“We can make compute so cheap it is impossible to monopolize.” - @DaLiberman, Creator of @gonka_ai At Verifying Intelligence (@token2049), David Liberman gave a deep and charismatic keynote arguing that today’s AI chokepoint isn’t algorithms - it’s control of GPUs and power. He drew a parallel to Bitcoin: when proof-of-work unleashed open, market-driven competition, hardware efficiency exploded and capacity decentralized. Apply that same incentive design to AI, he said, and you flip the economics. Instead of proof-of-stake networks that concentrate rewards with capital and starve real hardware growth, a proof-of-work AI network pays miners/operators directly, pushing cost per inference down and total available compute up. Gonka’s twist is a transformer-based PoW with “sprint PoW”, where only a small slice of work secures the network and the bulk is spent serving real AI tasks. The outcome David envisions: abundant, low-marginal-cost compute that’s broadly accessible - too cheap and too distributed for any single platform to lock up. Also available on YouTube: