Where Is The Crypto Alpha Hiding This Cycle?

Crypto was described as cautiously optimistic and possibly still early in a bull cycle, but more mature and selective than in earlier cycles. Because market moves can unfold in three to six months, investors need to weigh near-term momentum against 12-, 18- and 24-month horizons, asking whether a thesis is strengthening or merely becoming crowded. The search for alpha is therefore less about buying a broad narrative and more about identifying projects with durable use cases, growth and value capture.

Crypto-AI is moving beyond what speakers called a forced pairing in 2023, with some teams now showing clearer products. Arthur cited Nous Research, which built the open-source AI agent framework Hermes and has also worked in crypto-AI, as a credible AI team. He also pointed to Venice AI, whose privacy-focused inference routing is growing rapidly. Venice’s token spend and inference volume remain below OpenRouter’s, but its growth curve was described as “hockey stick” shaped.

The proposed crypto-AI edge is not necessarily greater efficiency than centralized systems today, but permissionless coordination, privacy and alternative economic models. Crypto may also help finance AI infrastructure: portfolio company USDA AI offers mid-market cluster financing, including for buyers seeking $150 million in GPUs. Whether crypto can make machine-to-machine transactions cheap enough, or capture value from distributed inference using smaller models, remains unresolved. Those open questions make the category promising but still difficult to underwrite.

Privacy has become a more prominent crypto narrative, with Zcash, NEAR and zero-knowledge-related assets cited as liquid opportunities. Arthur said crypto-AI projects tend to enforce encryption and privacy by default, and expects privacy concerns to grow. Speakers also raised the possibility that distributed, permissionless ledgers or storage networks such as Arweave and Filecoin could make messages durable or available to AI-agent swarms. The same permissionless properties could, however, make it easier for rogue agents to exploit those systems.

The fund has shifted from primarily private seed and pre-seed investing toward more liquid exposure, seeking broader asset-class coverage after finding too few early-stage founders it wanted to back. It may make only four or five private transactions a year, while crypto markets move faster. Building liquid positions is not frictionless: liquidity can be thin, OTC access uncertain and trades crowded. Arthur said his team budgets for roughly 5–10% variance in a position’s entry price, usually aiming for about 5%, and applies similar thinking to exits.

Speakers estimated that more than 80% of the liquid crypto hedge funds they know had outperformed Bitcoin and Ether, with most of those funds up roughly 20% to 100% year to date. They argued that market dispersion can reward active managers, while Hyperliquid illustrates that consensus ownership does not rule out further gains when fundamentals and growth remain strong. Bitcoin remains the most common benchmark for LPs and larger allocators, although investors may also judge absolute returns, peer performance or, if crypto becomes less distinct, technology equities and indices.

Private investing still offers distinct trade-offs: late-stage crypto deals are few, heavily overlapping among investors and hard to access, though they have generally done well. Seed opportunities may improve after investors were burned, but founder quality is difficult to assess and underwriting risk is high; Arthur said a well-executed crypto or AI seed investment could potentially deliver a 50x uplift in one to two years. His hedge fund would ideally hold 10–20 names but is usually restricted to 5–10. Meanwhile, AI may let early-stage companies make more progress without matching increases in headcount: some funding once directed mainly to hiring may shift to compute and API credits. A $5–10 million raise may go further for AI-enabled teams, though not necessarily for projects training models, and marketing and enterprise sales can still require people.