When Social Networks Miss the AI Wave: Why ChatGPT Is Becoming the New Town Square
Over the weekend, I finished watching the new interview with the Liberman brothers, and honestly, it’s one of those conversations that stays with you.
Daniil and David Liberman aren’t your typical Silicon Valley entrepreneurs. These guys are serial builders who’ve been at it for years - they ran products at Snap, sold their AR startup Kernel to Snap back in 2016, and have been shipping products since they were kids in Moscow.
Their parents were state biophysicists who proposed that the human brain is basically a quantum molecular computer, which explains a lot about how these brothers think.
But here’s the insight that hit me: they mentioned that social networks completely missed the AI wave, and ChatGPT is already transforming into a social network where people are spending more and more time.
Think about it. Users are now spending an average of 13-14 minutes per session on ChatGPT. That’s approaching YouTube territory. The platform has 700 million weekly active users as of late 2025, up from 300 million just a year ago. People are spending billions of hours there, and that number keeps climbing.
OpenAI has recently launched Group Chats - up to 20 participants with an AI moderator/co-author, currently piloting in Japan, New Zealand, South Korea, and Taiwan. They’re also testing Sora for generating and sharing short videos in a feed with friends. This isn’t a productivity tool anymore. This is social infrastructure.
The Libermans have always had this ability to see around corners. After working as directors of product at Snap focusing on AR and avatars, they created the holding company Libermans Co and launched Humanism (investing “in people”) and Product Science - a mobile app acceleration service that raised $17.5 million.
But here’s where it gets interesting: they’ve essentially put everything on pause to prioritize their new project, Gonka.ai.
The Decentralized Compute Play
Gonka is building a decentralized AI compute network where owners of computing power receive cryptocurrency as compensation. It’s not a company or a startup - it’s a protocol, governed by consensus, similar to how Bitcoin operates.
The brothers believe they can scale quickly, and they’ve already connected over 450 GPUs to their network (some sources say they’ve passed 60 active hosts including professional miners, VC firms, and GPU data centers).
Their bet mirrors what Durov is trying with TON and Cocoon - they’re convinced they can achieve rapid network growth by aligning incentives correctly.
The Economics Actually Make Sense
Here’s where my finance brain kicked in. The economics of Gonka look surprisingly reasonable:
Renting a GPU costs about $2/hour ($50/day). At roughly 100 units per hour, that’s a cost of approximately $0.02 per unit.
For corporations, depreciation and margins push this 2-3x higher. But in a decentralized network, you’re only adding a 10-20% commission (roughly $0.022-0.024 per unit).
Compare this to centralized providers who are marking up compute by 200-300% to cover their infrastructure costs and margins. The decentralized model fundamentally undercuts them because it eliminates the middleman layers.
I read through their whitepaper and documentation. They’ve made progress much faster than Durov with his 1 billion TON users and current focus on gifts and NFTs - which, let’s be honest, are pretty much the only tangible value for TON besides crypto exchanges at this point.
The Proof-of-Work Innovation
What’s clever about Gonka is their “Transformer-based Proof-of-Work” mechanism. Unlike Bitcoin, where computational power is essentially wasted on solving arbitrary puzzles, Gonka channels nearly 100% of computing resources into meaningful AI tasks - training and inference work that actually matters.
David Liberman explains it like this: when they were in college in the early 2000s, they built a community of 300,000 computers worldwide. That network had a total power of 10 teraflops - more than the largest supercomputer at the time. They realized then that the networks of the future should be decentralized.
The key insight is that most decentralized AI projects waste enormous amounts of computational power on verification and staking. Gonka eliminates that waste by making the consensus mechanism itself perform useful AI computation.
What They’re Really Building
The bigger picture here is about where power and infrastructure will live in the AI era.
Right now, AI compute is concentrated in the hands of a few massive players - AWS, Google Cloud, Azure, and a handful of others. This creates natural monopolies, high costs, and potential censorship chokepoints.
A decentralized compute network solves multiple problems:
It dramatically reduces costs for AI developers
It prevents any single entity from controlling access to AI infrastructure
It creates economic incentives for hardware innovation (similar to how Bitcoin mining drove GPU evolution)
It allows individuals and smaller players to participate in the AI economy
The Libermans aren’t just building infrastructure. They’re betting that decentralized AI compute become the standard - potentially creating infrastructure larger than Big Tech’s centralized stacks.
The Real Question
I’m genuinely curious about real-world experience with decentralized computing power.
The Libermans have this track record of being early to trends. They were building AR before it was cool. They understood social avatars before Bitmoji became ubiquitous. They’ve attracted investments from Marc Andreessen, Josh Kushner, and Arielle Zuckerberg, and their holding company is now valued at $400 million.
So when they say social networks miss AI and decentralized compute will reshape the industry, I’m inclined to pay attention.
And the main question is whether it happens fast enough to matter before the incumbents lock in their advantages. And whether projects like Gonka can actually deliver on the promise of making AI compute as accessible as Bitcoin made digital currency.
Time will tell. But the economics look solid, the team has credibility, and the timing might be perfect.