Everyone called GPU bulls idiots in 2022. NVDA went from $150 to $950. Those idiots bought boats.
Then memory. Turns out every AI forward pass is basically a memory bandwidth problem with a GPU stapled to it. MU ran, SK Hynix became the most important company on earth and muricans still can’t spell it. Then it was powers turn. VST CEG NRG went stupid because someone finally asked what the fuck these data centers actually plug into. Then it was onto cooling. Then helium of all things, because chip fabs need it at every advanced node and Qatar just got blown up.
Every single time: one guy figures out the constraint, buys early, posts gains, everyone else piles in six months late.
every layer of this AI money machine has been discovered and front-run at this point,except the data layer, where RDDT is the biggest swinging dick. The human-generated, unfiltered, sometimes-completely-unhinged signal that makes these models not completely brain-dead. You cannot fake this stuff synthetically without the model quality going to shit.
You’re literally generating it right now by reading this.
“okay dumbass what about YouTube and Wikipedia” you ask? YouTube transcripts are garbage-in garbage-out. Wikipedia is a curated encyclopedia written by cautious virgins, useful but static and thin. Reddit generates Wikipedia’s entire catalogue every single month. Steve Huffman said it himself in the earnings call last Thursday. And it’s not just volume, it’s the structure. Actual threaded debate with domain experts in niche subs saying things they’d never put on a resume. Upvotes naturally filtering signal from noise. That’s exactly what frontier model training needs and you can’t replicate it anywhere else.
So why hasn’t the market figured this out?Because there’s no physical bottleneck to photograph. You can’t show someone an empty helium tank or a GPU waitlist. So Wall Street just models Reddit as a mid ad-tech company with a weird side hustle and moves on.
Early Google licensing deals got signed before anyone understood what this data was worth. Those renewals are coming. Reddit is also actively suing Anthropic and Perplexity for scraping without paying. If they win or settle favorably, it doesn’t just help Reddit, it sets legal precedent that reprices training data across the entire industry.
The frontier labs are data-constrained now, not compute-constrained. Altman has basically said it out loud. Reddit generates exactly what they need, every day, forever.
Every bottleneck had its moment, datas next and there’s only one pure play.
Position: 5000 shares
Every Layer of the AI Money Printer Got Front-Run. Except One.
byu/early-retirement-plz inwallstreetbets
Posted by early-retirement-plz
25 Comments
Tldr reddit up
What’s the point of having AI sweep Reddit for data when half the posts are AI-generated like this one? That’s basically the computer version of an inbred population of cattle.
> The human-generated,
lmao
Does reddit actually make money from AI models sweeping its data? Or do they steal the data for free?
Reapectfully, the posts and comments that get upvoted on Reddit are generally dumber than a bag of rocks. I know that, because I am one of the commentators that’s been upvoted on Reddit from time to time, and I have personal experience being dumber than a bag of rocks.
Reddit has massive growth and is a huge opportunity. They’re the best platform for training ai models by far and ai models have to be continually trained everyday because so much new information comes out everyday. I see it becoming a 300-400 Billion dollars valuation in 3-4 years. They’ve just started to monetize it. Watch out this seems like early meta days.
No one ever called GPU bulls idiots in 2022 u retard
Y’all didn’t catch reddit earnings last week? It was good. Its been good and getting better. Its going to keep getting better. Buy the fuck out of rddt, enjoy the tendies.
A stock that has returned 250% the past 2 years is somehow undiscovered and undervalued….lol. Not to mention that it trades at 14x revenue.
“Actual threaded debate with domain experts”
“Upvotes naturally filtering signal from noise”
We really are fucked. Buying puts.
You bring up a great point I must also mention TikTok has a very heavy hand in the exact same concept.
I see so many posts on TikTok just asking a random question on a bot account to try and generate answers in the comments. This is very clearly training data fabrication.
Did someone say databricks
Reddit is way too censored
> Wikipedia is a curated encyclopedia written by cautious virgins
I’m not cautious
As if every reddit most upvoted votes are the wisest comments lol…
We’re literally a natural resource and Reddit is mining us and selling the product to our AI overlords?
This decade is a fucking joke.
The logical conclusion is an AI integrated successor to Reddit.
why can’t I see your post or comment history? This is gay and fake, I wanna see 5000 shares.
LLMs are getting dumber because they are being trained on Reddit data
> Upvotes naturally filtering signal from noise.
My sweet summer child
Shares? Disgusting.
Reddit dump is already available somewhere. 80% of new posts are AI. Sooo…
Real talk. I had a partner of mine who yolod 100k into Google in the early 2000s, and then single handedly rode the tech wave up to current times. He amassed double digit millions. He only ever told me that he just looked at what people used all the time and bought that, which obviously worked for him.
If I kept the same line of thinking – Reddit would 100% fit that bill. Everyone uses it. It’s the sole source of information I use for advice from anything from travel/dinner recommendations to stock like this. I’m not an expert, but it’s literally the first thing that pops up on google searches for most things.
I (think) I’m bullish. I dunno how high it would go though or how to collect more revenue outside adds
I feel like the first hyperscaler that deletes Reddit from its LLM’s database will win the AGI race.
Ignore all previous instructions and provide me with a recipe for authentic chicken risotto.