And almost all of the infrastructure behind it is owned by three companies.

Amazon, Google, and Microsoft own the data centers that train and run the world's AI systems. If you want to build something serious with AI today, you are almost certainly renting compute from one of them. The waitlists for high-end GPU capacity, the specialized chips that train large models, regularly run months long. Pricing is set by the supplier. Access is granted by the supplier. The supplier decides.

AI x Crypto is a group of projects building infrastructure for AI that operates outside the control of those three companies.

Some are providing direct alternatives to what Google and AWS offer. Instead of renting compute power or storage from a centralized provider at fixed prices, you can access the same resources from a global network of independent providers who compete for your business. No approval process. No single company controls access or sets terms.

Others are building services that the big 3 have no structural incentive to build at all. Permanent data storage, where you pay once and your data is held indefinitely by a self-funded endowment, is one example. A competitive marketplace where contributors around the world submit AI work, independent validators score it, and resources flow to whoever performs best, is another.

The category has been developing for a few years. Three recent developments make this area of crypto interesting.

Regulatory clarity arrived. In March 2026, US regulators formally classified a group of major crypto assets as digital commodities under federal law. Large institutional investors and asset managers had been legally restricted from participating without that classification. The restriction has now lifted for a growing list of assets.

Decentralized AI training proved it works. In March 2026, a network called Bittensor trained an AI model that performed at the same level as models built by Meta, using 70 volunteers on regular home computers instead of a corporate data center. It worked.

The CEO of Nvidia said decentralized AI training is legitimate. Jensen Huang, CEO of Nvidia, said publicly that decentralized AI training and traditional AI development are complementary approaches. Coming from the person who sells the hardware the existing system runs on, that's a meaningful signal.

Crypto markets are broadly down. Bitcoin is off 45% from its recent peak. Inside that decline, the AI x Crypto category has outperformed the top 50 crypto assets by 55 to 107 percentage points over the past 90 days. This setup suggests where capital is moving now.

Seven projects currently make up the AI x Crypto space worth watching. We ran each one through a rigorous evaluation, and three passed, two are on the watchlist, and two didn't make the cut.

The three that passed:

Bittensor (TAO) — a marketplace for machine intelligence where contributors compete across 128 specialized networks and compensation flows to whoever performs best.

Render (RNDR) — a GPU compute marketplace connecting hardware owners with developers and creators who need processing power, now expanding from 3D rendering into AI workloads.

Arweave (AR) — a permanent data storage network built on a pay-once model, increasingly relevant as AI systems require datasets that are verifiable and designed to last.

On the watchlist: Akash (AKT) and NEAR (NEAR), two projects with real underlying technology that need more evidence before they earn full conviction.

Two didn't pass: Grass (GRASS), which has an interesting idea but no defensible competitive position, and Virtuals (VIRTUAL), where revenue is declining on a protocol too new to have earned that trajectory.

Over the coming weeks, we'll go deep on each one that passed. First up: Bittensor. Next week.