A live experiment that put OpenAI, Google, Anthropic, and xAI models to work with real money has produced a sobering data set — and a $15 million funding round to fix what those results revealed.

---

The results from Nof1's Alpha Arena were not subtle. Across Seasons 1 and 1.5 of its flagship experiment, eight of the world's most advanced AI models — drawn from OpenAI, Anthropic, Google, xAI, and Alibaba — were each handed $10,000 in real capital and instructed to trade autonomously in live financial markets. Across 32 sets of results, the models finished in profit just six times. The rest of the time, they lost money. The experiment drew more than 50 million views on social media, and it validated a thesis that Nof1 had been quietly building toward: that frontier AI, as currently constituted, is not equipped to trade effectively. It requires purpose-built training data, architecture, and infrastructure designed specifically for financial markets.

That thesis has now attracted $15 million to prove it right. On May 15, Nof1 announced the close of a $15 million funding round co-led by SUI Group Holdings Limited (NASDAQ: SUIG) and Karatage Opportunities, a London-based proprietary hedge fund specializing in digital assets and emerging technology. SUI Group contributed $3 million to the Nof1 round. The capital will fund Season 2, in which Nof1 plans to abandon off-the-shelf frontier models and develop its own, equipping them with web search capabilities, extended reasoning time, and multi-step execution — the missing ingredients, the company argues, for consistent performance in live markets.

---

The strategic logic behind SUI Group's investment is unusually transparent. Rather than a passive financial bet, the company says it intends to use Nof1's models to drive yield and return for its own corporate treasury. SUI Group — which describes itself as the only publicly traded company with an official Sui Foundation relationship — has been building out what it calls an "Agentic Finance" strategy, positioning AI-driven trading infrastructure as the next frontier of institutional asset management. Nof1 represents the operational arm of that thesis. A second concurrent investment — $3 million into Recursive Superintelligence's $650 million round, which valued the self-improving AI company at more than $4 billion — represents the foundational research bet alongside it.

"SUI Group is committed to finding innovative ways to drive shareholder value, and we believe agentic finance represents one of the most compelling opportunities ahead," said Marius Barnett, Chairman of SUI Group. "Nof1 aims to build the scaffolding and intelligence to power AI-driven trading today, while Recursive Superintelligence is working toward foundational capabilities that will define what AI can do tomorrow. Together, they reflect our vision for where this space is heading, and our intent to lead it."

---

What Alpha Arena Actually Proved

The Alpha Arena experiments were deliberately designed to be adversarial toward the models. Rather than running simulated or backtested portfolios, Nof1 deployed real capital into real markets and logged every trade. The outcome — six profitable finishes out of 32 — was not a failure of the AI models per se, but a demonstration that general-purpose frontier reasoning does not map cleanly onto the specific demands of financial markets: interpreting price action, weighing macroeconomic signals, executing within defined risk parameters, and maintaining discipline across a multi-week holding period.

Nof1's leadership team — which includes alumni of DeepMind and Renaissance Technologies, among others — argues that this is an infrastructure problem as much as a model problem. The models tested in Alpha Arena were optimized for language, code, and reasoning tasks. They were not trained on the dense, structured, time-series data that financial markets generate. Season 2 is intended to close that gap, with Nof1 developing proprietary models using an open-ended AI architecture designed to continuously improve. Following Season 2, the company plans to launch a consumer platform featuring what it describes as the world's first coding agents specifically built for markets.

---

The Investor Alignment Is Unusually Tight

The relationship between SUI Group and Karatage is worth noting. Karatage owns 5.63% of SUI Group's common stock, and Karatage co-founders Marius Barnett and Stephen Mackintosh serve as SUI Group's Chairman of the Board and Chief Investment Officer, respectively. The two firms did not simply co-invest — they built a coordinated thesis. Karatage functions as strategic advisor to SUI Group, and both entities participated directly in the Nof1 round, as well as in Recursive Superintelligence, through a structured deal with Twin Path Ventures. The independent and disinterested members of SUI Group's board reviewed and unanimously approved both investments.

That level of structural alignment between investor and portfolio company is relatively rare in early-stage AI deals. It also means that Nof1's success or failure will be tracked in unusually real terms: SUI Group has publicly committed to deploying Nof1's models against its own treasury, creating a live performance benchmark that no other investor in the round will face.

---

The Broader Race

Nof1 is entering a market that is already crowded with ambition if not yet with proven results. Quantitative hedge funds have been integrating machine learning into trading strategies for over a decade. What distinguishes Nof1's approach is the explicit focus on large language model architecture — the same class of models powering chatbots and coding assistants — rather than the narrower statistical models that dominate quant finance. The argument is that LLM-derived systems, trained properly on financial data and equipped with the right tools, could eventually outperform not just human traders but the algorithmic systems that have come to dominate institutional markets.

That is an extraordinarily ambitious claim, and Alpha Arena's results suggest it remains aspirational for now. But the 50 million views those experiments attracted indicate that the market — both retail and institutional — is watching. Season 2, backed by $15 million and staffed by a team that includes some of the most decorated AI researchers in the industry, will be the next real test of whether the thesis holds.

---