Name: The Intersection of AI and Crypto: What Worked, What Didn’t, and What’s Next | Roundup
Uploaded: 2026-01-23T09:00:40+00:00
Duration: 3608 s

Speakers

Xavier

Mike Ippolito — Co-founder, Blockworks

Myles O'Neil — Guest

Summary

The hosts discuss the intersection of AI and crypto, focusing on why early compute and decentralized infrastructure bets struggled and where value may accrue instead. They highlight Bittensor as a rare working infrastructure coordination layer, remain skeptical of decentralized compute and training, and see more promise in applications, agents, DePIN data, stablecoins, and AI-driven creator tokens. They also debate how AI will disrupt SaaS pricing and change how businesses and agents coordinate data and payments.

AI-crypto's first wave was dominated by hype and failed infrastructure bets.
Decentralized compute and training face latency and centralization disadvantages.
Bittensor is viewed as a rare successful crypto AI infrastructure coordination layer.
Value is shifting toward applications, agents, and crypto rails.
DePIN data networks may sell proprietary or anonymized data to AI customers.
AI may disrupt SaaS per-seat pricing toward usage or outcome pricing.
AI agents may use stablecoins and crypto rails to hold and move money.
AI-driven creator tokens are discussed as a long-term speculative theme.

Ideas

Xavier 4:11

Bullish crypto AI applications and agents.

The most interesting and valuable part of crypto x AI is the application/agentic layer, not infrastructure. Agents will need 24/7 permissionless, open, non-siloed crypto rails for payments and actions, and crypto applications can offer better UX and serve demand that centralized incumbents are not meeting. This is where crypto can actually compete.

Xavier 4:11

Bearish crypto AI infrastructure.

Crypto AI infrastructure is commoditizing quickly and most funded infrastructure projects have not found product-market fit. It is very hard to compete with centralized incumbents that spend billions on compute, training, and inference, so value is more likely to accrue at the application/agent layer.

Mike Ippolito Co-founder, Blockworks 8:53

Avoid decentralized compute and training networks.

Decentralized compute and distributed training networks are structurally disadvantaged. Compute and training are fixed-asset heavy, highly specialized, and latency-sensitive, and centralized Nvidia-style data centers do it better. Crypto incentives have not proven they can coordinate these resources better than centralized firms.

Mike Ippolito Co-founder, Blockworks 15:31

Bearish per-seat SaaS pricing model.

AI agents will shift software pricing away from per-seat subscriptions toward usage- or outcome-based pricing, breaking the traditional SaaS model. The transition will take years because buyers are used to subscriptions and big incumbents must move first, but the direction is clear.

Xavier 25:53

Bullish Bittensor and its subnet ecosystem.

Bittensor is one of the few crypto AI infrastructure projects that works. It uses token incentives to coordinate a cheap open supply side of compute, data, and testing, subsidizing builders' capex so they can launch products more cheaply and target indie AI developers. Some subnets show early PMF and the token has held up relatively well, but it still depends on token appreciation and productization.

Myles O'Neil Guest 42:41

Bullish Geodnet's precision GPS data.

Geodnet has a real product for non-crypto users: more accurate GPS data for precision agriculture sold to farmers and large data buyers. It is cheaper and better than alternatives, giving DePIN a credible wedge outside crypto.

Xavier 43:42

Bullish DePIN data networks.

The data side of DePIN is promising. Networks like Grass, Bless, and Vana can use token or edge-compute incentives to collect permissioned, anonymized, proprietary data, such as how users interact with LLMs, and sell it to AI companies outside crypto. Bless has a new go-to-market for anonymized LLM usage data and is getting early enterprise AI interest, though token optics and distribution to non-crypto customers remain difficult.

Mike Ippolito Co-founder, Blockworks 46:18

Watch Hivemapper as proprietary geo-data.

Hivemapper's dashcam-mapped geographic data may be a proprietary, recurring data source for LLMs and large data purchasers because it is not easily scraped from the open web and must be continuously refreshed. The risk is becoming dependent on an LLM aggregator or selling commoditized data, which would squeeze the business.

Myles O'Neil Guest 54:40

Bullish stablecoins for AI agent payments.

As companies adopt AI and agents automate workflows, agents will need to hold and move money. The most natural rail for agent value transfer is digitally native tokens and stablecoins, likely abstracted behind normal payment experiences such as Stripe checkouts, so stablecoin usage should grow.

Xavier 58:44

Bullish AI-driven creator tokens.

Long-term bullish on AI-driven creators and AIXBT-like experiments. As LLMs improve and content generation becomes far better and cheaper than human creators, AI personas with large audiences should monetize. Japan's hologram concerts are an early precedent.

The Intersection of AI and Crypto: What Worked, What Didn’t, and What’s Next | Roundup

Summary

Ideas

Up Next