AI Is Getting More Powerful—Who Will Get Richer? Gonka Co-Founder on Compute and Power
If future AI can help us conduct research, write software, drive cars, and even take on everyday work through robots, then what determines our situation is not just how smart AI is.
There are also several more fundamental questions: Who can access these capabilities? Who controls the infrastructure that supports them? And where does the new value created by technology ultimately flow?
In a conversation published by MaeveKnows, Gonka co-founder Daniil Liberman starts from decentralized AI to discuss compute concentration, Bitcoin's organizational mechanisms, the productivity potential of AI and robots, and how ordinary people can participate in this transformation.
His core argument can be summarized in one sentence: "We should all use AI, but we should also own AI."
Here, "own" refers not just to a product account, but to the infrastructure behind AI, as well as the opportunity for people to participate in building it and sharing in its development.
1. Decentralized AI Is First and Foremost Not a Crypto Industry Problem
Daniil begins by distinguishing two concepts: the significance of decentralized AI, and the role that crypto mechanisms play within it.
In his view, discussing decentralized AI is not necessarily about discussing crypto assets themselves. Crypto mechanisms are more like a tool that enables dispersed participants to jointly build and maintain infrastructure.
What really needs to be answered is: As AI becomes an increasingly important source of capability, can we make it accessible to more people, rather than letting a few institutions decide who can use it?
Daniil believes that access to advanced models will affect multiple domains. Those who can use the latest models may gain advantages in economic activity, politics, media, scientific research, and creativity.
As model capabilities grow, this advantage may also expand. AI is no longer just helping someone write an email; it may directly influence what an individual, a company, or even a society can accomplish.
Therefore, what concerns him is not the label of "decentralization," but how access to technology is distributed, and what consequences that distribution will produce.
2. Beyond the AGI Timeline, What Matters More Is Who Can Access It
Daniil holds a positive judgment on the pace of AI development. He believes that artificial general intelligence capable of solving a broad range of human tasks may emerge within the next few years to a decade.
This is his judgment on technological progress. But in this discussion, what he wants to probe further is not exactly which year AGI will arrive, but rather: When it arrives, will we have already established the conditions for more people to be able to use it?
If only a few institutions can access the most powerful models, they may be the first to gain stronger research, production, and organizational capabilities.
For everyone else, the question is not just "my tools are a generation behind," but whether they still have the ability to participate in the same competition.
Daniil therefore connects the popularization of AI with underlying compute. In his argument, to make advanced AI more widely available, one cannot focus only on upper-layer products; one must also consider who provides and who controls computational resources.
3. What Does Compute Concentration Mean?
The estimate Daniil gives in the conversation is that over 80% of AI compute chips are located in the United States, with about 70% held by three to five companies.
What he emphasizes through these numbers is that AI infrastructure is already highly concentrated.
This concentration involves not only the location of hardware, but also who has the authority to decide the access conditions for model services. He mentions that access to some advanced models has already been restricted, and that national policies have begun to influence these arrangements.
When AI capabilities are still mainly reflected in chat and content generation, these restrictions may manifest as products being unavailable, models being inaccessible, or service rules changing.
But if future AI can drive robots, autonomous vehicles, and take on more software development and scientific research work, then control over the relevant capabilities could affect the supply of a broader range of goods and services.
Daniil describes this potential influence as "the nuclear weapons of the new era."
This metaphor emphasizes the weight of control: If a technology can reshape the economy, those who master it may possess enormous power.
He therefore advocates pushing decentralization down to the compute layer, rather than stopping at the application or interface layer. Otherwise, even if users are dealing with different products, the key resources supporting those products may still be concentrated in the hands of a few entities.
4. What Bitcoin Offers Is Not Just an Asset, but a Building Mechanism
Decentralizing control sounds simple, but actually building an AI compute network requires hardware, electricity, facilities, and long-term investment.
Daniil readily acknowledges that this kind of infrastructure requires investment on the scale of hundreds of billions of dollars. The question is: Without a unified company to raise funds and build it, how can dispersed participants complete such a massive undertaking?
He uses Bitcoin as his primary reference.
According to his estimate, over roughly fifteen years, Bitcoin built computational infrastructure on the scale of tens of gigawatts from scratch, with related construction investment reaching at least $300 billion.
In this network, participants are spread across the world, may not know each other, and do not need to obtain permission from a central authority before contributing resources.
For Daniil, what is noteworthy is this coordination capability: a shared rule system that allows strangers to continuously contribute real physical resources without management by a unified company.
Bitcoin completes its proof of work through hash computation. What Daniil hopes to borrow is not letting these hash computations directly perform AI tasks, but the organizational and incentive logic behind them.
In the model he describes, participants contribute compute to the network and receive corresponding network token incentives. Early contributors can participate in network building earlier and receive initial incentive allocations.
Such a mechanism attempts to link infrastructure building with participants' interests, so that the network's development does not rely entirely on the capital expenditure of a single company.
