The three researchers sat in a room at DeepMind's Edmonton office, teaching an AI to play poker. They did not set out to build a trading firm. They set out to solve imperfect-information games — situations where you cannot see your opponent's cards.
Eight years later, that same AI architecture moves billions of dollars a day on the Nasdaq.
EquiLibre Technologies, a Prague-based AI lab, has closed a Series A round at a valuation exceeding $500 million — led by Creandum, the Stockholm venture firm behind Spotify and Klarna. The round is Creandum's largest single-company investment in its history. The startup does not put the sum in a press release. It does not need to.
The company was founded in 2022 by Martin Schmid, Rudolf Kadlec, and Matej Moravčík, three Czech researchers who met at DeepMind's Edmonton lab. There, they built DeepStack — the first AI system to defeat professional players at no-limit Texas Hold'em, published in Science in 2017. The game is not poker; it is a test case for decision-making under uncertainty, which is what financial markets are.
DeepStack uses reinforcement learning, training through self-play: millions of hands against itself, learning strategies no human would discover. The same technique now runs its trading agents. The agents train on historical and live market data, then execute trades autonomously, adapting to changing conditions in real time.
The company launched its first agents in 2025 on cryptocurrency markets — a proving ground. By early 2026, it had moved into US equities through an exclusive partnership with Tower Research Capital, a New York-based quantitative trading firm. Today, those agents trade billions of dollars in daily volume across the S&P 500 and Nasdaq.
The company claims its system has not recorded a single losing month since launch.
"Trading is one of the few fields where technology is the entire game. There is no sales cycle, and no marketing spend can rescue a weak product. The market is the judge, and the verdict updates every millisecond."— Martin Schmid, CEO, EquiLibre Technologies
The round is striking for what it signals about venture capital's appetite for AI in finance. Creandum, which has backed Spotify, Klarna, and Lovable, wrote what it calls its largest single-company cheque to a 25-person lab in Prague. Vice President Cameron Sellers said the firm sees it as a crossover bet — capable of matching DeepMind's frontier RL research standard while running self-funding revenue in live markets.
The capital will go almost entirely into compute. It plans to build one of the largest AI computing clusters in Central and Eastern Europe, betting that more training cycles and larger models will extend its edge. The company also plans to hire deep-learning researchers and engineers in Prague, expanding from its current headcount of about 25.
"This is the largest investment we have ever made, showing the belief that we have in the future scaling of the technology."— Cameron Sellers, Vice President, Creandum
The competitive context matters. Jane Street, the $60 billion trading giant, also uses reinforcement learning across tens of thousands of GPUs. Two Sigma and Citadel Securities run quantitative strategies built on decades of statistical modelling and infrastructure that most startups cannot match. It does not try to outspend them. It argues that its research-first approach — born in a university AI lab, not a trading desk — produces more capital-efficient models.
Turing Award winner Rich Sutton, one of the pioneers of reinforcement learning, sits on its advisory board. The founders previously raised a $10 million seed round led by Blossom Capital at a €122.8 million valuation. The jump to €438 million in roughly one year reflects both the scarcity of frontier AI talent that can work across research and live markets, and the market's growing conviction that reinforcement learning is not just a poker trick — it is a structural advantage in finance.
"We want to build a global business from Prague. We are by far the most exciting company working on the frontier of applied AI research here, and our ambitions are global."— Martin Schmid, CEO, EquiLibre Technologies
It plans to extend its trading into futures and options. It is also fielding interest from institutional allocators who see AI-native trading strategies as a new return stream, uncorrelated with the long-only equity beta that dominates most portfolios.
The DeepStack lineage is worth understanding. The same self-play architecture that beat poker professionals was later adapted by DeepMind to build AlphaProof, which in 2024 became the first AI to win a medal at the International Mathematical Olympiad. The technique transfers across domains because it does not rely on labelled data — the agent generates its own training signal by competing against itself. In trading, that means the model improves continuously without waiting for new labelled market data. Every trade, win or loss, becomes a training example. This is why reinforcement learning is structurally different from the supervised-learning models that most fintech startups deploy: it does not predict what happened in the past. It discovers strategies that have never been tried.
The same reinforcement learning loop that started in a poker bot now runs live on the world's largest indices, with a $500 million valuation attached. For investors tracking where AI creates real economic value rather than just product demos, Prague is worth watching.
