Charles Hoskinson: Crypto Will Eat AI as the Data Center Boom Goes Dark
Charles Hoskinson says the cryptocurrency industry is about to do to artificial intelligence what it once did to cryptography, and he thinks the spending spree behind today’s AI data centers is heading for a bust.
In the September 16 episode of the Deeptech Insights podcast, the Cardano founder argued that blockchains could give AI payment rails, data ownership, provenance, and distributed computing as the infrastructure boom runs into economic limits.
Why Hoskinson Thinks Crypto Eats AI
Hoskinson said spending 10 times more on data centers every year cannot continue because there is not enough electricity to support that pace. Companies such as OpenAI and Anthropic also need to become profitable at scale, he said, with pre-training creating much of the financial pressure.
The developer compared AI’s position today with cryptography when he entered the industry, saying that cryptographers objected to being associated with cryptocurrency, a stance that ended once cryptocurrency had the money to hire the best cryptographers. He expects AI to follow the same path within five to ten years.
“Cryptocurrencies are going to eat AI because we solve all the hard problems that AI can’t solve,” Hoskinson said.
The problems in question are payments, alignment and data provenance.
His alignment argument is that blockchains create shared rules among participants, while AI companies make their own decisions about issues such as free speech and acceptable behavior.
A blockchain-based system, in his view, could provide a shared mechanism for those rules rather than leaving them to individual companies. Blockchains could also track who created data and how it changes hands, creating records for intellectual property and automated royalties when AI systems use someone else’s work.
The Input Output CEO also raised the idea of pooling ordinary phones and GPUs together as a training resource, arguing that would beat building new data centers altogether.
He compared it to the fiber optic buildout of the late 1990s, when about 90% of the cable laid nationwide sat unused for close to a decade before demand caught up. He expects something similar with data centers: overbuilding now, then a shift toward smaller local models running on personal hardware, like Apple’s M5 Mac Studio.
If frontier AI increasingly runs on networks of smaller machines instead of centralized data centers, Hoskinson argued, cryptocurrency is “the only coordinating technology that exists to do that.”
