The Outsized Bet on Prague's Poker Minds
Most venture capitalists spend their weeks chasing the same dozen Stanford dropouts building wrapper apps in San Francisco. Creandum decided to look elsewhere. Specifically, they looked to Prague, writing their largest single ticket ever to back a 25-person team of reinforcement learning purists.
The company is EquiLibre Technologies. The valuation is a stunning $500 million.
For a Series A, that is a staggering number in this market. But if you look at the pedigree and the capital efficiency, the math changes. The team previously grabbed a $10 million seed round at a $140 million valuation led by Blossom Capital. Now, they are scaling. They aren't trying to build another chatbot. They are building models that trade financial markets, and they are doing it with algorithms originally designed to take professional card players for everything they own.
Creandum’s lead position here tells you everything you need to know about the current state of AI venture investing. The low-hanging fruit of generative text is gone. Corporate buyers are pushing back on expensive SaaS seats that don't produce measurable ROI. But a machine that prints money by outsmarting human traders? That is a business model with zero friction.
From Edmonton to the Felt
Before EquiLibre was a half-billion-dollar enterprise, its founders were visiting PhD students sitting in DeepMind’s Edmonton, Alberta office—a satellite lab that Alphabet unceremoniously shut down in 2023 during its global efficiency drive. That cutback looks like an expensive mistake in hindsight.
At the center of the startup are three researchers: CEO Martin Schmid, CTO Rudolf Kadlec, and CSO Matej Moravcik.
While at DeepMind, this trio built DeepStack. For those who don't follow the history of game theory, DeepStack was the first AI program to consistently beat human professionals at no-limit Texas hold’em. It was a massive breakthrough because, unlike chess or Go, poker is a game of imperfect information. You don't see your opponent's cards. You have to calculate probabilities, sniff out bluffs, and make decisions under extreme uncertainty.
They aren't working alone in a vacuum. Rich Sutton, the godfather of modern reinforcement learning and the 2024 Turing Award winner, is actively advising them. When Sutton attaches his name to an RL project, the research community notices. It gives EquiLibre a level of technical credibility that practically eliminates the usual diligence hurdles VCs face when investing in complex algorithmic trading systems.
Why Poker Models Hunt Best in Financial Markets
To a layman, shifting from Texas hold’em to the S&P 500 seems like a leap. To a reinforcement learning researcher, it is the exact same game.
The underlying mathematics of decision-making under uncertainty don't change because you swapped playing cards for exchange-traded securities. In both environments, you have incomplete information, moving targets, and a noisy environment. Most importantly, both systems offer a clean, brutal, and unambiguous reward signal. Profit or loss. Up or down.
Enterprise AI startups struggle because defining "success" for a marketing copy generator is highly subjective. It requires human-in-the-loop feedback and endless tuning. In quantitative finance, the feedback loop is instantaneous. The system places a trade, the market moves, and the P&L statement updates.
EquiLibre designs its models to treat the financial system as a massively multiplayer game of imperfect information. The algorithms don't read news articles or scan Twitter sentiment. They observe market microstructure, manage risk, and learn how to extract margin from less sophisticated players. It is pure reinforcement learning applied to capital allocation.
The Flawless Trading Sheet
Plausibly sounding slide decks don't survive contact with Wall Street. EquiLibre proved its algorithms by partnering with Tower Research Capital, one of the premier proprietary trading firms in the world.
They deployed their reinforcement learning models across the S&P 500, the NASDAQ, and various cryptocurrency exchanges.
The results are hard to ignore. According to their disclosures, the startup has recorded zero negative months since inception. Every single month of operations has yielded a net positive return. That is a track record that seasoned hedge fund managers would give their left arm to replicate.
Of course, we have to look closely at these claims. "Zero negative months" does not mean they never lose a trade. It means that across their portfolio of markets, the gains systematically outweighed the losses over every thirty-day window. In cryptocurrency, where volatility is extreme and retail traders frequently act on emotion, EquiLibre’s models found a highly lucrative playground. Now, they are taking those learnings and pushing deeper into traditional, highly liquid equity markets.
The Prague Talent Moat
If you build an AI company in San Francisco, you are constantly fighting a war of attrition. Your lead engineer will get recruited by OpenAI, your product manager will leave for a stealth startup, and your rent will eat up a third of your runway.
EquiLibre set up shop in Prague. It was a deliberate strategy. Martin Schmid and his co-founders realized they could build a team of 25 elite engineers, largely built from former DeepMind and Google colleagues, without dealing with the Bay Area circus.
In Prague, talent retention is exceptionally high. Engineers join to solve complex mathematical problems, not to flip stock options in eighteen months. This geographic isolation creates a highly focused, collaborative culture that is incredibly hard for Western competitors to disrupt.
Furthermore, the cost of top-tier talent in Central Europe is a fraction of what you pay in California. EquiLibre’s $10 million seed round stretched much further than it ever could have in Silicon Valley. It allowed them to focus on research and infrastructure development rather than constantly fundraising to keep the lights on.
Winning by Compute Efficiency
The popular narrative in AI right now is that the firm with the most GPUs wins. Jane Street, one of EquiLibre's primary competitors, operates tens of thousands of high-end GPUs to run a mix of reinforcement learning and large language models. They are brute-forcing the market.
EquiLibre is taking the opposite path. They are obsessed with compute efficiency.
You do not need to burn megawatts of power if your underlying algorithms are mathematically elegant. The DeepStack team knows how to optimize models to run on constrained hardware. After all, the original DeepStack AI ran on a single GPU while beating players who had spent their entire lives perfecting their game.
By focusing on compute efficiency, EquiLibre is building a business model that is highly scalable. They aren't spending all their venture capital paying cloud providers for compute time. Instead, they are building one of the largest specialized computer clusters in Central and Eastern Europe. This local infrastructure will give them the physical capacity to train much larger models without being beholden to the pricing whims of external cloud vendors.
The Lab That Shuns the Finance Label
If you ask Martin Schmid what EquiLibre is, he will tell you they are "a lab first, not a finance firm." This is more than just clever marketing. It is a philosophy that directly affects how they recruit and build.
Finance firms are notoriously short-sighted. They care about the next quarter’s P&L, the next bonus cycle, and the immediate payout. Labs, on the other hand, care about fundamental scientific breakthroughs. By positioning EquiLibre as an AI research lab that happens to fund itself through trading, the founders can attract scientists who would otherwise refuse to work on Wall Street.
This structural setup creates a virtuous cycle. The trading revenues fund the basic scientific research. The scientific research improves the trading models. It is a self-sustaining engine.
As they expand from crypto to traditional liquid equities, this research-first mindset will be tested. Making money in crypto is relatively simple compared to the highly efficient, heavily front-run world of U.S. equities. But if their reinforcement learning models have truly cracked the code of decision-making under imperfect information, the $500 million valuation Creandum just stamped on them might look like a bargain in two years. Keep an eye on Prague. The future of quantitative finance is being written in Czech crown territory.