While the global debate on artificial intelligence focuses on the speed of development and technological capabilities, the question of the infrastructure on which AI rests is raised far less frequently.
Daniil and David Liberman, co-creators of the Gonka protocol, futurists, entrepreneurs and investors, in an interview for NIN open up the issue of the centralisation and decentralisation of AI as a strategic choice with long-term economic, social and geopolitical consequences.
Most of the world today speaks about centralised AI infrastructure. Why and how did you decide to move in the completely opposite direction – towards decentralised AI infrastructure?
Centralised AI infrastructure appears inevitable only if its underlying assumptions are not questioned. For a long time, discussions about artificial intelligence have focused on models, ethics or regulation. However, beneath all of that lies a more decisive layer – computing power. Who owns it, who controls access to it and under what conditions it can be used ultimately determines how AI functions and whom it serves.
When you look at AI from that perspective, the current picture becomes difficult to ignore. OECD research and other publicly available data show that advanced AI computing infrastructure is increasingly concentrated in the hands of a small number of cloud providers and within a limited number of countries. This has created a growing gap in computing infrastructure between those who have access to the foundational layer and those who do not.
This concentration is not accidental. Today, access to advanced graphics processing units is controlled by a small number of providers and is increasingly shaped by state priorities. The United States and China are actively consolidating control over chips, energy resources and data centre capacity, effectively determining where large-scale artificial intelligence can be built and who can build it. The result is computing power that is expensive, limited in capacity and geographically unevenly distributed – precisely at the moment when AI is becoming the foundation of science, industry and society.
At the same time, decentralisation in its current form has not automatically solved this problem. Many decentralised systems still consume a large share of available computing power on consensus and security mechanisms, while incentives often reward capital rather than actual computational contribution. This discourages hardware suppliers and slows innovation at the infrastructure level.
This is where our thinking diverged. We did not approach decentralisation as an ideological position or as a reaction against centralised actors. We arrived at it by asking a practical question: what would AI infrastructure look like if efficiency, access and contribution were aligned rather than in conflict?
That question led us to a model in which most computing power is directed towards actual AI work rather than technical overhead; in which participation and governance are based on verified computational contribution rather than capital; and in which access to global GPU resources is open and permissionless. In practice, many of these assumptions are being tested through continuous, open dialogue with GPU operators, system builders and researchers collaborating in real time, including within our Discord community.
Artificial intelligence has never been merely software. It has always functioned as infrastructure. And infrastructural decisions tend to set societies on trajectories that last for decades. Establishing that foundation within a small number of corporate or national jurisdictions is not a neutral technical outcome; it is a systemic decision with long-term economic and geopolitical consequences. If intelligence is conceived as something abundant and accessible, the infrastructure that supports it must be designed for abundance from the outset.
How difficult is it to build something that goes against the beliefs of dominant technological communities? From your perspective, what does success look like?
It is difficult above all because you are not arguing with people, but with default settings. Dominant technological communities usually optimise what works in the short term: speed, capital efficiency, centralised control and growth through consolidation. Those decisions make sense in context, but once they become the default, they are rarely questioned. When you challenge them, it can seem as though you are speaking a different language, not because the ideas are extreme, but because they touch the incentives on which many careers, companies and strategies have already been built.
What makes this even harder is the moment in which everything is happening. Centralised systems often appear successful long before their long-term costs become visible. Although enormous investments and infrastructure costs are already evident, deeper consequences usually emerge later in the form of dependency, loss of flexibility, the transfer of pricing power to a small number of providers and the inability to change direction once systems become deeply embedded.
For us, success does not mean winning an argument or replacing existing players. It looks much quieter than that. Success is the moment when decentralised infrastructure ceases to be a stance or a message and becomes something ordinary: when people use it not because they believe in decentralisation, but because it is simply the most practical option available. Ultimately, success arrives when the very subject of the conversation changes – when the question is no longer whether intelligence should be centralised, but why we ever assumed it had to be. At that point, beliefs no longer need to be directly challenged; they evolve on their own.
Why is it important for states and companies to think about decentralised artificial intelligence? At what point should governments and companies decide between a centralised and a decentralised approach?
Because AI infrastructure is no longer just a technical layer; it is becoming a strategic dependency. For both states and companies, centralised AI infrastructure creates lock-in that is difficult to escape. When key systems begin to depend on a small number of providers, control gradually shifts from users to infrastructure owners. Over time, this affects pricing, availability, the pace of innovation and the scope of real strategic choice.
For states, this is increasingly a matter of sovereignty and resilience. Today, artificial intelligence underpins public services, scientific research, defence and economic planning. Full reliance on external, centralised infrastructure introduces single points of failure – technical, political and geopolitical – which often become visible only in moments of crisis. In the conversations we have with governments, control over computing resources is increasingly viewed as a key component of AI sovereignty, alongside energy and data.
For companies, this is a question of strategic flexibility. Centralised infrastructure may function extremely well in the early stages, but over time, it often turns into long-term dependency. Costs become harder to control, alternatives more complicated to introduce and architectural decisions more difficult to change at scale.
The key moment for making the decision comes earlier than most expect. Infrastructure decisions become locked in long before their consequences are obvious. Once AI moves from the experimental phase into routine use, changing the underlying architecture becomes exponentially more expensive. The right time to decide is not when centralised systems begin to fail, but while they still appear to function perfectly. Early consideration of decentralised approaches preserves the possibility of choice; waiting often means the decision has already been made, just without conscious intent.
