Data
The $10 million offered by Google may seem small compared to the billions of dollars of AI investments of the technology world.
But what is purchased here is not a physical asset.
Digital records that were once created as a byproduct of the company's daily activities.
According to Axios, the archive up for auction includes emails, chats, calendar information, documents, spreadsheets and other corporate records. In more detailed news, it is stated that the data set can reach an extraordinary scale of approximately 100 million e-mails and around 80 thousand e-mail accounts.
The real striking part is this:
Spirit did not write these emails years ago to sell them to Google.
Employees did not send Teams messages for the purpose of creating AI training data.
Excel files were not prepared for training models.
All this arose as a result of the normal activities of the company.
But today, this past alone is an asset worth millions of dollars.
This situation may seriously change the concept of data in company balance sheets in the coming years.
Could Archives Become the Most Valuable Asset of Companies?
It is not a new idea that data is valuable in modern companies.
Customer lists.
Sales history.
User behavior.
CRM data.
Their economic value has been known for years.
But generative AI and agentic AI make a different type of data valuable:
Organizational behavior data.
Who wrote to whom and when?
How did a project progress?
How were the problems solved?
What documents were created before a decision was made?
How did the departments coordinate?
What stages did a task go through?
Such data could theoretically provide artificial intelligence with not only information but also a way of doing business.
Therefore, it would not be surprising if not only customer base or intellectual property, but also new concepts such as "AI training value of data" will gain importance in company mergers and acquisitions in the coming period.
A company may go bankrupt.
The brand may disappear.
But its 30-year operational memory may still be valuable to another company.
Why is the Spirit Case More Interesting in the "Synthetic Data" Era?
AI companies are increasingly turning to synthetic data today. A model can create a question, answer, task or example for another model.
This method is scalable.
But it's hard to fully emulate the messiness of the real world.
Real company data is not clean.
People write half sentences.
Sends the wrong file.
Adjourns the meeting.
An employee says, “Let's do this like last week's file.”
Someone else knows the context.
Information is distributed across different channels.
Organizational reality is not like a tidy benchmark data set.
This is exactly why real job data can be of particular value in agent training.
Showing AI controlled and sterile tasks is not the same as providing millions of examples of how real people work over years.
It is noteworthy that other AI companies are also interested in the Spirit data set. According to Reuters, Google beat AI data company Mercor's $7.5 million bid at the auction. It was later reported that AI startup Micro1 was also preparing to submit a higher offer of $12.5 million.
So there is no opportunity that only Google sees.
The AI industry is increasingly placing greater economic value on real enterprise data.
But does saying "We Anonymized" Solve the Problem Completely?
The most controversial part starts here.
Spirit and Google state that personally identifiable information will be purged from the data set before sale. Customer information and personally identifiable data will be excluded from the sale or anonymized, according to Reuters.
However, the Association of Flight Attendants-CWA, which represents former Spirit employees, is of the opinion that this is not enough.
The union argues that the information of approximately 5,500 former cabin workers is in the archive and that the sale of employee data poses a serious risk in terms of privacy. The records under discussion are not just ordinary emails; Some news reports state that payroll, working hours, tax information, travel records and potentially sensitive employee information are also included in the archive.
Upon the union's objection, the court postponed the approval hearing of the sale to September 9.
This case raises a very important question in the age of AI:
A company's data may be owned by the company; So what will be the rights of the people speaking in that data?
Does an Employee Come to Work Thinking That Their Emails Will One Day Be Sold as AI Training Data?
I think the most disturbing part of the Spirit case is exactly this question.
When an employee emails a colleague in 2017, he probably doesn't think like this:
“This message could be sold to another tech company for AI training nine years after the company goes bankrupt.”
But today we are entering a world where this is technically possible.
In most cases, corporate accounts are not the personal property of the employee.
Data created in company systems is subject to certain company policies.
But legal ownership is not the same as a reasonable expectation of privacy.
The fact that it is legally possible to use an employee's conversation with his boss for AI training does not automatically make it acceptable for the employee.
That's why the Spirit agreement isn't just a data privacy debate.
Debate over economic ownership of employee data.
Why Is Employee Data More Vulnerable While Consumer Data is Protected
One of the most interesting issues in the union's objection is right here.
While the agreement includes clear mechanisms to protect customer information, it is argued that there is not the same level of protection for employee records. AFA demands that confidential information belonging to cabin crew be either excluded from sales altogether or given assurances similar to the protections provided to customer data.
This situation reveals an interesting gap in data protection law.
Most of the big privacy debates around the world over the last decade have revolved around the consumer:
Can websites track us?
Can we use cookies?
Can it hide our location?
Can I create a profile for advertising?
