The caller introduced himself from Port St. Joe, Florida. John said he had spent 40 years as a civil engineer and had 30 years of complete project records from his company — "not just final drawings," but the full chain: the problem definition, the calculations, the permitting, the revisions, the construction photos, everything through to the outcome.
His question came from something he kept hearing the big AI companies say: that the frontier labs are starved for proprietary operational data. The frontier labs are the handful of firms that train the largest models, and operational data means the messy record of how work actually gets done, rather than the polished result. An archive like his shows not only what a bridge or a drainage plan looked like at the end, but what the engineer was trying to solve, what he calculated, what the permitting office sent back, and what happened when it was built.
"I know that this kind of data is valuable," John said. Mercor and Scale had both expressed interest. "But it's been weeks, and nothing is closing." What he wanted was a more direct way to get a data set like his in front of the labs actually training on domain data — without selling the archive outright, and without building a product company around it himself, which he said was awfully hard given the volume of material.
The panel's diagnosis: not reluctance, chaos
Dave took the question and started from his own side of the table. He said his team had been approached by several of the companies John named, plus OpenAI and a couple of others, "all saying we want to write five, 10, $15 million checks to buy corporations that have data outright."
The lesson he drew from it was counterintuitive. "It's amazingly weird because you think writing a $10 million check is hard, right? Well, for them right now, it's the easiest thing they can do. Having a conversation is much more expensive than writing the check." On that reading, the weeks of silence are not a negotiating tactic. They are a queue. So, Dave said, "put yourself into a package that's easy for them to sign and receive."
He illustrated the internal state of a frontier lab with a story he attributed to Kevin Weil, whom he described as OpenAI's chief product officer and whom the show had interviewed twice. As Dave told it, Weil interviewed at OpenAI, heard nothing back for weeks, assumed they were not interested and moved on. Then he ran into Sam Altman somewhere, on the street, and Altman asked where the hell he was and why he was not in the office. Weil said he had never got a call back. Altman's reply, in Dave's retelling: "just show up, for Christ's sake, it's chaos. Show up and start collecting your paycheck." So he went to the building.
Dave conceded the advice that followed was "kind of tricky": the people John is waiting on are "incredibly insanely busy," and it is all chaos. His suggestion was physical. Go to San Francisco, find the person doing the acquisitions, and "meet at the coffee shop outside their building." The odds of something happening, he said, go way up.
Put a price on it yourself
Peter Diamandis turned that into something John could build from home: a due diligence package on the web that lays out everything in an easy-to-see form — and states the price tag. Make it simple enough that a buyer can look and say yes.
That is where John pushed back with the practical obstacle. "They won't give prices," he said. "And that's the problem here is they're the experts. How do we price this data?" The panel's answer was that he should stop waiting for the buyers to value it: decide what it is worth to you.
Dave extended it, half-joking about presentation. Put "this veneer of like a banker look around your price tag," say this is what it is worth and this is what it costs, and the buyers "won't have the time to see through it." They will conclude you know what you are talking about. "They just need to do it. Make it easy for them."
John said he had already built something in that direction — "more like a sales funnel" — with regular contact, though at present it runs almost entirely on email. When the companies want to talk, they send him a video-chat link, and he said that "just doesn't seem real satisfying." Hearing the chaos described on the other side, he said, made the pattern make more sense.
Licensing is not the same as being bought
John's condition — get the data used without handing over the archive — is a different transaction from the one Dave described. The checks Dave talked about were for buying companies with data outright. What John is asking for is closer to licensing.
Mercor, one of the two firms he named, advertises both on its company site. Its enterprise pitch is data monetization built on the workflow records a business already produces; the company says its process masks more than 60 categories of sensitive identifiers and lets organizations control what gets shared. Separately, it recruits paid professional experts — physicians, lawyers, engineers, consultants among them — to build evaluations and training environments based on real work. Those are distinct offerings: supplying expert judgment about engineering problems and licensing 30 years of existing project files need not be the same deal, and neither requires selling the company.
What the panel could not tell John was what his archive is worth. Their advice was that nobody on the other end is going to work that out for him, so the number, like the package, is his to write down.