Wall Street is quietly tuning into Nof1, an AI trading lab backed by SUI Group (SUIG) and Karatage. The startup’s flagship project, Alpha Arena, puts frontier AI models from OpenAI, Anthropic, Google and xAI head-to-head in autonomous trading competitions using real capital.
Early results are messy. Most models lost money, overtraded or struggled with risk management. Yet for sophisticated investors, that failure may be the signal. AI has conquered language over the past two years; trading may be the next frontier, and a far harder one. Financial markets reward consistent capital allocation under uncertainty, not just sounding intelligent.
Alpha Arena: Real Capital, Real Losses
Nof1’s thesis is that today’s frontier models were trained to predict language, not allocate capital. Alpha Arena bridges that gap by exposing models to live financial environments where they are punished by actual market outcomes. SUI Group, a Nasdaq-listed company building Sui blockchain exposure, and Karatage, a London-based prop hedge fund focused on digital assets, AI and gaming, co-led Nof1’s $15 million funding round. They also backed Recursive Superintelligence alongside TwinPath Ventures—a self-improving AI startup reportedly valued above $4 billion.
A Coordinated Bet on Autonomous Agents
The investments look less like isolated venture bets and more like a coordinated thesis. Both Nof1 and Recursive are grounded in Open-Endedness Research. The wager: if autonomous agents eventually become core to investing, the firms building the infrastructure and training environments today could shape finance’s future. Nof1 plans to launch a consumer platform after season two, calling it the world’s first coding agent for markets, built on a self-improving AI architecture designed to eclipse human and algorithmic traders.