Is it too late if most systems have already been built on a centralised infrastructure?
It is rarely truly too late, but the longer centralisation functions as the default model, the more exponentially difficult change becomes. When most systems are built on centralised AI infrastructure, the challenge is no longer technical; it becomes institutional. Workflows, incentives, budgets, compliance requirements and even talent development begin from the assumption that centralisation is simply how things work. At that point, change no longer means merely migrating infrastructure, but unlearning habits, contractual models and ways of thinking that have become deeply embedded over time.
Research on infrastructure lock-in confirms this. Industry analyses consistently show that after several years operating in centralised cloud environments, switching costs rise sharply – not linearly, but in leaps. The reasons include long-term contracts, regulatory frameworks, deeply integrated internal processes and workforce specialisation. OECD research further indicates that states and organisations without early access to AI computing resources risk falling into a vicious cycle of lagging behind, losing not only competitiveness but also architectural freedom – the real ability to choose alternative infrastructure models.
At the same time, history shows that infrastructure transitions almost never happen all at once. They typically begin at the margins of systems. New use cases, new actors and new constraints create pressure points where centralised systems cease to be sufficient – becoming too expensive, too slow, too restrictive or too fragile. It is precisely there that alternatives begin to gain practical significance.
What is truly lost over time is the possibility of choice. The longer centralised infrastructure dominates, the fewer real options remain available. Dependencies harden, and decentralisation shifts from a proactive design decision to a reactive correction that is always more expensive, more complex and harder to control. The real risk is not that it is already too late. The real risk is waiting until decentralisation is no longer a choice but a necessity imposed by systemic failure. The earlier it is considered, even in parallel with centralised approaches, the more room there is to shape outcomes thoughtfully rather than under pressure.
What should future generations know about this issue?
They should understand that technology does not become neutral simply because it becomes more advanced. Every generation inherits infrastructural decisions made before it, often without awareness that those decisions were once deliberate choices rather than inevitabilities. For future generations, artificial intelligence will feel as natural as electricity or the internet do to us today. That is precisely why the underlying architecture matters so much: it shapes not only what is possible, but for whom it is possible.
Future generations should know that access to intelligence can be organised in fundamentally different ways. It can be treated as a shared foundation – open, accessible and difficult to appropriate. But it can also be closed, metered and controlled, even when at first glance it appears practical and efficient. Both approaches can produce impressive technology, but only one preserves long-term freedom, resilience and genuine choice.
It is also important to understand that centralisation most often arrives quietly. Not by force, but through convenience. Early compromises seem harmless: slightly lower costs, faster implementation, simpler coordination. The consequences appear later, when changing course becomes prohibitively expensive or practically impossible.
It is equally important to recognise that infrastructure directly affects social mobility. Systems that appear technically neutral can either reduce unequal starting positions between people and generations or subtly entrench them for decades to come. As you likely know, this is also a major issue for us. Younger generations today already face significantly worse starting positions compared with previous generations at the same stage of life. Artificial intelligence in its current form does not resolve that problem and has the potential to deepen it further. In that sense, architectural decisions determine not only efficiency, but who truly gets the opportunity to experiment, build and shape the future.
Most importantly, future generations should know that these systems are still designed by people. Not by fate, not by the market and not by the machines themselves. Questioning default assumptions, asking who benefits from a particular architecture and insisting on preserving the possibility of choice is not resistance to progress. It is a way to ensure that progress remains open.
Why did you decide to join the Unlockit conference and share your story?
Unlockit seemed to us a space where the conversation is not centred on hype, launches or predictions, but on why people make certain decisions in the first place. That was important to us. Our story is not really about a single project or technology, but about recognising structural patterns in time and choosing not to accept them as inevitable.
For years we operated within dominant systems – building companies, investing, working with large organisations and using the advantages of centralised infrastructure. We understand how those systems function from the inside. At a certain point, it became clear that repeating the same structures while hoping for different outcomes rarely leads to anything fundamentally new. Sharing that insight openly felt more honest than remaining silent or packaging it into yet another “success story”.
At the same time, we did not come to Unlockit only with reflection, but with concrete, practical experience that may be relevant to different people in the room. For founders, these issues translate into control over infrastructure, dependency on providers and the ability to grow without losing flexibility. For investors, they concern long-term risk, infrastructure lock-in and models that genuinely create sustainable value. For corporations and technology leaders, the questions involve cost structures, reliability, regulatory constraints and strategic freedom in a rapidly changing environment.
We wanted to share an approach that already functions as a practical alternative – not as a universal solution, but as a different way of thinking about how AI infrastructure can be built with less dependency, greater transparency and more long-term choice. Equally importantly, we wanted to hear feedback from people who make real decisions with real consequences for businesses, capital and institutions.
We also believe that such conversations should not remain confined within a narrow circle of insiders. The moment infrastructure decisions cease to be publicly examined, they quietly become defaults. Unlockit creates space to reflect on those choices before they become irreversible, and that is precisely what makes participation in this conversation worthwhile. In the end, joining the Unlockit conference was less about explaining what we do and more about explaining why it remains important to question default assumptions – especially at a moment when technological progress appears rapid, powerful and inevitable. It was equally about listening to the perspectives of those actively shaping the future of business, technology and social systems.
(NIN, 26.02.2026)