Allocators: alpha compresses fast, but fundamentals now drive price action more than positioning

Crypto and AI overlap more legitimately now, say the guests

Arthur Cheong of Defiance Capital argues that after years of "force fitting" AI and crypto themes since 2023, genuine overlap has emerged. He points to Nous Research, developer of the open-source Hermes agent framework, as an example of a crypto-native team now taken seriously in AI circles, and to Venice AI, a privacy-focused inference router he compares to OpenRouter, which he says has a "hockey stick" growth chart despite smaller volumes.

Rajiv Patel-O'Connor of Framework Ventures highlights financing as crypto's clearest AI-adjacent opportunity, citing portfolio company USDA which does "mid-market cluster financing" for GPU purchases. Smac of Compound frames the overlap through talent and ideology rather than efficiency, saying crypto AI competes less on raw performance and more on "coordination" and "permissionlessness" versus centralized frontier labs.

On whether AI agents will actually transact on blockchains, Rajiv raises the idea of agent swarms needing durable, available messaging, suggesting Arweave or Filecoin could be tested for agent activity. Arthur counters that permissionless rails could also let a "rogue AI agent" exploit systems more easily than a bank API, but maintains such an agent would still be "definitely going to be crypto" over a fintech API if it went rogue.

Allocators size into illiquid positions by budgeting for slippage

Arthur says building size in illiquid tokens isn't a major issue below roughly $500 million to a billion dollars in fund size, as long as managers budget for variance.

"we basically budget for like uh 5 to 10% uh variance in terms of our final entry price when we are building any position" — Arthur Cheong

He adds he generally prefers closer to 5% slippage tolerance on both entry and exit.

Rajiv says Framework looks for "venture style outcomes" whether in liquid or private assets, accepting long holding periods and volatility. Smac notes Compound shifted toward liquid markets years ago partly because private seed deals only generated four or five transactions a year, too slow relative to the asset class's pace, and because the investable universe has grown.

Dispersion is rewarding selective managers, though crowding remains a risk

The guests agree this cycle shows unusual dispersion compared to prior "everything up, everything down" markets. Arthur says this is "the best time" for active managers, noting hype has stayed a consensus name for over a year yet kept outperforming, which he sees as healthy evidence that consensus positioning doesn't preclude continued gains if fundamentals hold.

Smac warns that ideas compress into crowded trades faster than in the past, so investors must model "who will be the marginal buyer" and size into convictions over time rather than after the idea becomes obvious. Rajiv adds that timing "probably matters a little bit" when many managers share similar theses.

Arthur also describes this as a "golden age" for long-short strategies, since failing projects can be shorted for repeated declines, and some managers run cross-asset trades, long crypto against short equity or vice versa, with fundamental drivers overpowering technical signals this year.

Bitcoin remains the default benchmark, though views diverge on its relevance

Arthur says Bitcoin "is still the best benchmark we have" for most LPs, alongside absolute returns, calling these "the only two things people care about." Smac notes that in the ETF era, underperforming a simple IBIT allocation undermines the case for active management, making Bitcoin a "bare minimum baseline," though he floats the idea crypto could eventually be benchmarked against broader deep-tech equity indices. Rajiv is less committal, saying peer benchmarking is also reasonable for sophisticated LPs.

Seed-stage opportunity may be reopening as growth-stage deals get crowded

Arthur, describing himself as not a venture investor, observes heavy overlap among late-stage deals in both equities and crypto due to scarce opportunities, while seed-stage investing carries more risk now that AI is pulling away quality founders. He argues crypto and AI remain rare spaces where a 50x return is achievable within one to two years if seed investing is done right.

Rajiv disagrees that founder quality has declined, saying "the number of extremely high quality founders at early stages is still there," even as fewer founders raise seed rounds amid capital right-sizing. Smac says Compound will stay disciplined at pre-seed and seed stages rather than drift into growth, and speculates some previously "written off" crypto sectors could work in a future attempt with the right founder and timing. All three touch on AI reducing headcount needs, with Rajiv noting a meaningful share of raised capital is shifting from salaries toward compute and API credits.