EquiLibre Technologies, the Prague startup built on the same AI that cracked no-limit Texas hold'em, has raised a Series A at a $500 million valuation, with Creandum leading what the firm calls its largest single check ever written.
The founders didn't start out thinking about markets. Martin Schmid, Matej Moravcik, and Rudolf Kadlec were visiting PhD students at Google DeepMind's Edmonton research office when they co-authored DeepStack in 2017, the first AI to beat professional players at no-limit Texas hold'em. The system worked by solving incomplete information in real time, running deep reinforcement learning across billions of simulated poker hands to develop a strategy that accounted for bluffing, deception, and multi-round uncertainty. It didn't just win. It dismantled the notion that imperfect information games were permanently out of reach for machines.
Poker and markets, it turns out, share a deeper structure than they appear to. Both punish rigid rules. Both reward agents who can reason under uncertainty, adapt to an adversarial opponent, and make sequential decisions with incomplete information where mistakes compound across rounds. That insight is the intellectual foundation of what EquiLibre is now building. According to TechCrunch's reporting on the funding, the startup's algorithms have been trading billions in daily volume across the S&P 500 and Nasdaq, and the company claims a zero negative months record since inception, first on crypto markets starting in 2025, then on equities.
The seed round, led by Blossom Capital, priced EquiLibre at $140 million. The jump to $500 million on the Series A is steep, and Creandum's Cameron Sellers confirmed the round size was the firm's largest single investment ever, though the exact dollar figure wasn't disclosed. That kind of capital commitment from a European VC into a quant infrastructure play is unusual, and it reflects something broader: the appetite for AI-native quant shops has become serious money.
Traditional quantitative finance is not short on compute or talent. Renaissance Technologies built Medallion into arguably the greatest trading record in history using statistical arbitrage, pattern recognition, and a culture of secrecy that makes the NSA look chatty. Two Sigma and Citadel have each spent a decade hiring machine learning researchers from academia. And yet, the specific flavor of reinforcement learning that DeepStack pioneered, training an agent by letting it play itself millions of times in adversarial conditions rather than fitting it to historical data, has not been the dominant approach in quant finance. Most ML in trading is still supervised: train on past price sequences, predict future returns, repeat. Reinforcement learning is fundamentally different. The agent learns by acting, not by fitting.
That distinction matters most in exactly the market conditions where traditional quant models struggle: regime changes, thin liquidity, adversarial price discovery where other participants are actively working against you. A poker agent trained purely on historical hand histories would be mediocre. An agent trained by playing billions of hands against itself, building a strategy that holds up against an opponent who is always adapting, is something else. EquiLibre is betting that same gap exists in markets, and its trading record, if the company's own figures hold up to scrutiny, suggests the bet is paying off.
The Tower Research partnership announced in 2024 was the first public signal that the technology had cleared institutional standards. Tower is not a charity for interesting research projects. It's one of the largest high-frequency market makers in the world, sourcing signals from independent teams who succeed or fail on actual returns. EquiLibre contributing exclusively to Tower through that arrangement was, frankly, the most credible validation the startup could have received short of publishing audited returns.
Creandum's check takes that validation and turns it into a growth mandate. The firm's history skews toward consumer software and developer tools, which makes this a significant departure, and the largest single investment number Sellers cited suggests it's not a toe in the water. Whether EquiLibre expands its exclusive Tower arrangement, brings on additional hedge fund clients, or begins building its own fund structure will define the next chapter. What's already clear is that three researchers who spent years teaching a machine to read a poker table have done something most of the quant world has been trying to do: make reinforcement learning work at scale in live markets, not just in research papers.
Also read: Etched bets $800 million that transformer silicon will outlast the GPU era • Bitcoin ETFs just posted their worst month on record and the buyers stepping in are not who you think • Bending Spoons prices its Nasdaq IPO above range as Wall Street bets on AI-powered software roll-ups
