Uncovering Crypto Alpha: Investment Approaches in an Evolving Market

Arthur Cheong, founder and CEO of Defiance Capital, describes his firm as a crypto liquid fund that has been actively investing in the space since 2020. Their strategy aligns with a classic liquid venture style, focusing on acquiring tokens of applications that show significant potential.

The fund's approach involves buying into promising crypto projects and holding for the long term, indicative of a venture-like investment horizon but applied to liquid assets.

Rajiv Patel-O'Connor, a General Partner at Framework Ventures, elaborated on his San Francisco-based firm's investment strategy. Framework Ventures primarily engages in early-stage venture investments, complementing this with some liquid token investing.

The firm's dual approach allows it to participate in the foundational development of crypto projects while also capitalizing on more mature, liquid opportunities in the market.

Arthur Cheong observed a shift in the crypto-AI intersection, moving past early 2023 struggles with 'false fitting' narratives. He highlighted that legitimate applications are now emerging, such as Nous Research, which developed Hermes, the largest open-source AI agent framework, and also works on crypto AI.

Another example is Venice AI, described as a crypto version of Open Router that provides privacy-focused inference routing. Venice AI has shown significant, hockey stick-like growth compared to its peers, indicating a strong uptake of genuine crypto-AI solutions.

Rajiv Patel-O'Connor discussed how AI is enabling early-stage startups to achieve more with smaller teams, showcasing unprecedented operational leverage. He emphasized crypto's particular strength in capital formation, which is crucial for the extensive capital expenditure (CapEx) required in AI infrastructure.

As an example, he cited USDAI, a Framework Ventures portfolio company specializing in mid-market cluster financing. This company helps promising startups and public entities acquire significant GPU resources, such as $150 million worth of GPUs, potentially offering more attractive rates and smaller down payments than traditional financing options.

Smac from Compound argued that crypto AI's competitive edge does not currently lie in raw efficiency compared to centralized labs. Instead, its value is derived from underlying societal trends that favor coordination, diverse economic models, and permissionlessness.

He highlighted the high quality of talent entering the crypto-AI space. Arthur Cheong added that crypto AI projects prioritize encryption and privacy by default, a significant differentiator from many traditional AI projects that often overlook privacy in their early development stages. This focus on privacy is expected to become increasingly important globally.

The panelists discussed the core strengths of crypto, particularly its ability to facilitate coordination and provide effective economic incentives. As blockchains become more performant, there is a strong argument for their role in enabling machine-to-machine transactions and communication within AI systems.

Currently, centralized systems incur a minimum cost for agent interactions. The value proposition for crypto AI is to significantly reduce these transaction costs, potentially to a level where centralized providers cannot compete. However, the full feasibility of achieving this cost reduction is still to be determined.

There is a growing concern among users about data ownership and privacy within traditional AI models, leading to a pushback against frontier labs and conventional AI approaches. This trend creates an opportunity for crypto, particularly with decentralized ledgers offering a permissionless environment for AI agents to store data securely.

The concept extends to agent swarms potentially utilizing decentralized storage solutions like Arweave or Filecoin to ensure messages are durable and available. Experimentation is needed to confirm if current agent swarms are already leveraging such decentralized storage options.

The future of AI may involve a shift towards more local or smaller models, reducing reliance on massive frontier models, with a greater portion of inference being conducted at distributed levels. This directional trend aligns well with the early core beliefs of crypto, which center on distributed networks and privacy.

The question remains whether crypto projects can effectively capture the value generated from these evolving AI trends. This intersection represents a significant area of exploration, with its potential for value capture still an open question.

Arthur Cheong stated that building a liquid crypto position is not a significant issue for funds, unless the capital involved exceeds $500 million to $1 billion. He advises budgeting for a 5% to 10% variance in the final entry and exit prices to avoid the impracticality of perfectly timing the market.

Rajiv Patel-O'Connor emphasized the need to seek 'venture-style outcomes' even from liquid assets, which necessitates a long-term holding horizon. He noted that market consensus now forms more rapidly, causing prices to react quickly to collective analysis. Smac added that their firm expanded into liquid markets to gain asset class exposure, as there were insufficient high-quality early-stage founders to meet their investment volume.

