Gonka Protocol is positioning itself as a decentralized network for high-efficiency AI compute, built to coordinate GPU capacity distributed around the world. According to the project’s description, the goal is to give builders and researchers permissionless access to compute resources while rewarding participants through its native token, GNK. In a recent podcast appearance, co-creator David Liberman outlined why he believes AI infrastructure should evolve in a direction more reminiscent of Bitcoin than today’s concentrated cloud model.
Liberman’s core argument is that decentralized infrastructure can unlock underused GPU power globally and route it toward meaningful AI workloads. Rather than relying on centralized gatekeepers to allocate scarce compute, a network like Gonka aims to create a market where hardware providers are directly incentivized to contribute resources. In his view, this could improve utilization and expand access for smaller builders who may otherwise struggle to compete for compute against large platforms and well-capitalized incumbents.
A Bitcoin-Inspired Model for AI Compute
The protocol draws inspiration from Bitcoin and its ability to bootstrap massive decentralized infrastructure through aligned incentives. Liberman said Gonka is designed around a transformer-based Proof-of-Work (PoW) model for AI compute, connecting GPU owners globally into a network that performs useful computational work. The comparison to Bitcoin mining is central to the project’s thesis: if a blockchain network could coordinate physical infrastructure at global scale for monetary security, a similar architecture might coordinate hardware for AI inference and related workloads.
This framing is also a critique of conventional cloud infrastructure. Liberman argued that the current centralized cloud model for GPUs is inefficient, with resource access often controlled by a limited number of providers. In that system, pricing, allocation, and availability can become bottlenecks. A decentralized network, by contrast, could lower the barrier to participation on both the supply side and the demand side, allowing more hardware owners to monetize idle or underused capacity and more developers to tap into compute without needing approval from a centralized intermediary.
Why Gonka Focuses on Inference
One of the more notable points in the discussion was Gonka’s emphasis on inference rather than training. That distinction matters because the AI market often centers public attention on large model training runs, which require enormous capital and highly specialized clusters. Liberman’s framing suggests that inference — the ongoing process of serving, querying, and running models in practical applications — may be a better fit for decentralized coordination.
Inference represents recurring real-world demand. By focusing on this layer, Gonka appears to be targeting a segment of AI compute where distributed resource supply may be easier to organize and where demand could scale with application usage rather than only with frontier model development. That approach also supports the project’s broader case that decentralized networks can serve practical and economically meaningful workloads, not just experimental or residual compute markets.
PoW Versus PoS in Hardware Networks
Liberman also made an explicit case for Proof-of-Work over Proof-of-Stake when the objective is to grow physical compute infrastructure. His argument is that PoW aligns incentives more directly with the people who actually provide hardware. In a hardware-dependent network, he suggested, rewarding real-world contribution matters more than rewarding capital ownership alone.
From that perspective, PoW can function as a mechanism that channels value toward infrastructure expansion. If participants are compensated for deploying and operating machines, they are incentivized to improve performance, add capacity, and invest in better hardware. Liberman contrasted this with Proof-of-Stake systems, where returns may flow primarily to token holders rather than the operators responsible for delivering compute. The implication is that for AI infrastructure, an incentive model tied to productive hardware could be more effective in accelerating network growth.
From GPUs to ASICs
Another important theme was hardware specialization. Liberman argued that AI compute may follow a path similar to Bitcoin mining, which evolved from general-purpose hardware toward increasingly specialized machines. In that analogy, the long-term trajectory of decentralized AI compute could involve a shift from general GPUs to more specialized hardware, including ASICs, where efficiency gains become a major competitive advantage.
The significance of that claim is twofold. First, it frames AI compute as an industry likely to reward optimized hardware design over time. Second, it suggests that tokenomics should be built in a way that encourages innovation in hardware efficiency rather than purely speculative activity. If network rewards are directed toward productive compute and technical improvement, the protocol’s economic structure could help drive the development of more efficient machines and more scalable infrastructure.