But in the age of AI, another question arises:
What can our employer do with our work data that we have accumulated over the years?
The answer to this question may be one of the important topics of employee rights in the coming years.
Can Anonymous Data Really Remain Anonymous?
Anonymization of data is usually:
name,
e-mail address,
phone,
employee number,
address
It means removing direct identifiers like
.
But with large, interconnected data sets, the problem is more complicated than that.
Let's think of an example:
“The crew manager who served on the Las Vegas-Chicago flight on March 12 took leave for health reasons the next day.”
The name may not be written here.
But when matched with other data sources, the person's identity can be inferred.
This is exactly one of the union's concerns. In the objection reported by Reuters, it is argued that the fact that the sales agreement requires the preservation of connections between data sets may allow the reconstruction of information about certain individuals or small groups.
This is a significant data privacy problem known as re-identification.
This is why data anonymization techniques will need to be rethought in the AI era.
Will Spirit Data Turn Gemini into a Better Travel Assistant?
At this point, Google's travel products naturally come to mind.
Google Flights.
Google Travel.
Google Search.
Gemini.
Maps.
It is clear that the aviation operation data that Spirit has accumulated over the years could be interesting for Google's travel technologies.
However, Google has not yet said that the data was purchased for a specific product.
The company states that the data can only be used for product development and improving AI models.
Therefore, a statement like “Google will teach Gemini to predict flight prices with this data” would be speculation at the moment.
But it is possible to evaluate potential areas of use.
Operational aviation data AI systems:
flight operations,
corporate processes,
complex timing problems,
interdepartmental coordination,
demand and productivity relationships
It may help model it better.
So the value may not be just in travel search.
Real World Company Data Could Be a Gold Mine for “Agentic AI”
I think the real big technology story is here.
In the AI world of 2026, the focus is increasingly shifting from chatbots to agents.
Chatbot does this:
“Explain the problems airlines may experience in bad weather conditions.”
If Agent:
“Find the operations that will be affected by tomorrow's bad weather conditions, identify the relevant teams, create an alternative plan, prepare notifications to the necessary people and record the changes in the calendar.”
The second task requires understanding real company processes.
And decades of emails, calendars, documents, and operational records from large companies can be outstanding training material for such systems.
Therefore, the Spirit case may be the beginning of a market that we may see much more frequently in the future:
Enterprise data for AI training.
The Next AI Race May Be About Data More Than Models
OpenAI.
Google.
Anthropic.
Meta.
xAI.
DeepSeek.
The performance gap between larger models is gradually narrowing in some tasks.
Model architectures are circulating in the open literature.
Researchers move between companies.
Open source models are getting stronger.
In this case, sustainable competitive advantage is sought elsewhere.
Computing power.
Distribution.
User network.
And unique data.
Data on the Internet is largely accessible to everyone.
Spirit's more than 30-year history of internal communications is not like that.
So in the future, an AI company's valuable asset may not just be its larger GPU cluster.
Data sets that competitors cannot access can also create a powerful moat, or competitive moat.
This Situation May Also Change Company Acquisitions
We may see interesting acquisitions in the future.
An AI company might buy a small software company not because it wants its product, but because it wants 20 years of customer support conversations.
The value of a logistics company may be in its operations data rather than its vehicle fleet.
A law firm's anonymized case workflow data could be valuable to AI legal systems.
A design agency has accumulated over the years:
brief,
customer revision,
presentation,
design decision,
project management
The archive may be of value for training creative AI systems of the future.
The result of this may be that data due diligence processes may also change.
When purchasing a company, only:
“How much income does it generate?”
not,
“What unique data do you have?”
The question will be asked.
There's a Huge Lesson Here for Design and Advertising Agencies
When it comes to data in the creative industries, Google Analytics or advertising performance usually comes to mind.
However, an agency's project history over the years can be a much more valuable data set.
For example, let's consider thousands of customer revisions.
Which design was rejected?
Why was it rejected?
Which did the customer choose in the end?
Which campaign performed better?
How many revisions were made in a logo project?
What was the difference between the initial brief and the final?
If an AI system can learn these relationships, it can begin to understand creative decision processes rather than just producing designs.
The difference here is huge.
Generative AI today:
“Create a logo for a coffee brand.”
may say.
Agent trained with corporate creative data will be in the future:
“This client rejected minimalist options in the previous three projects, opting for warm colors and turning to serif typography in the latest revisions. So let's try these three directions in the first presentation.”
He may say.
This is much more advanced than simple image generation.
But Opening Agency Data to AI Brings the Same Privacy Problem
There is a huge responsibility in front of this technology opportunity.
