Theta Labs Launches Hybrid Edge Cloud Architecture with Dynamic Supply-Demand GPU marketplace to Democratize Access to Computing

New decentralized system delivers enterprise-grade AI computing at up to 70% lower cost by intelligently combining cloud and distributed resources

Theta Labs today announced the beta release of its hybrid edge cloud architecture for its Theta EdgeCloud network, a computing platform that combines traditional cloud-based GPUs with a distributed network of over 30,000 community-operated edge nodes. The platform is designed to provide cost-effective access to high-performance computing resources for AI model training and inference, video processing, financial modelling, and other GPU-intensive tasks. In this new release, a new decentralized GPU marketplace is introduced that keeps compute pricing competitive and transparent across the platform.

The hybrid architecture addresses a fundamental challenge in modern computing: the rising cost and limited availability of specialized hardware needed for AI and machine learning tasks. Traditional cloud providers charge premium rates for GPU access, often pricing out smaller research teams, startups, and academic institutions. By integrating distributed computing resources from community members alongside conventional cloud infrastructure, the platform will provide similar capabilities at significantly reduced costs.

A dynamic GPU marketplace for efficient compute routing

Theta EdgeCloud is a decentralized marketplace that connects the supply and demand for GPU computing power. It empowers anyone with idle GPUs to contribute their resources and earn rewards, while providing developers and AI teams with a scalable, cost-efficient platform for running containerized workloads.

The platform now allows customers to choose the most suitable infrastructure for different types of computing tasks. For instance, training a large AI model that requires substantial GPU vRAM can be directed to powerful cloud or data center based GPUs. In contrast, tasks like burst model inference workloads, which are inherently parallelizable, can be distributed across many community-operated gaming machines, providing a flexible and cost-effective alternative.

To ensure a fair and dynamic pricing model, Theta EdgeCloud allows node operators (the supply side) to set their own hourly rental rates. Meanwhile, users (the demand side) can select nodes that meet their performance requirements and budget constraints when launching workloads. This market-driven approach helps keep GPU compute pricing competitive and transparent across the platform.

The system includes backup mechanisms with automatic failover logic to reassign work if any community device goes offline, ensuring reliable completion of computing tasks across the heterogeneous network of GPU types.

Powering Leading Academic and Enterprise Customers

The platform currently serves customers including Stanford University, Seoul National University, KAIST, Yonsei University, the University of Oregon, Michigan State University, and NTU Singapore for academic AI research. Enterprise clients include major sports teams such as the NHL’s Las Vegas Knights, NBA’s Houston Rockets, and global esports teams FlyQuest and Evil Geniuses.

The beta release includes features requested by existing customers, including persistent storage for AI model training, improved job prioritization, and a developer API interface for job submission and analytics.

“The reality is that GPU costs have become prohibitive for many organizations doing important AI research,” said Jieyi Long, CTO of Theta Labs. “Universities are telling us they’re having to scale back projects or wait months for access to affordable computing resources. By tapping into the unused GPU power sitting in gaming computers and workstations around the world, we can deliver the same computational capabilities at a fraction of the cost. This means our partners at Stanford, KAIST, and other institutions can run more experiments, iterate faster, and push the boundaries of what’s possible in AI research without budget constraints limiting their ambitions.”

Technical Capabilities

The hybrid architecture supports containerized computing tasks including AI model training and inference, video encoding and transcoding, 3D rendering, financial simulations, and scientific computing applications. The platform provides over 80 PetaFLOPS of distributed GPU compute power through its combination of cloud partnerships with Google Cloud and Amazon Web Services and its distributed edge network.

The EdgeCloud client node is a lightweight software package that enables community members to contribute their idle GPU capacity. Advanced job containerization ensures high-efficiency computation across different GPU types and specifications. To learn more about the EdgeCloud client software, please check out this link.

About Theta Labs

Theta Labs is the leading provider of decentralized cloud infrastructure for AI, media and entertainment powered by a global network of 30,000 distributed edge nodes and a native blockchain. Backed by Samsung, Sony, Bertelsmann Digital Media Investments and Creative Artists Agency, Theta is among the top 10 DePIN blockchains by market capitalization on Coingecko and top AI tokens on Binance.com. Theta’s enterprise validator and governance council is composed of global market leaders including Google, Samsung, CAA and Binance.

Recently launched Theta EdgeCloud is the first hybrid cloud-edge computing AI platform with over 80 PetaFLOPS of on-demand distributed GPU compute power. EdgeCloud now counts 25 global customers including 4 of the top 5 South Korea universities, top professional sports teams including NHL’s Las Vegas Knights, NBA’s Houston Rockets and global esports teams FlyQuest and Evil Geniuses among others.

Source