Akash Network Enables Decentralized GPU Access
Akash Network launched early access for Homenode, a platform for bringing decentralized compute to everyday devices. The initial phase allows individuals with high-end GPUs, such as the RTX 4090 or 5090, to contribute their compute power to the network. The initiative aims to expand access to compute resources for AI developers and enterprises, addressing a key bottleneck in the industry.
- Akash Network was founded in 2015 by Greg Osuri and Adam Bozanich. Before Akash, Osuri founded AngelHack, a global hackathon organization with a developer ecosystem of over 200,000 people across 164 cities. - The network functions as a peer-to-peer, open-source cloud computing marketplace where individuals and data centers can lease their underutilized computing resources. Those needing compute power, the "tenants," are connected with "providers" who have the capacity to lease. - Decentralized GPU networks are often 60-86% cheaper than traditional centralized cloud providers like AWS, Microsoft Azure, and Google Cloud, which is a significant draw for startups and individual developers. For example, an RTX 4090 on Akash can be found for around $0.40/hour, while H100 GPUs can be leased for as low as $1.20/hour. - The company has raised a total of $2.8 million over three funding rounds. Its investors include Digital Asset Capital Management, P-OPS Team, and Blockin.Ventures. - The network's roadmap for 2026 includes several key milestones, such as implementing stable pricing and smart contracts, launching support for Virtual Machines, and introducing confidential computing to attract enterprise clients with sensitive data. - Competitors in the decentralized GPU marketplace include io.net, Render Network, and Fluence. The broader decentralized physical infrastructure networks (DePIN) industry has an estimated market capitalization of over $19 billion and is projected to reach $3.5 trillion by 2028. - Greg Osuri, the CEO, has a background as an open-source developer with over 25 years of experience and has contributed to widely used technologies like Kubernetes and Terraform. He also provided expert witness testimony that was instrumental in passing California's first blockchain law. - While decentralized networks are cost-effective for AI inference and tasks that can be broken down, they are generally not suitable for training the largest AI models, which require thousands of GPUs to be tightly synchronized in a single location.
