Founders Fund's Bridget Harris breaks down the Prime Intellect raise and where crypto investing goes from here

Summary

Bridget Harris of Founders Fund breaks down the firm's investment in Prime Intellect, a distributed compute marketplace, and lays out where she sees genuine opportunity in crypto over the next four years.

Prime Intellect

Most AI crypto projects Harris had seen before Prime were effectively copy-paste products with a token grafted on top — character AI with a token, LangChain with a token, Hugging Face with a token. Prime Intellect stood out because the founders, Vincent and Johannes, were tackling distributed training directly. Founders Fund reached out after Prime publicly demonstrated a 1.1 billion parameter model trained across three continents. By the time of the segment, the team had followed that with a 10 billion parameter run.

The core product is a peer-to-peer compute marketplace. On the demand side: AI startups, independent developers, and labs that need burst capacity. On the supply side: data centers, individuals, and startups with idle GPUs — Harris names Hugging Face and SemiAnalysis as examples. The technical underpinning is an open-source implementation of Google DeepMind's DiLoCo framework, a distributed low-communications training protocol, which Prime released as OpenDiLoCo and which Harris describes as a meaningful step-function improvement in cross-geography training.

The crypto layer matters here for the same reason it mattered in prediction markets: settling compute payments across geographies, currencies, and counterparties without the overhead and markup that compute brokerages currently charge.

Crypto reserve reaction

On the U.S. strategic crypto reserve announcement, Harris says the XRP and Cardano inclusions drew genuine frustration across the community. Two explanations circulated. The first is unit bias — Cardano trades at a fraction of a cent versus Bitcoin at six figures, and retail buyers psychologically prefer cheaper tokens, so Trump may have been playing to that base. The second, attributed to Udi Wertheimer of Taproot Wizards, is that Trump opened with an extreme position knowing he would roll it back, ultimately landing on a Bitcoin-only reserve while appearing to make a concession — which would itself be a significant act of Congress.

Investment posture and deal pace

When Harris and her colleague Joey joined Founders Fund roughly two years ago, sentiment was at a low — six months post-FTX, with crypto-only funds and generalist funds alike sitting on their hands. They moved quickly and, in retrospect, the timing was close to optimal. Now the picture has reversed: sentiment is soft but new deal prices are, in Harris's words, "asymmetrically high." Founders Fund has not done a net new deal since Prime Intellect in October or November 2024, a gap of roughly five months at the time of recording.

The filter Harris applies is whether crypto is a step-function improvement over the non-crypto version, not just an add-on. Polymarket is the model: prediction markets existed before it, but liquidity was thin, counterparty risk was high, chargebacks were a problem, and currency conversion was costly. Crypto solved all of that structurally.

DeFi and stablecoins

DeFi founders had the hardest regulatory environment of the past four years, with many leaving the U.S. entirely. Harris expects a new market structure bill to include regulatory exemptions for DeFi protocols specifically, on the logic that a set of smart contracts cannot be regulated the same way a broker-dealer can. If those exemptions arrive, she thinks it reopens the sector. Founders Fund has invested in Lava, a Bitcoin lending protocol, which Harris sees as a natural first-use-case given the volume of Bitcoin holders who want yield without selling.

On stablecoins, the Bridge acquisition by Stripe triggered roughly 50 pitches to Founders Fund alone. Harris's read is that the most credible near-term opportunity is B2B cross-border payments, where volume is in the billions annually, fees are high, and settlement times are slow. Consumer-side stablecoin payments are more complex — Visa and Mastercard's 2% interchange flows back to users as rewards, creating a different incentive structure — and stablecoins may end up running under the hood of existing card rails without users knowing. The remaining friction on the institutional side, which Harris acknowledges without resolving, is error recovery: crypto transactions are irreversible, which makes fat-finger risk a genuine infrastructure problem for any bank-scale deployment.