In the agency archives:
Products that have not yet been released,
Projects within the scope of NDA,
customer strategies,
price offers,
campaign budgets,
brand research,
personal data
Available.
Therefore, even if it is technically easier to say "let's teach past projects to AI", it requires serious governance in legal and ethical terms.
The Spirit case is also a strong warning for creative companies:
We need to think about how the data we produce today in the normal workflow can be used in the future.
Data is now a product that needs to be designed
Here we come to an interesting design problem.
Companies have been producing data for years, but much of it is scattered.
File names are inconsistent.
E-mails are in different folders.
Presentations are in old formats.
Metadata of the documents is missing.
The same customer's information is in five different systems.
In the age of AI, the quality of this archive can directly translate into economic value.
Therefore, in the future, companies will need to create not only a brand guideline, but also some kind of data architecture guideline.
Which data is named and how?
How long is it kept?
What information is sensitive?
What data can go into AI training?
What data can never be used?
Is employee permission required?
Is there an AI use clause in customer contracts?
These are becoming part of the company strategy rather than small details of the IT department.
The Analogy of “Data is as Valuable as Oil” Is No Longer Enough
There is a cliché sentence that has been used for years:
“Data is the new oil.”
Data is the new oil.
The Spirit incident shows why this analogy is a bit incomplete now.
Oil is consumed when used.
Data does not run out as it is used.
Same data:
analysis can be done,
AI can train the model,
can be matched with other data,
Can develop new products,
Can be re-licensed.
Moreover, the archive, which was considered worthless at the time, may gain value years later when new technology emerges.
Spirit's old emails might not have been thought to have such economic value in 2020.
The value equation changed when the AI agent race began in 2026.
That's why data is starting to look more like intellectual property than oil.
When a Company Goes Bankrupt, Its Digital Memory Doesn't Die
This is perhaps the most striking aspect of the Spirit Airlines example.
The company no longer operates flights.
Their planes can go to other places.
Employees can move to different jobs.
The brand may disappear over time.
But the digital memory it created over the decades lives on.
And another company can buy this memory and use it to develop new technologies.
This situation will also create new questions in bankruptcy law.
A company:
customer data,
employee data,
emails,
corporate chats,
AI educational value
Can it be sold like other assets in the bankruptcy estate?
Under what conditions?
Who should give permission?
Do former employees have a say?
Can the data be used for a completely different purpose later?
The decision in the Spirit case could therefore be an important precedent that could affect not only the sale of an airline but also similar transactions in the future.
The Biggest Market of the AI Era Perhaps Will Be “Legacy Data”
Today, the biggest investment of the technology world is going to GPUs.
However, after a point, finding quality data to train powerful models may become a bigger problem.
Internet contents were scanned repeatedly.
Books were used.
Code repositories were used.
Academic publications were used.
The next valuable resource may be closed corporate data.
Hospitals.
Banks.
Airlines.
Factories.
Logistics companies.
Call centers.
Software companies.
Each of these has data sets that describe how a particular job is actually done in the world.
Therefore, it would not be surprising if a new data economy emerges in the next few years.
Companies can license AI training data.
Private data markets may emerge.
Anonymization companies can grow.
Preparing an “AI-ready dataset” may turn into a new consultancy category.
And companies may begin to look at their historical archives as financial assets for the first time.
What is the Real Significance of Google's Spirit Move?
$10 million is not a big acquisition for Google.
Spirit Airlines is not a technology company.
The agreement is not a company acquisition in the classical sense.
But the symbolic value of the event is very high.
Because it shows us the new equation of the AI economy of 2026:
Model + calculation + real world data.
The first two elements have been talked about for a long time.
The third is becoming increasingly critical.
The Internet taught AI what the world said.
Corporate archives can teach AI how the world works.
And the second type of data will perhaps be even more important for the future of agentic AI.
Result: Companies Are Now Producing Not Only Products, But Future AI Training Data
The Spirit Airlines case is, at first glance, a strange bankruptcy news.
An airline went bankrupt.
Google offered $10 million to get its data.
But look a little closer and it covers several of the most important AI discussions of 2026 at once:
data economy,
employee confidentiality,
anonymization,
agentic AI,
institutional memory,
bankruptcy law,
AI education rights.
Perhaps the most important thing that companies need to realize today is this:
Every email, every meeting recording, every operational report and every work flow is no longer just a record of the past. Data that can be training material for an artificial intelligence in the future.
This is a huge economic opportunity.
But it is an equally great ethical responsibility.
Because in the age of AI, it is no longer enough to just ask this question:
“Whose data is this?”
Another question is coming:
“This data can be used to teach what to whom in the future?”
This is where the debate between Google and Spirit Airlines really matters.
Nur Oğuz