Tom Schmidt of Dragonfly on why decentralized AI compute is overhyped and agent payments are the real crypto x AI thesis

Summary

Tom Schmidt, general partner at Dragonfly, thinks most of what passes for AI-crypto investing is misguided — and he's willing to say so.

The category has collapsed largely into decentralized inference: GPU marketplaces promising to distribute AI compute across consumer hardware worldwide. Schmidt dismisses the premise. Training a state-of-the-art model requires a single large co-located A100 cluster, not a million consumer-grade GPUs scattered globally. Decentralized networks simply don't have the right characteristics for the compute that actually matters. He draws a direct line back to Golem, which made the same pitch for video rendering years ago — the idea keeps returning without getting more viable.

Agent payments are where Schmidt sees the genuine AI-crypto intersection. Agents making micropayments — a fraction of a penny per API call, potentially one-time, from counterparties that will never interact with the service again — are structurally incompatible with how traditional SaaS payment infrastructure works. Stablecoins offer deterministic finality and the ability to handle those transactions at a unit economics level that legacy payment rails can't support. Dragonfly is an investor in Gold Sky, which is building in this category.

On the borderline, Schmidt flags Exo, another Dragonfly portfolio company, which allows sharded inference across a local network of consumer devices — currently optimized for Apple Silicon. Run DeepSeek R1 across a few Mac Studios on your LAN without managing a cloud dependency. Whether that architecture eventually extends to a global network is an open question, but Schmidt frames local edge compute as the directional bet regardless.

Stablecoins get more credit from Schmidt than their flat price performance implies. The real story isn't payment flow — it's the stock of new Treasury holders. Tether is now the seventh-largest holder of U.S. Treasuries, and Schmidt argues that most of that demand is genuinely incremental: offshore users in Argentina, Turkey, and across Asia who wouldn't otherwise have accessed dollar-denominated assets through a traditional broker. Rob Hadick at Dragonfly published research on this recently that Schmidt says supports the incremental-demand thesis.

Asia is running ahead of the U.S. on centralized exchange product sophistication, a gap Schmidt attributes partly to regulatory headroom. Dragonfly was an early investor in Bybit and Bitget, now two of the largest exchanges globally with billions in revenue. A structural trend he calls "CeDeFi" — a centralized front end wrapping DeFi protocols on the back end — was standard in Asia years before Coinbase launched Bitcoin-backed loans via Morpho a few months ago.

Crypto fund structure has bifurcated sharply since Dragonfly's founding in 2018. In the early days, a fund manager bundled Bitcoin, ETH, equity rounds, and SAFTs into a single vehicle because even buying Bitcoin required exotic custodians. Now that liquid access is trivial, Dragonfly doesn't buy major liquid assets in its fund at all — charging 2-and-20 to hold Bitcoin would be indefensible. The firm focuses on seed-through-Series-B equity, with some liquid treasury positions where Schmidt sees venture-style upside. Separately, dedicated liquid funds — delta-neutral, credit, or long-only discretionary — have filled the other lane.

The pressure on deployment is real. AI has made operators so capital-efficient that even well-funded startups don't need much money, flipping the dynamic from a decade ago when founders competed for capital. In crypto specifically, the fundraising arc now often terminates at a token launch rather than a Series B, which compresses how many equity checks a large fund can write. Schmidt acknowledges that some of the outsized rounds that look strange from the outside are partly a symptom of capital raised by VCs outpacing the deal flow available to absorb it.

Two risks keep Schmidt up. The first is regulatory: failing to pass a stablecoin bill or market structure legislation isn't catastrophic in the short term — stablecoins have grown through the gray area — but the longer-term question is whether the U.S. ends up with a privacy-preserving digital cash or a fully surveilled panopticon. The second is Bitcoin treasury companies: Schmidt is comfortable when prices rise, but worried about what happens to the convertible structures when they don't.