The demand for cost-effective computational power is rapidly increasing in emerging sectors such as AIGC, Web3, and the Metaverse. Computational power, often likened to the “oil” of future society, is a fundamental necessity. However, many individual data centers globally are not fully utilized, with the United States reporting an average utilization rate of merely 12% to 18%. This underutilization results in bottlenecks, leading to escalated prices for GPU computing. Consequently, several initiatives are underway to establish a decentralized cloud network that optimizes these dormant computing resources, with a specific emphasis on AI business applications. This article highlights some of the most esteemed projects in this field.
Overview of AI Model Training and Inference
The primary utilization of computational power is in AI model training and inference. AI model training involves the utilization of machine learning algorithms to build models using training datasets. Conversely, AI inference involves applying the trained model to new datasets to produce outputs or “predictions” based on the acquired knowledge during training. This process enables the comprehension of new data and facilitates real-time decision-making. Each project discussed in this article is assessed based on whether it prioritizes model training, inference, or both.
Akash Network (Established in 2017 – Training, Inference)
Akash Network, an innovative open-source supercloud network, provides a decentralized marketplace for cloud computing services. It enables users to securely purchase and sell computing resources through a “reverse auction” system, where customers can specify their desired price, prompting providers to compete for their business. This approach leads to competitive pricing. By leveraging technologies like Kubernetes and Cosmos, the network utilizes the Akash Token (AKT) for transactions and governance. It offers features such as unlimited storage, dedicated IP addresses, and high levels of privacy and security.
Cortex Labs (Established in 2017 – Inference)
Cortex is a decentralized, peer-to-peer blockchain platform that seamlessly integrates AI models. AI developers can upload their models to the blockchain and integrate them into smart contracts. Cortex has achieved significant milestones, including launching the world’s first AI on the blockchain and enabling on-chain AI inference. It is renowned for its innovative blockchain technology that facilitates AI integration and machine learning capabilities. Cortex empowers decentralized applications to leverage on-chain inference and fulfill the objectives of smart contracts.
Bittensor (Established in 2019 – Inference)
Bittensor is an open-source protocol that powers a decentralized blockchain-based machine-learning network. Similar to Bitcoin’s mining network, it provides censorship-resistant access to a decentralized network of machine learning models. Bittensor aims to establish a transparent market for AI models, creating incentives for the exchange between consumers and producers of these valuable models. Its emphasis on encouraging AI model participation and fostering value-creating markets sets it apart within the industry.
Gensyn (Established in 2020 – Training)
Gensyn is a machine learning computing network that aims to consolidate the world’s computational resources into a global supercluster accessible to all. It enables developers to train deep learning models across a network of interconnected devices, utilizing idle computing power in a verifiable manner. By utilizing blockchain technology to validate the completion of deep learning tasks, the network initiates token-based payments. Gensyn aims to democratize computing power through decentralized technology, providing AI developers with a cost-effective, scalable, and environmentally conscious solution.
OORT (Established in 2021 – Training, Inference)
OORT is an innovative decentralized data cloud designed to enhance privacy and cost-efficiency by integrating global computing and storage resources. It offers enterprise-grade decentralized cloud solutions tailored for generative AI and data-driven businesses. With a vast network spanning over 29,000 nodes across 100+ countries, OORT has prestigious clients and partnerships, including DELL, Lenovo Image, BNB Chain, Tencent Cloud, and Cardano. OORT’s ‘Talk to Data’ (TDS) service has significantly improved Lenovo Image’s customer support, elevating customer satisfaction rates from 30% to over 90%. OORT leads the transition of traditional non-crypto users to blockchain-based decentralized cloud solutions.
Io.net (Established in 2022 – Training, Inference)
Io.net has developed a decentralized physical infrastructure network to source GPU computing power for AI and machine learning applications. It simplifies infrastructure deployment by utilizing an open-source library, similar to the one used by OpenAI to distribute ChatGPT training across more than 300,000 CPUs and GPUs. Additionally, Io.net leverages Solana’s blockchain for transactions, rewarding machine learning engineers and miners for contributing computing power. In contrast to centralized services like AWS, Io.net offers a flexible approach, akin to Kayak in the airline industry, by facilitating the booking of compute resources.
