Paweł Czech on Money Moves: why AI compute will go decentralized

Paweł Czech joined hosts Samantha Lewis of Mercury Fund and Ian Epstein of Profitr on Money Moves for an episode titled "Globalized intelligence: How AI Can Be Built for Humanity, Not Just Hyperscalers".

Ian Epstein opened with the idea that the internet spread communication, commerce and finance to everyone, and that intelligence is next. Paweł's case: AI compute will have to go decentralized, and three things decide where AI runs, namely latency, data privacy and ownership.

What this video covers

Samantha Lewis and Ian Epstein host, and Paweł is the guest. Samantha sets the topic in the first minute. The headlines are all about centralized data centers, and Paweł has spent years working toward a world where compute can be decentralized.

Ian frames the discussion this way. The internet began with global, decentralized communication, then commerce and social life. Crypto brought decentralized finance. Intelligence is next, he says, and it will need decentralized compute behind it.

The idea Paweł puts on a pedestal is ownership. He prefers the idea of universal basic ownership to universal basic income. If people can own the technology, and part of the hardware and infrastructure behind it, either personally or as a community, access can be more democratic.

Samantha then asks why centralized compute is not the ideal way. Paweł starts from physics: how fast signals travel, and how much power you can unlock in one place and move to another. Complex questions to a central service can mean long waits, and agents will ask such questions all day. Sometimes capacity runs out and the request times out. In an operating theatre, he says, nobody can be told to come back when the tokens reset. That is where he says compute will have to go decentralized.

Asked whether latency is the main reason enterprises hold back, he calls it the visible side. The deeper issues are data privacy and data ownership. Then he names the three things: latency, data privacy, and ownership or sovereignty of the compute.

Much later he comes back to control. The question, he says, is who controls the compute and the power, and who can cut you off. A small business with a handful of agents should be able to choose between corporate infrastructure and something its community provides.

Key moments

• 1:25 Ian Epstein frames the idea: from decentralized communication to decentralized intelligence
• 6:19 Universal basic ownership
• 20:13 Why centralized compute is not the ideal way
• 23:01 The hospital operating theatre example
• 23:37 Why compute will have to go decentralized
• 24:18 Latency is the visible side; privacy and ownership run deeper
• 25:36 The three things that decide where AI runs
• 44:24 Who controls the compute

Quotes

“If we want to benefit from these improvements in our day-to-day lives, it means that compute will have to go decentralized.”