Factor Capital Update - November 2025

It can be helpful at times when projecting future progress to look at examples today that may be a microcosm of future adoption within a niche application. This month the example that caught my attention was an automated trading competition which I think foreshadows how AI and crypto are likely to complement eachother over the coming months and years.

The Alpha Arena

An intriguing experiment unfolded in late October that reveals something profound about where we are with AI capability and crypto’s relationship with that technology. Nof1's Alpha Arena competition gave six leading LLMs (OpenAI's GPT-5, Google Gemini 2.5 Pro, Anthropic's Claude Sonnet 4.5, xAI's Grok 4, DeepSeek v3.1, and Alibaba's Qwen3-Max) each $10,000 and access to trade perpetual futures on Hyperliquid's exchange. With identical prompts supplying market data updates every 2-3 minutes and a simple objective of maximizing PnL, the models autonomously managed positions, sized trades, selected leverage, and executed risk management across Bitcoin, Ethereum, Solana, BNB, Dogecoin, and XRP.

The results are striking not necessarily because performance was exceptional, but for what this demonstrates about accessible intelligence and programmable money. The competition ran through November 4th, and the final results tell an interesting story:

Qwen3-Max outperformed all other models and BTC over the trial. The other models either performed poorly from the start (mostly the US models) or got crushed when the market collapsed over the past week, with a 30% drawdown wiping out any gains they had previously accrued.

Performance not withstanding, what these general-purpose models accomplished with zero specialized training or traditional CIO guardrails would have required years of painstaking work to build inferior specialized systems just a few years ago. Today, anyone with API access can deploy this level of capability in minutes.

As an aside, with just one semester of intro computer science experience from 25 years ago, I was able to generate through vibe coding an exact replica of this Nof1 trading experiment with the assistance of Cursor and Claude Code, and effectively test firsthand how one can not only build but work with these models through prompt design to improve and shape their performance for these trading simulations. As a pure educational exercise and demonstration of the enablement that those coding tools offer, it was frankly amazing.

Critics might argue, "But the performance is pretty bad." I think this misses the point entirely.

The Exponential Progress We Take for Granted

The real insight from Alpha Arena isn't the trading outcomes. It's what this reveals about the coming convergence of crypto and AI. These general-purpose models process quantitative data, reason about market dynamics, manage leveraged positions, and adapt to changing conditions autonomously in live markets. None were fine-tuned for trading. A few years ago, building even a mediocre algorithmic trading system required teams working for months. Today, anyone can deploy six frontier models and watch them develop distinct trading personalities within hours.

It's the same pattern we've seen with every transformative technology through hedonic adaptation that the (now-cancelled) Louis CK incredibly articulated 10 years ago.

When airplane WiFi launched, the miracle was that you could check email at 35,000 feet. Quickly people complained their shows were being interrupted mid-stream. Amazon Prime's two-day shipping seemed impossibly fast… until same-day delivery made it feel unacceptable. The Alpha Arena experiment reveals we're already at this inflection point with AI: the models accomplish things that would have seemed impossible a few years ago, yet we're immediately focused on where they fall short. But that's sort of how progress works. The shortcomings we identify define the next frontier of improvement and capability.

Programmable Money Meets Programmable Intelligence

The Alpha Arena experiment works today because crypto infrastructure was built for autonomous, permissionless interaction. But broader agentic commerce adoption requires infrastructure that doesn't exist in traditional finance. That's changing rapidly through stablecoin adoption.

Over the next couple of years, businesses across the spectrum (from coffee shops to enterprise software companies) will receive stablecoin bank accounts and begin accepting stablecoin payments as standard operations. When that happens, two dynamics converge.

First, programmable money means AI agents can transact directly with merchants through instant, final settlement with transparent costs. The infrastructure barrier dissolves.

Second, businesses gain access to an entirely new financial stack: tokenization becomes trivial, and treasury management tools that were exclusive to sophisticated firms become available to any business on these rails.

Portfolio companies like Plural are demonstrating this today and are on the cusp of some very cool announcements over the next few weeks that will illustrate this more clearly.

This convergence of AI's amazing capabilities and crypto's unique programmability makes the next wave inevitable. Models with remarkable and rapidly improving capabilities meet infrastructure being deployed that naturally supports autonomous agents with capital. The combination unlocks applications we're just beginning to imagine.

An Uncomfortable Parallel

There's a second convergence between crypto and AI worth noting, though it's less exciting. The qualities that characterized crypto venture markets in 2021 and 2022 (accelerated deal timelines, extreme valuations, minimal diligence, speculative excitement overriding fundamentals) are showing up in AI funding markets today. The same investors who mocked crypto's excesses are now actively participating in similar dynamics around AI.

Stories are emerging about teams raising substantial seed rounds and effectively walking away with the capital, or robotics companies garnering interest despite presenting what appears more like vaporware than viable near-term products. The parallels to late-stage crypto funding are hard to ignore. That cycle ended violently in 2022 and 2023, with widespread markdowns and collapsed projects and market confidence - just like the dot-com crash and the housing bubbles before it.

It's unclear whether these examples from the AI market are isolated anecdotes or systemic warning signs. The funding environment around AI is certainly overheated. People are calling into question things like OpenAI's commitments of more than $1 trillion in capital on less than $20 billion in revenue (with expenses far exceeding that) and the circularity of many of the compute deals from Nvidia, AMD, and others we've seen announced over the past month.

When excitement concentrates around transformative technology, capital tends to flow indiscriminately. Bad actors capitalize on hype. Valuations detach from fundamentals. A correction eventually rationalizes the market, allowing the next phase to build on more solid footing.

I don't know whether these red flags are in fact a signal of some systemic concern or coming AI correction, or what specifically those series of dominoes would be. Crypto had specific, fraudulent catalysts like FTX and Terra/Luna that accelerated the unwind. AI may follow a different path, or the frothiness may prove less concerning since this is such a catalytic technology. But nevertheless, it is a dynamic worth paying attention to and was cited as the reason for broader markets selling off this week.

Technology Cycles and Infrastructure Convergence

Despite these concerns, none of this changes the fundamental assessment that AI represents perhaps the most transformative technology shift ever. The models are improving at an exponential rate and costs are plummeting. That acceleration continues, and the shortcomings revealed by pushing models to their limits become the targets for the next wave of improvement. The same was true about blockchain technology during the 2021-2022 bubble.

The underlying premise remained valid even as the funding environment became unsustainable. That pullback was ultimately productive for crypto, clearing out unsustainable projects and refocusing capital on productive applications like the ones gaining widespread adoption today.

The stablecoin adoption that followed the collapse has led to programmable money infiltrating commercial society, with businesses across every vertical gaining access to crypto-native financial services and tokenization. AI models with elite capabilities are becoming universally accessible. The convergence isn't speculative. It's happening now, and the Alpha Arena experiment demonstrates what becomes possible when you connect capable models to capital and let them operate in real environments.

The models are getting refined, the infrastructure is being quickly deployed and adopted. The funding environment will sort itself out through market discipline. And I think that we’re going to see the convergence of programmable intelligence with programmable money opening more possibilities we're just beginning to explore.

As always, thanks for reading.

- Jake

Jake Dwyer
Founder & Managing Partner
Factor Capital
Website