Ideas
Decentralized training costs create open-source AI edge.
Decentralized training is becoming more interesting because centralized AI platform risk (government pulling Fable, data collection) and cost pressure from token maxing are making open-source models more attractive. Distributed GPU clusters in universities and elsewhere can share the massive upfront training costs, enabling cheaper, competitive open-source models.
Decentralized inference always worse than centralized.
He remains bearish on decentralized inference because it will always perform worse than centralized inference, making it a structurally inferior solution for inference workloads.
Geonet's GPS capex moat is compelling.
Geonet is Myles's favorite DePIN project because it deploys GPS sensors globally, creating a product that requires massive capex and could not exist without a large company or network investing heavily in supply-side infrastructure.
AI business privacy use case drives adoption.
Mike believes the business use case for privacy and censorship resistance in AI is much stronger than the individual philosophical case. He sees this beginning with privacy assets like Zcash and confidential computing setups, and expects business platform risk to drive adoption.
Nous Research harness owns valuable AI layer.
Nous Research is executing an open-source, decentralized AI strategy and its Hermes harness is arguably better than OpenAI's and the best in the harness space. The harness owns the user relationship, and since Nous is also training its own models, it can vertically integrate, route to its own models, and capture significant value.
Prime Intellect could power decentralized training.
Prime Intellect could become the long-term decentralized training engine for crypto AI if it can compete with frontier labs on capability. Its model training layer would be the engine, while the harness is the car, making it a key infrastructure watch.
Index Network shows Hermes agent PMF.
Index Network is a crypto AI team that went live at ETH City as an agent within the Hermes harness. It showed product-market fit by connecting investors with relevant startups, and it illustrates how crypto teams can build valuable agents inside a harness.
Crypto AI entering next-cycle S-curve.
Crypto AI has been early, but it is now entering an S-curve with many layers of the stack—stablecoin payments, infrastructure, and applications—coming together. He is very excited and expects this category to be a major theme in crypto's next cycle.
Venice leads private consumer crypto AI.
Venice is probably the best-performing consumer crypto application right now. It demonstrates what users actually want: private, uncensored AI access through open-source models, client-side privacy, and no identity leakage, and its early insight into privacy is finally seeing timing align.
Nym targets private organizational AI agents.
Nym built Iron Claw, a private Hermes-style agent, and Xavier sees a real wedge for crypto cryptography in privacy for organizational AI brains that touch sensitive data. Hyperscalers may add some privacy, but startups can earn customers quickly by providing privacy and security beyond training into agentic workflows.
Bittensor's team leads decentralized training.
Bittensor is a network template working on decentralized training with a strong team. As crypto AI infrastructure becomes more consensus and adoption grows, it remains a relevant project to watch in the decentralized training stack.
Pluralis token launch could be significant.
Pluralis is a decentralized training research project with crypto investors. If it launches a token, it will likely be post-PMF and not a small cap, making it a potential token launch to watch.
Crypto AI tokens will return strongly.
He is bullish tokens, especially at the crypto-AI intersection. These teams are likely to launch tokens, and because many are reaching PMF first, they will not be ultra-speculative small caps; consolidation and demand for access should lead to fundamentally driven valuations for these networks.
Ethereum sustainable fees are out-of-consensus setup.
Ethereum sustainable fee generation is out of consensus right now. The market does not believe ETH can generate sustainable fees, but if Jevons paradox plays out and demand eventually materializes, the current disbelief creates a contrarian setup, though there is a timing gap before demand arrives.
This Bell Curve video, published June 19, 2026,
features Myles O'Neil, Mike Ippolito, Xavier
discussing Decentralized training, Decentralized inference, GEOD, Privacy coins, Nous Research, Prime Intellect, Index Network, AI-SECTOR, VVV, NYM, TAO, Pluralis, Crypto AI tokens, ETH.
14 trade ideas extracted by AI with direction and confidence scoring.
Speakers:
Myles O'Neil,
Mike Ippolito,
Xavier
· Tickers:
Decentralized training,
Decentralized inference,
GEOD,
Privacy coins,
Nous Research,
Prime Intellect,
Index Network,
AI-SECTOR,
VVV,
NYM,
TAO,
Pluralis,
Crypto AI tokens,
ETH