The panelists agreed that the objective is to achieve venture-level returns even from liquid tokens, which requires maintaining investment discipline despite high market volatility. Liquid assets offer a viable way to gain market exposure when the volume of private transactions is insufficient to meet a firm's growth targets.

The focus is on generating specific alpha, moving beyond merely seeking levered beta. Participants in the market must be prepared to accept volatility as a natural, two-way phenomenon when accumulating positions over time.

The increased dispersion among assets presents a new challenge for active managers, making it harder to manage crowding effects. Unlike previous cycles where assets tended to move uniformly, not all assets now follow the same directional movement.

While intuitively, dispersion should benefit active managers by allowing them to demonstrate alpha through selective investments, the practical difficulty of execution due to rapid crowding-out effects offsets some of these benefits. This current bull cycle is notably characterized by significantly more dispersion than prior cycles.

In the past, crypto investing was characterized by a lower level of sophistication and a large influx of capital, making it easy for almost any investment to appreciate. However, the market's maturation means it is now more challenging to identify and size assets early.

This shift reflects a growing professionalization of the market. Currently, over 80% of liquid crypto hedge funds are outperforming both Bitcoin and ETH, with year-to-date returns for known crypto funds ranging from 20% to 100%. This indicates a market that increasingly rewards active management and fundamental analysis.

Arthur Cheong highlighted Hyperliquid as a consensus asset that has maintained its strong market cap and continued to outperform for over a year, despite its widespread recognition. This phenomenon suggests that an asset's consensus status among 'smart money' does not necessarily limit its growth, provided it maintains strong fundamentals and growth prospects.

The market is displaying greater sophistication by rewarding robust performance over mere order book positioning. Furthermore, retail investors are increasingly seeking extreme returns (10-20x) and moving away from mid-cap assets, pushing professional capital to concentrate on major or fundamentally sound mid-cap tokens.

Investors, particularly 'smart money,' are increasingly avoiding speculative meme coins. Instead, professional capital is concentrating on major cryptocurrencies and mid-cap assets that possess solid fundamental value.

The objective for these investors is to consistently outperform Bitcoin, a strategy that helps facilitate future capital raising and demonstrates value in a maturing market.

Arthur Cheong asserted that Bitcoin remains the primary benchmark for most Limited Partners (LPs), who often view the crypto space as a singular entity despite growing sectoral dispersion. Alternative indices like the CCI 30 or the Bloomberg Crypto Index have not achieved comparable popularity or acceptance among allocators.

For institutional allocators, success is measured by both absolute returns and outperforming Bitcoin. In the current era of accessible Bitcoin ETFs, such as IBIT, funds are increasingly pressured to justify their existence by demonstrating superior performance compared to these passive investment vehicles.

The venture landscape is evolving with the crypto industry's maturation, bringing debates over whether to prioritize seed or growth-stage investments. There is a clear recognition of projects that have achieved product-market fit and key use cases, prompting discussions about funding later stages.

Arthur Cheong observed an extreme overlap in late-stage deals across both equity and crypto markets, mainly due to a limited number of such opportunities. This situation leads investors to re-evaluate the optimal concentration between seed-stage and growth-stage allocations for the coming years.

Arthur Cheong believes that seed-stage opportunities in crypto are currently abundant because many investors were previously burned in market corrections, leading to reduced competition. Despite the AI industry attracting many high-quality founders, the overall quality of crypto founders remains high.

When correctly seeded, crypto projects can potentially achieve a 50x uplift within one to two years, a growth rate comparable only to certain AI ventures. This high growth potential makes seed-stage crypto an attractive, albeit risky, area for investment.

Rajiv Patel-O'Connor noted that AI is significantly reducing the capital intensity of early-stage startups, enabling teams to operate effectively with fewer employees. Portfolio companies are increasingly adopting AI-native tools, leading to a prioritization of team members who can leverage AI efficiently.

Historically, a large portion of funding went to headcount, but now, a meaningful chunk of those costs is being reallocated to compute and API credits. Consequently, $5 to $10 million in early-stage funding now stretches much further, with the exception of companies dedicated to training large-scale models, which still require substantial capital.