Decentralized GPU networks promise scalable and affordable AI: AMA with Nosana

AI relies on computing power like our brains on oxygen – without it AI systems simply cannot function and improve. Popular AI computing solutions rely on cloud platforms like AWS and Google Cloud or companies using their own GPUs and TPUs. Although offering flexibility and scalability, they can also be costly and technically challenging, especially for startups.

🚨 Missed our AMA with @nosana_ai?

Watch the recap now – find out how decentralized GPU networks drive costs down for devs & businesses and how can you jump in as a GPU host.

[Brought to you by Nosana]https://t.co/P4vtUQbzzL

— Cointelegraph (@Cointelegraph) February 11, 2025

Decentralized compute networks may provide a favorable alternative, according to Sjoerd Dijkstra, co-founder of Nosana. He outlined the advantages during the Cointelegraph live AMA.

Launched in January 2025, Nosana’s GPU marketplace allows developers to access the power of a decentralized GPU network for training and inference of their AI models while allowing GPU owners to rent out their hardware.

“We’re tapping into underutilized resources around the world, making it much more affordable for companies and developers who need scalability without breaking the bank,“ Dijkstra said. “Scalability is another reason. Also, by processing data closer to the source, we minimize delays and speed up AI testing.“

He continued: “Decentralized GPUs have a wide range of applications in different industries, and they are particularly excellent for AI inference tasks. In Web3, we see a lot of platforms with large language models and image generation systems that really benefit from these enhanced computational capabilities of decentralized GPU networks.“ Other applications Dijkstra sees include healthcare for drug discovery and diagnostics, automotive for autonomous vehicles, finance, and AI agents.

Building AI models from scratch can be time and resource-consuming, so the platform offers a set of templates with pre-built AI models that support LLMs for different needs. “Businesses can use our templates as a starting point, tailored to common industry needs,“ Dijkstra emphasized. “These are state-of-the-art AI solutions that are already optimized for your specific use case.“

He mentioned some active projects benefiting from Nosana’s resources: a platform for the creative industry Sogni AI has reduced costs and scaled operations, Ocada runs AI agents integrated with Nosana’s computing power to optimize their application layer, a decentralized marketplace for AI models and data sets AlphaNeural has accelerated model deployment.

Nosana’s network is supported by a mix of independent GPU owners and large providers: approximately 75% of the GPUs in the network are from individual contributors worldwide, with the remainder from established GPU providers such as Render and PikNick, which enhance Nosana’s security, reliability, and performance.

“They make Nosana’s network not only decentralized but also enterprise-ready, which allows us to serve a broader range of clients, from startups to larger organizations that need robust and compliant compute solutions,“ said Dijkstra.

GPU hosts on Nosana are required to stake $NOS to ensure that only trusted participants contribute compute power. They receive incentives as well as performance-based bonuses.

“We’re just scratching the surface of the AI revolution, and decentralization is key to making AI more accessible,“ Dijkstra concluded the conversation. “But we should keep some important points in mind - ensuring security and trust, setting standardized protocols and APIs, and using scalable solutions, which is what we’re doing at Nosana. That’s how we can pave the way for a more accessible and efficient AI landscape.“