When Anthropic complied with U.S. export restrictions and shut down its Fable 5 and Mythos 5 models for foreign users regardless of location, Jake Brukhman saw a pattern. The CoinFund founder argued that AI development is increasingly centralized, making advanced models natural targets for government control.
Export controls expose weak points of centralization
Brukhman pointed out that Anthropic suspended both models after receiving a directive affecting foreign nationals. He said this episode proves that concentrated AI systems buckle easily under policy pressure. “If training compute is held by one or two giants, regulators can flip the switch for the whole ecosystem.”
Distributed training: coordinating global GPUs with new algorithms
According to Brukhman, the industry's biggest hurdle is not demand but access to large-scale computing power. Enough commodity GPU capacity exists worldwide to rival centralized clusters — if somebody can solve the coordination problem. He named several teams working on it: Gensyn, Prime Intellect, Bagel, Pluralis Research, Nous Research, Macrocosmos, and Covenant. Prime Intellect raised seed funding co-led by CoinFund in 2024 and later added more capital.
Open-source AI’s sustainability problem and the tokenized fix
Beyond compute, Brukhman flagged another issue: open-source projects lack durable business models despite wide accessibility. Pluralis Research is taking a different approach by splitting model weights across network participants. That structure could support tokenized AI models while preventing any single entity from controlling the whole system. “No one party can run the model alone,” he said.
The broader debate now centers on whether AI development remains concentrated among a few organizations or expands through open decentralized networks. Which path the industry takes will determine who ultimately controls intelligent systems.