Fundamental raises $275M for a foundation model for tabular data and is using the World Cup to benchmark it against sports books

high such an impactful company. So good. Well, yeah.

Spain was locked in doom scrolling.

He invented the Spanish group doom scroll. No, I'm sure it was uh it was probably messaging driven, probably uh more post-driven, but that is Yeah, that is remarkable. Uh anyway, we have our next guest, Jeremy Frank, the fundamental co-founder, CEO in the way. Let's bring him in to the team.

What's going on?

How you doing, Jeremy?

Good to be here.

You guys, thank you for having me on.

Uh I want to get straight into World Cup predictions, wagers, all of this. But first, introduce yourself and the company a little bit. What are you working on? and then let's go through uh the the broader news and applications.

Awesome. Yeah, I'm Jeremy. I'm the CEO and co-founder of Fundamental. Uh Fundamental built a foundation model for tabular data. So when most people think about the AI boom, they think about LLMs, you know, like CHP cloud. But what's less appreciated is that LLMs work well with unstructured data. So think about text, audio, video, but the global economy um and enterprises really run on structural data on tables such as you know spreadsheets, databases um CRM and that's you know the part of the enterprise that really never had a chip moment and that's a market opportunity that we're focusing on and so what we're doing is for tables is what chipd language we pre-trained the model on millions of tables to understand how the world works um uh in rows and columns. And so we created one model that generalizes across industries um and use cases to um allow companies to make better predictions and decisions on their data.

Yep. Okay. So everyone is going to say, "Oh, the frontier models like they can spin up a spreadsheet. They can speak CSV. They can write Python that generates and manages tables and tabular data." Uh and the frontier is the frontier. It's always the best. But you're going to effectively be benchmarking your product against the frontier models around the World Cup. Walk us through the the strategy, the wager, what's actually at stake there.

Yeah, so you know, I grew up in Belgium and I'm a huge soccer fan and a lot of us at the company are also big soccer fans. And so what we realized is that um you know, historical team performances and games can be structured in rows and columns. And so we started thinking around the office like can we actually use our model Nexus to make predictions on the World Cup and we said like you know maybe betting markets are off and we are right. So like we wanted to see how good our model was and the funny thing is at the core our model you know our company is focused on enterprises. So we're not really a consumer focused company but we you know thought we could have some fun with it. And so before doing anything we wanted to see how good our model was um historically. And so what we did is we train a model to uh with data up until the 2022 workshop. And so data such as uh which who won uh in different international games, the quality of the teams, whether games were home or abroad. And we ran more than 20,000 simulations. And what we what our model Nexus predicted or identified is that Argentina would be the eventual winner way before book markers did. Mhm.

So if you look at what the market actually predicted, the market predicted that Brazil was going to win the 2022 World Cup up until the quarterfinals when Brazil got knocked out. And our model was much more accurate than that. And so we said, you know what, let's apply the same thing for the 2026 World Cup. And our model um already differs meaningfully from what betting markets are predicting. So I'm excited to see next win, right?

Yeah. I wish America America is America has a 0.5% chance of winning the World Cup according to a chance.

I like those odds. I'll take those odds any day.

John loves betting on himself.

Yes.

Uh, how's Belgium fair?

Belgium. Belgium is at slightly less than 2% but I honestly have I have faith.

Okay. And then

Belgium.

Yeah.

How what's the population of Belgium roughly?

11 million. How many? 11 million.

11 million.

So population of 11 million absolutely mogging population.

So So are you also benchmarking the other frontier models just going to the different models, open source models, and just asking them the same questions? Who do you predict will win? Build a probability table and then benchmark those as well. Because I think you're going to want to not just say that you're better than the sports books, but you're also going to want to say you're better than the frontier models. Correct.

Yeah. But the frontier models themselves don't really work well with like predictions, right? So like

but they could take a I mean they can predict the next token. So they can like take a wild guess then if you smoke them it looks good, right?