A Skeptical View of AI Valuations
While Liberman was broadly optimistic about AI’s long-term significance, he also offered a cautionary note on current market behavior. He said AI’s eventual impact is likely to be enormous, but some company valuations may already be in bubble territory, drawing a comparison to the dot-com era. That is not a dismissal of AI’s importance; rather, it is a reminder that transformative technologies can coexist with overheated pricing during early cycles of adoption and speculation.
Within that context, Liberman emphasized the importance of economic models that prioritize hardware providers and innovators. His view appears to be that token networks tied to real compute output may offer a healthier foundation than narratives driven solely by capital inflows or sentiment. By grounding rewards in the actual cost and delivery of computation, a protocol may be better positioned to avoid artificial inflation in perceived value.
Linking Token Value to Real Compute
A recurring idea in the conversation was that a PoW-based AI network can connect coin value to the real cost of compute. Liberman argued that this makes the ecosystem healthier by reducing the chance of value being detached from underlying productive activity. In practical terms, that means the network’s economics would be anchored by hardware expenditure, energy use, operational performance, and actual computational demand.
This argument reflects a broader philosophical position often associated with proof-of-work systems: that digital asset issuance and network security are more robust when they are tied to measurable real-world costs. Applied to AI compute, the same principle would mean that token incentives are not merely financial abstractions but mechanisms that coordinate tangible infrastructure and useful work.
Decentralized AI as a Competitive Necessity
Liberman expanded the discussion beyond protocol design to the geopolitical and social importance of decentralized AI. He argued that distributed compute systems are essential if smaller countries, startups, and independent builders are to compete with centralized national champions and large technology platforms. In this framing, access to compute is not just an operational issue; it is a strategic one.
If advanced AI capabilities become concentrated in the hands of a few firms or states, the result could be a severe imbalance in innovation power and economic leverage. Decentralized infrastructure offers an alternative path, one in which compute is more broadly accessible and less dependent on the policies or priorities of dominant intermediaries. For founders and researchers outside major power centers, that could be the difference between participating in the AI economy and being locked out of it.
Autonomy, Abundance, and the Future of Work
The conversation concluded on a more philosophical note. Liberman argued that the future should not be defined by a small number of centralized AI systems controlling disproportionate wealth and influence. Instead, he said society may need billions of independent, decentralized AIs to preserve autonomy, distribute opportunity, and help create a more abundant economic future.
That vision connects the protocol’s technical architecture to a broader societal thesis. Decentralized compute is presented not simply as a cheaper or more efficient market structure, but as a safeguard against concentration of power. Whether that vision proves achievable will depend on execution, adoption, and the economics of decentralized AI networks at scale, but the underlying premise is clear: infrastructure design today could shape who benefits from AI tomorrow.
Background on the Founders
The podcast also highlighted the background of David and Daniil Liberman. Based in Los Angeles, they were described as futurists, serial entrepreneurs, investors, former Snap directors of product, and founders of ventures including ProductScience.ai and Humanism Co. by Libermans Co. Their earlier work with siblings included startups across computer graphics, finance, and AI. They also founded Frank Money and launched Kernel AR, which was later acquired by Snap in the same year.
The article noted that the founders signed the Founders Pledge and directed their economic output into Libermans Co., which it said is now valued at $400 million. It also cited backing from investors including Marc Andreessen, Josh Kushner, and Arielle Zuckerberg. These details provide additional context for the team behind Gonka, though the more important test for the project will remain technical delivery and sustained network participation.
It is also worth noting that the source material came from a sponsored podcast. As with any promotional interview, readers should distinguish between the project’s vision and independently verified outcomes. Even so, the discussion captures a growing debate in crypto and AI: whether the next generation of computation should be mediated by a handful of hyperscalers, or coordinated through open networks that reward real hardware contribution and broaden access to machine intelligence.