In today's AI ecosystem, computing infrastructure has become a critical bottleneck. Large centralized providers dominate GPU access, limiting opportunities for small teams and independent developers. In this context, Sjoerd Dijkstra, co-founder of Nosana, shares his vision on how decentralization can rebalance the market and open new opportunities for building more open and sovereign AI applications. This interview is part of the Judge Spotlight series for the Decentralize AI Hackathon, where we explore the work, perspectives, and advice of the judges for participants.
Nosana has emerged as one of the most relevant decentralized GPU networks, powering AI workloads in production. Dijkstra explains that it all started with a simple observation: there is a huge amount of underused computing power worldwide, while access to compute has become one of the biggest hurdles for developers. Initially, Nosana's focus was broader, centered on general distributed computing, but the rapid growth of AI made the opportunity much clearer. 'AI teams need reliable and affordable GPU access, but the market is still highly concentrated among a small number of large providers', he points out.
Nosana's vision has evolved from simply providing decentralized compute to building infrastructure that developers can genuinely use in production. Today they focus on making it easier to deploy, run, and scale AI workloads without forcing teams to rely entirely on centralized cloud platforms. This perspective resonates with what we promote at Q2BSTUDIO: helping businesses adopt AWS/Azure cloud services flexibly, combining the best of centralized environments with decentralized solutions when they bring real value.
One of the most surprising lessons for Dijkstra has been that technology is rarely the hardest part. 'The real challenge is making decentralized infrastructure feel as simple and reliable as the tools developers already use', he states. Builders don't want decentralization for its own sake; they seek better access, greater flexibility, competitive pricing, and less dependency on a small group of providers. This philosophy aligns with developing AI solutions that truly solve business problems, not just for tech trends. At Q2BSTUDIO we know that developer experience is key, whether implementing custom applications or integrating artificial intelligence into existing processes.
Dijkstra also shares his vision on the role of hackathons in advancing the ecosystem. He considers them one of the best ways to move an ecosystem forward because they encourage people to build, experiment, and test ideas in the real world. Decentralized AI is still a young space, and many of its most important use cases have probably not been discovered yet. Events like the Decentralize AI Hackathon give builders the opportunity to explore new models, infrastructure, agents, and applications without being limited by conventional assumptions.
When evaluating projects, Dijkstra first looks for whether they solve a real and clearly defined problem. 'A technically impressive product is interesting, but it becomes much more compelling when the team can explain who needs it and why'. He also observes how well the project uses the available technology. The strongest submissions don't add decentralized infrastructure or blockchain elements simply because they are part of the hackathon; they demonstrate why those components make the product better, more accessible, more resilient, or more open. Execution matters: even if the scope is small, a working product with a clear direction is usually more convincing than an ambitious concept that hasn't been properly demonstrated.
Regarding what separates a good submission from a truly standout one, Dijkstra notes that a good submission shows the team can build a working product, but a standout one demonstrates a clear and meaningful use of Nosana's decentralized GPU compute. The strongest projects don't just deploy a workload to the network; they show why decentralized compute is important to the product, whether by making AI inference more accessible, reducing reliance on centralized providers, supporting continuous agent workloads, or enabling an application to scale more efficiently. Clarity also matters: judges should be able to understand what is running on Nosana, how the compute is integrated into the architecture, and what role it plays in the user experience. A strong live demo, supported by clear deployment details and evidence of real GPU usage, can make the project much more convincing.
Dijkstra warns about common mistakes, such as trying to do too much. Hackathons have limited time, so it's usually better to deliver a focused and functional product than a very broad platform with several unfinished features. Another mistake is failing to explain the technical architecture. Judges shouldn't have to guess how the project works, what was actually built during the hackathon, or how the infrastructure is being used. He also recommends testing the full experience before submitting: make sure the links work, the repository is accessible, the instructions are clear, and the main use case can be demonstrated without unnecessary friction.
Looking ahead, Dijkstra believes decentralized AI can play an important role in making compute more sovereign, open, and resilient. Today, much of the AI ecosystem depends on a small number of large cloud providers, giving those companies significant control over pricing, access, infrastructure availability, and the conditions under which AI products can be built. Overreliance on these providers creates strategic risks for developers, companies, and even entire regions. Decentralized networks can offer an alternative by distributing compute across independent providers and giving builders more choice over their infrastructure. 'This doesn't mean centralized clouds will disappear, but it can create a healthier and more competitive market', he says.
Among the areas that most excite Dijkstra for builders to explore are AI agents. 'Agents create ongoing demand for compute rather than a single training or inference task'. They need infrastructure for deployment, execution, monitoring, evaluation, and coordination, and there is still plenty of room for innovation across that entire lifecycle. He is also interested in open AI services that can be deployed across distributed infrastructure, including inference APIs, model marketplaces, privacy-focused applications, and tools for communities or languages currently underserved by mainstream platforms. Another important area is infrastructure orchestration: making distributed GPUs easier to discover, benchmark, schedule, and manage will be essential if decentralized AI is to scale.
Finally, Dijkstra offers direct advice to participants: use the time to build something that could exist beyond the hackathon. A ten-month program gives the opportunity to go much further than a prototype. The strongest teams will validate the problem, test their product with real users, improve it through multiple iterations, and show that the project can grow into something genuinely useful. Judges will value clarity, but they will also look for depth of execution. Show how the product evolved, what you learned, how Nosana's compute supports the core use case, and why the project has long-term potential. A strong submission should not feel like a weekend experiment, but rather the beginning of a serious product.
At Q2BSTUDIO, as a software and technology development company, we understand the importance of integrating solutions like custom applications with AI capabilities, cybersecurity, cloud, and business analytics (BI/Power BI). Decentralization opens new avenues for building more robust and ethical systems, and we are confident that initiatives like the Decentralize AI Hackathon will drive the next generation of open AI infrastructure. As Dijkstra says, don't be afraid to experiment. Decentralized AI is still being defined, and builders have a genuine opportunity to influence what the ecosystem will become.
