Hey there. Welcome to the very first edition of this manmade newsletter. The goal here is to make it just a bit easier for all of us to understand what the hell is happening in the training data market. Who is buying? What data? Why?

Frontier labs want to turn their models into workers, which requires real-world process, judgment, and outcomes. What a coincidence! Businesses have that data! So the race has begun to collect valuable business data, build benchmarks, and conduct evaluations.

This week Micro1 announced a commitment to spend $1B on business data, with some help from their friends at Citi and Hercules (both very serious institutional lenders) over the next 12 months. Mercor and Micro1 are in a head-to-head competition with who can pay more for business data referrals with reports of up to $100K per successful referral! Expecting at least a $3B debt facility from you soon Mercor 👀.

Rumor mill: most course creators on Whop and the last two batches at YC have all immediately pivoted into becoming business data brokers 😂

On a more serious note, this is an exciting new revenue stream for SMBs, but there are a LOT of outstanding questions on how the agreements are being structured, if data is being sufficiently anonymized, and who exactly is buying this data.

I imagine there’s a real market opportunity for a Business Data Advisor, who is fiduciary to the business, not the lab, and represents them as a talent firm would. The WME of SMBs, if you will.

Training data started with services cash flow and BPO roots (Appen, iMerit, etc.). Then venture capital got involved with labeling software companies (ScaleAI, Labelbox, etc.) and RL environments (Mercor, Micro1, etc.). Finally, we’re seeing the private credit market emerge! Notably, Labelbox has also announced a major financing including debt this week.

Innovative lenders, like Masterpiece Labs, are starting to pop up with creative working capital solutions (factoring, invoice financing, etc.). This is exciting, but it feels like the private credit market is very underdeveloped in this asset class.

Where is our merchant bank for training data?! Gold led to Wells Fargo, railroads to JPM, Hollywood to BoA, crypto to Galaxy. There is a massive market opportunity for a firm to put their balance sheet to work in financing new training datasets!

But wait, isn’t that venture capital firms? Historically, VC firms haven’t been able to stomach the customer concentration of data companies, and I can’t tell you how many times I’ve heard, “but what’s the long-term moat, won’t all data just be simulated?” or “labs will just handle all data internally.”

Turns out, customer concentration is OK when they’re paying enough (MicroAGI got to $100M annualized run rate in just six months!). And it turns out that data collection, RL environments, benchmarks, and evals are very good cash-flowing businesses (Micro1 crossed $1B in three years). Oh, and we’ve also learned that more compute doesn’t help if the data is bad. Therefore, quality data is becoming the real bottleneck to hill climbing. Venture capital funds are now stampeding into data lab financings. The frenzy has only just begun.

But the plot thickens! Traditional venture capital financing isn’t very attractive to Founders building data companies. There’s less market risk and technical risk to underwrite. Closing just one deal can catapult the company to typical Series A or B revenue targets. I’m starting to see investors discuss new models, but it’s shocking to me how few venture funds understand the training data value chain and are positioning themselves to be helpful.

My prediction is you’ll see a bifurcation between venture funds either investing at the earliest stage and helping their companies navigate novel working capital options OR doing very late-stage deals to create liquidity for teams.

Thanks for reading! You can reply directly to this newsletter with what I got wrong, what you want to see next week, or any other dank training data memes. See you next week 🫡