When Spirit Airlines went bankrupt this year, Google paid $10 million for about 600 million of its employees' work emails and Teams chats. The people who wrote those messages didn't get a dime.
Hold onto that, because it explains why the startup in this post is offering its VP of semiconductors up to $9 million a year.
Grepped is a data broker. A company hands over its internal records, like its Salesforce data and accounting books, and Grepped cleans it all up and sells it to AI labs. Erik says he was at a $4 million monthly run rate about two weeks after he started.
The math on this job post is kind of wild. The hours are 9am to 9pm, seven days a week, so 84 hours. At the top of the range the VP makes roughly $2,000 an hour, and the founding scientist on the same list makes about $70.
Why would anyone pay a dealmaker 10 to 30 times what they pay a scientist? Look at the question he wants applicants to answer. "What does humanity know that it doesn't know it knows?"
That's basically his whole pitch in one sentence. AI labs already trained on the entire public internet. What they haven't seen is sitting inside companies, like the Slack thread where a senior engineer explained why the last fix broke production, or the ledger where an accountant made a judgment call and never wrote down why. People did that work for decades without ever thinking of it as knowledge. For a model it's gold, because it shows how experts actually make decisions when it counts.
Spirit put a price on it. Google beat Mercor's $7.5 million bid, and $10 million spread across 600 million messages comes out to less than 2 cents each. The money went to Spirit's creditors. The flight attendants' union went to court over the sale and pointed out the airline was selling their messages while it still owed them vacation pay.
Chips are where this gets really expensive. A chip company's design history might be the most protected IP in the world. You can't scrape it off the web, and the expert marketplaces paying engineers $85 to $200 an hour can only rebuild little pieces of it. Somebody has to walk into a chip company and talk them into selling 20 years of design files and failed chips. If an airline's chat logs went for $10 million, that archive goes for a lot more, and a broker only needs its cut on a couple of those deals to cover a $9 million salary.
So if you've spent the last ten years typing into Slack at your job, you've been building one of the most valuable datasets in AI, and whoever gets it sold is the one making $2,000 an hour.
we are hiring!
9am to 9pm, 7 days a week, in SF.
vice president, semiconductors: $4m–9m a year + equity
neural data lead (bci): $1.5m a year + equity
chip design data lead (semiconductors): $1.5m a year + equity
founding scientist, autonomous lab (sf): $300k–400k + equity
member of technical staff (sf): $340k–450k + equity
to apply, answer:
"what does humanity know that it doesn't know it knows?"
send your answer, resume + things your proud of to a@grepped.ai