A Bankrupt Airline's Emails Went for $10 Million. Then micro1 Said It Would Spend $1 Billion on Yours.

In August, Google won an auction for the digital remains of a dead airline. Its $10 million bid covers roughly 100 million Spirit Airlines emails, about 500 million Teams messages, and 516 code repositories holding some 30 million lines of software.

A startup called micro1 wanted that pile too. On Sept. 3 it filed a late bid of $12.5 million. It hasn't won. Beri reports that its CEO, Ali Ansari, called Google's price "actually quite low." Five weeks later, on Oct. 9, micro1 announced what it wanted to do next time: spend $1 billion over 12 months buying data from companies that aren't bankrupt, and paying them for it.

That's the story. Here's how it works, who's at risk, and where it fits.

The Auction That Started It

Spirit stopped flying on May 2. Its records went on the block, with a privacy ombudsman, Lucy Thomson, reviewing the sale for the bankruptcy court. Google won at $10 million, and Mercor, another AI data company, is the alternate bidder at $7.5 million.

The ombudsman recommended approval on Oct. 5, provided the personal information of Spirit's 97 million customers is left out. Google told PYMNTS: "We are not looking to buy any personal information from Spirit." It said the information "will either be completely excluded or will be deidentified by an independent third party before Google receives any data."

Then came the pushback. More than 100 members of Congress, led by Rep. Steven Horsford and Sen. Elizabeth Warren, wrote to the CEOs of Spirit and Google on Oct. 8 about employee data. "No employee data should move until real, enforceable safeguards are in place," Horsford said. A software maker, Springshot, also objected.

This week the flight attendants' union, AFA-CWA, announced a deal with Google that excludes medical and disability history, drug and alcohol testing records and payroll records, and adds an independent verification agent chosen with the union's input. "Without our union, there would have been no one to speak for workers in this fire sale of their private information," union president Sara Nelson told Business Insider. BigGo reports a judge is expected to approve the sale next week. We haven't seen a ruling.

What micro1 Is Offering

The $1 billion program is called Company Data Partnerships. Citi and Hercules Capital are providing the capital, according to The Next Web. Nobody has said whether $1 billion is a firm commitment or a ceiling.

What micro1 wants is the ordinary paperwork of running a business: standard operating procedures, internal documents and knowledge bases, CRM data, project histories, QA processes, support tickets, and records of how people made decisions. It names six areas: software and engineering, finance and accounting, customer support, sales and CRM, legal and contracts, and operations and logistics.

Who can sell: companies with 30 or more full-time employees and established documentation, mostly in English, with the US first and other Western markets after.

What it pays, per micro1's current data page:

At launch, The Next Web reported the tiers as more than $100,000, more than $500,000 and more than $1 million, so the page has changed since. Separately, micro1's page says it licenses agent trajectories, the step-by-step records of how an AI agent worked through a task, for $1 million to $2 million a year. Beri reports a referral program that pays up to $50,000.

On the terms, micro1 says the company keeps ownership of its data, personal information is removed "when relevant," and some datasets are rewritten synthetically to loosen their ties to the original records. Companies can review samples first. micro1 says it won't resell or distribute data until personal information is anonymized, and that it licenses data only to labs and enterprises in the US and allied countries. On payment, its process page promises payment in full within 30 days of acceptance, while its FAQ describes a review period after transfer, then an invoice, then payment within 30 days.

Why would anyone want it? The data becomes training environments, simulated versions of real business operations where an AI agent can practice a workflow and get scored. As micro1 put it in its announcement: "Real business operations involve incomplete information, competing priorities and exceptions that require judgment."

Who micro1 Is

micro1 describes itself as a San Francisco data lab that helps AI research companies and enterprises train frontier models and evaluate agents. It runs three in-house labs, one of them for robotics.

The money figures are secondhand. Beri, relaying Forbes Australia, says micro1 raised more than $100 million at a $4 billion valuation, citing two people familiar with the deal, and lists Microsoft, Amazon and the robotics company 1X as customers. Beri also relays a TechCrunch report of a $500 million gross annual run rate. micro1 declined to comment on those figures, and we haven't seen the primary reports.

The Bigger Picture

Think of AI training data in three phases.