The decentralized AI compute direction explored by Gonka can also be understood in this context: how to organize independently provided computational resources to provide services for AI applications.
5. Who Will Own the Growth Created by AI and Robots?
The discussion then shifts from compute building to economic distribution.
Daniil paints a future scenario: Everyone can access a powerful AI assistant, receive action recommendations, and have robots help complete real-world tasks.
In this scenario, AI is not just answering questions, but participating in the process of people achieving their goals. It can help people work more effectively, make choices, and accomplish things that originally required more time and resources.
To illustrate the productivity potential, he proposes a simplified time model:
A person works eight hours a day, while AI and robots can work twenty-four hours continuously. Combined, this could give an individual the equivalent of four times their original work capacity.
"Four times" is a hypothetical he uses to explain future possibilities, not economic growth data that has already been achieved. The key point is that the ability of machines to work continuously could greatly expand what people can accomplish.
But Daniil immediately raises another question.
If only a few entities control these capabilities, they could capture most of the new value. Even if everyone else lives in a world with higher productivity, they may not be able to share in the corresponding growth.
In other words, the entire economy becoming larger is not automatically equivalent to everyone's situation improving.
Daniil hopes that AI will drive broader abundance, rather than perpetuating a structure in which a few control most resources and everyone else can only share limited benefits.
His focus on decentralization is therefore not only about preventing power concentration, but also about finding a way for more people to participate in new value creation.
6. What People Oppose May Not Be AI, but Losing Income and Autonomy
When asked whether people should start using AI now, Daniil shares a story about a friend's child.
This thirteen- or fourteen-year-old child says that he and his classmates oppose AI. But during the conversation, he also mentions that he has learned some things from games.
According to Daniil's account, the child encountered chatbots in Minecraft, Roblox, or similar gaming environments, but did not realize that the gaming experience he enjoyed already included AI.
Thus a contradiction emerges: saying they oppose AI while enjoying the help AI provides.
What Daniil expresses through this story is that people's attitudes toward AI cannot necessarily be simply divided into support and opposition.
Many people can feel the convenience of AI, yet worry about losing their jobs, sources of income, or the space for autonomous choice. These concerns and whether the technology is easy to use are two different levels of issues.
He says he can understand this fear.
His response is: If people merely face technological replacement passively without being able to share in the value AI creates, then anxiety is understandable; if more people can participate in and share growth, technological progress can become an opportunity rather than just a threat.
7. "Owning AI" Requires Shifting Our Gaze to Physical Infrastructure
"We should use AI, and we should also own AI" is one of the most striking viewpoints in this conversation.
Daniil extends this statement further to data centers, electricity, and chip manufacturing.
In his vision, the question of AI ownership cannot be discussed only at the software interface level. Models depend on real facilities to run, and the construction, control, and benefit structures of these facilities will also shape the future of AI.
He hopes that more people can participate in infrastructure building through distributed mechanisms, rather than always purchasing services from a few companies.
This does not mean everyone needs to personally build data centers. The core is to explore a different organizational approach: a broader range of participants jointly contributing resources, so that the network does not belong entirely to a single enterprise or a single government.
Daniil uses the internet as an analogy to express his expectation for open, broadly participatory infrastructure.
He also believes that how people use and support AI services will influence whether this structure can take shape. Consumer demand is not isolated; it can support one infrastructure model and can also give another model room to develop.
Therefore, choices around AI are not just about "which model answers better," but can further ask:
What kind of network are we supporting? Who can participate in building it? Who holds decision-making power? And how does the value created by the network connect with contributors?
8. Involving More Women in Tech Also Gives the Future More Perspectives
At the end of the conversation, Daniil is asked whether more women should be encouraged to enter the tech industry.
His answer is an unequivocal yes.
He believes this is not only a matter of equality, but also a way to solve problems. Different people have different experiences, perspectives, and understandings of needs.
When products and technologies are shaped by a richer range of participants, people have a better chance of discovering neglected problems and a better chance of finding good solutions to those problems.
In his view, more diverse ways of thinking and perspectives are very important for finding technological solutions that can address human problems.
This also echoes his earlier discussion of decentralization: If AI will affect everyone, then the people involved in shaping it should not be only a small group with resources.
Conclusion: The Future of AI Should Be Discussed Not Only in Terms of Capability, but Also in Terms of Participation
Daniil's thinking runs through three levels.
First, access to advanced AI may become an increasingly important advantage, and this access depends on underlying compute.
Second, Bitcoin demonstrates the possibility of dispersed participants jointly building large-scale infrastructure, and AI compute can draw on its coordination and incentive ideas.
Third, productivity growth itself cannot answer the question of benefit distribution. To make technological progress bring broader abundance, it is necessary to discuss who participates in building, who controls resources, and who shares in new value.
Therefore, decentralized AI is not just another way to provide model services, but also an exploration of how future infrastructure should be organized.
AI can change everyone's life. What is truly worth continuing to ask is: Will people be merely users of this change, or can they also become its builders and participants?
Original video: Conversation between MaeveKnows and Daniil Liberman