Yeah. Yeah. We we could try that. We've not done that. We could definitely try it. But like you know it's funny because like the frontier models like especially for predictions they sound right

but they are not actually correct. So

yeah and and and with chat it has the personalization and in the personalization I said like never bet against America so I'm sure it would just say like 100%. So yeah you probably smoke it.

Um who who are like how are you who are you projecting to to win? What what are kind of like the kind of key predictions that uh that we should be tracking as a World Cup kicks off? So the betting markets have Spain and France neck and neck and we uh heavily favor Spain over France.

The market also predicts that uh Argentina I think has like the fifth largest chance of winning according to the market and we predict them as strong runups.

Oh interesting. Okay. So Argentina Spain winning would look really good for you relative to the sports book.

Correct. Would be but I would be happy if Messi win his second World Cup.

Sure. Um

uh tell us about the company. You you recently raised some money.

Yeah, so we raised uh $275 million uh 1.5.

That's fantastic.

Who did you raise it from?

We raised it from um from Oak Valor, Battery Ventures, uh Salesforce and then um a bunch of big angels like Alvin Sinvas from Perplexity. Oh yeah. as rap from Whiz

and a few others.

That's very strong. Uh keep us posted as the the World Cup progresses. Uh and uh we're rooting. I'm not even rooting for any team. I'm just rooting for your predictions. I want you to know that.

Yeah,

I appreciate it. And everyone can go and see them on soccer.fundamental.te.

Amazing.

Soccer.funddamental.te. Very cool. Great to meet you, Jeremy. It'll be fun to follow along.

We'll talk to you soon. Have a good rest of your day. Uh, in other fundraising news, Thrive is out with $10 billion. Uh,

got a little mogged by Bezos with the 12 billion raise, but not bad.

Not bad.

Thrive 10 and it's 10 billion, exceeding 10 billion, the largest ever. Nearly double the size of the last fund for Thrive. Uh, 1 billion for early stage, nine billion for growth. uh biggest bats, just to remind you, OpenAI, Stripe, SpaceX, datab bricks, and cursor are all in the Thrive portfolio. Um and a bunch of other companies uh and so those have become the defining private companies of the current cycle. Uh the strategies concentrated founder loyal allergic to indexing the market. The fifth avenue strategy as discussed on invest like the best and Patrick Oski is ti is chiming in. Uh Josh Krishna says in markets to go long on something is to bet it grows more valuable over time. Much of the conversation today is short on humans wagering that AI makes people redundant. We believe the opposite is true for industries Thrive Holdings operates in. And Patrick Oshanosy shares a transcript from invest like the best. They said the north star for the next decade of Thrive is if we are a company and we have a product what are the products that we can create that make us look more like the businesses that we ultimately invest in. The second part is the story of the story is after we invested in OpenAI the business had $50 million in API revenue and several and several of us went around New York City and started pitching a bunch of the private equity firms on this idea that they could use the API to create greater efficiency within their businesses. We were somewhat surprised by the fact that we did not get meaningful traction around these ideas. So Kareem, to his credit, said, "Let's just start doing this ourselves." We've always looked at the world that we've lived in as one that needed to disrupt from an outside in perspective. If you were to think about the paradigm shift of reinforcement learning, the things that you need are both data and that is proprietary to the company, but also the experts that exist at the company to fine-tune the model. We believe that going forward disruption will happen from the inside out. We set up this permanent capital vehicle that enables us to buy these businesses and hold them in perpetuity because if you actually have a differentiated unique lens and cost of capital around these businesses and you're able to transform them in ways in which you ultimately want to hold on to them forever. So congrats to Thrive on the new round.

Uh congrats to the whole team.

Yeah, fantastic. The new fund. Anyway, leave us five stars on Apple Podcast and Spotify. Sign up for our newsletter at tvpa.com and we will see you tomorrow for the SpaceX IPO.

We're going to just be hanging out. We're going to have the stock price on the big board.

Be fun. We'll see you tomorrow. Goodbye. Throwing flashbang.