Phase one was the open web. Labs scraped everything. That has limits. Epoch AI's research estimated the stock of high-quality public human text at around 300 trillion tokens and projected that models could use all of it somewhere between 2026 and 2032. Rights holders noticed, and the market for licensing text opened. When Reddit went public, it disclosed AI data licensing contracts worth $203 million in total.

Phase two was paying experts. Labs hired lawyers, doctors, scientists and coders to write examples and grade answers. Scale AI, Surge AI, Mercor and others grew into the middlemen. BigGo, citing The Information, reports Mercor's first-half 2026 revenue at about $614 million, roughly 91% of it from foundation-model companies, with OpenAI, Anthropic and Google DeepMind among its reported customers. The same report says Mercor was planning a $500 million raise at a $20 billion valuation. We haven't seen confirmation that it closed.

Phase three is paying for process. The scarce thing isn't text anymore. It's how work actually gets done: how a team handles a refund, closes the books or triages a ticket. That's what agents need to learn, and it doesn't exist on the public web. So the competition is moving toward reinforcement learning environments. BigGo and RL List report Mercor acquired Deeptune, a startup that builds them, in July. Labelbox announced a Series E of more than $100 million on Oct. 15 aimed at reinforcement learning and enterprise AI agents.

micro1's contribution is the check size and the pitch. It's the first we've seen to say publicly, with a number, that a company's internal documents are an asset it can sell.

What Could Go Wrong

Start with who actually owns what. A company's CRM is full of customers' information. Its support tickets quote users. Its contracts bind vendors to confidentiality. Its employees wrote the documents. Beri's analysis notes that micro1's public terms don't name a downstream buyer or address exclusivity, liability or indemnity.

De-identification is the main safeguard, and it has limits. "Removed when relevant" is a loose standard, and micro1's own page says its anonymization model "has benchmarked as best in class," which is the company's claim. Beri points to a 2019 study in Nature Communications finding that 99.98% of Americans could be correctly re-identified in a dataset using 15 demographic attributes, a reminder that scrubbed records aren't always anonymous. In the Spirit case, The Next Web reports that the ombudsman wanted to review micro1's plan to de-identify the data itself rather than use an independent party. It also reports that if the court turned to Mercor instead, she wanted to look into a security incident at the company earlier this year.

Then there's the awkward one. Employees are unlikely to be paid, or even asked, when their employer sells the record of their workday. That's the fight the flight attendants' union just had, and Nelson said the union would take the issue to its other employers for contract protections.

Our Take

The cleanest way to see this is as a sale of recipes.

A company's SOPs, tickets and decision records are how it does its job. micro1 is offering to pay for those recipes, and the buyers want them to train agents that can follow the recipes without the cooks. Some of the companies selling will be selling the training set for the thing that eventually does part of their work. That might be fine. It might even be smart, a hedge. But it's a real trade, and the buyer gets to decide how the recipe gets used.

The second thing we noticed is the order of events. The most visible sale of this kind so far came out of a bankruptcy, which is the one situation where a company's records are sold without its say. micro1 is trying to build a version where the company says yes. Whether that stays clean depends on the contract terms nobody has published.

What We Cannot Conclude

• We don't know who's buying. micro1 names no downstream buyers, and we haven't seen a seller named.
• We can't say the $1 billion is real money. It may be a commitment, a ceiling or a headline.
• We don't know how much has been paid out. We haven't seen a completed deal reported.
• micro1's valuation, revenue and customers are secondhand. They come from reports relayed by Beri, and micro1 declined to comment.
• The market numbers come from mixed sources. The Mercor figures are relayed by BigGo from The Information, not read at the source.
• The Spirit sale isn't done. We found no court ruling as of Nov. 11 coverage, and no update on micro1's bid.

Why It Matters

• Data is turning into an asset class. The question for companies is shifting from "can we use our data?" to "should we sell it?"
• The scarce input is moving from text to process. Agents learn from how work is done, so workflow data is where the money is going.
• Rights are unsettled. Customers, employees, vendors and software partners all have claims on corporate data.
• Bankruptcy is a preview. Spirit shows how corporate records can become training data without everyone's consent, which is the case regulators will study.