Every leap in technology is reshaping the world, and in recent years, leapfrog technological innovation has once again triggered profound changes, that is, artificial intelligence (AI).
Artificial intelligence has surpassed all previous technical concepts and has become the next huge wave after the Internet. From self-driving cars to personalized recommendations, from machine learning to generative large language models, AI has gradually penetrated into every aspect of our lives, and has also triggered extreme social attention and investment booms.
Industries and enterprises are actively looking for ways to use AI to improve efficiency. They all understand the value of AI and look forward to using it to improve their productivity and competitiveness. However, the application of AI is not easy, and enterprises must initially invest a lot of resources, including time, manpower and funds, to build and maintain it.
Different from the traditional sublinear relationship between R&D investment and product value, the development of AI shows a unique characteristic: increasing computing resources directly leads to improved product performance. For example, the training of large models, AI-assisted development in the gaming field and cloud gaming all require massive computing resources. The biggest challenge is the high cost of obtaining computing power.
In the calculation process of AI models, GPU (Graphics Processing Unit) plays a core role. At present, the supply of efficient GPUs is almost monopolized by giants such as Nvidia, making it difficult for small and medium-sized enterprises to obtain the required computing resources. Although many companies are eager to try AI empowerment, in fact, the gap between supply and demand is widening.
According to CAPEX analysis and forecasts, in 2024, the GPU procurement demand of the four major cloud vendors in North America (Amazon, Microsoft, Google, and Meta) alone will reach 3 million pieces. Due to the imbalance between supply and demand, GPU prices are difficult to drop and delivery cycles continue to lengthen. Many companies’ spending on computing resources exceeds 80% of their total capital investment in the AI field. The shortage of computing power has become a key factor restricting the development of the industry.
Tesla founder Musk has publicly stated that with the rapid development of artificial intelligence technology, especially in the fields of deep learning and natural language processing, the demand for computing power has shown explosive growth. However, the current global supply of AI computing power is difficult to meet this demand, which will have a profound impact on the development and application of AI technology. The problem of AI computing power shortage exists not only in the business field, but also involves important fields such as security and medical health. Therefore, solving the problem of AI computing power shortage is not only the responsibility of technology companies and research institutions, but also a challenge faced by the whole society.
Based on the above problems, computing power leasing (cloud computing power) has become the key to solving the problem. The lower upfront cost, the ability to scale up and down, regional availability, and avoiding the distraction of building your own data center are very attractive to most startups and even some large companies. Different organizations and companies can rent computing power on demand without having to bear the heavy burden of purchasing hardware and maintaining it.
There are many cloud computing service providers on the market, such as Amazon Web Services (AWS), Microsoft Azure and Google Cloud Platform (GCP), which all provide GPU instances, but these international giants often charge high prices due to their market monopoly. Although there are some new service providers that specialize in AI workloads, they are often not very competitive.
In particular, the United States and other leading countries in the field of AI have restricted the export of top chips and the supply of computing power in order to protect their own advantages. As a result, customers in some countries and regions are unable to purchase enough GPUs or obtain services from service providers.
Aethir: Building decentralized cloud infrastructure for the AI era
As a new generation of cloud computing solution provider, Aethir adopts AI cloud computing + GPU DePIN solution to solve the shortage of computing power in the existing AI field. Aethir provides efficient, scalable and flexible computing power rental services that match demand for enterprises and individuals around the world by establishing a new, distributed, AI-based cloud computing network.
Aethir can optimize GPU utilization in compute-intensive fields such as AI, ML, and cloud gaming: First, through resource pools, owners can contribute underutilized GPUs to the network to form a powerful collective pool of computing resources, achieve global distribution of GPUs, reduce costs, and democratize advanced computing capabilities; second, through decentralized ownership, transcend the limitations of traditional ownership structures, achieve distributed resource possession, cultivate a fair and open technology landscape, eliminate barriers for new consumers and entrepreneurs to use AI, and contribute to the globally interconnected digital ecosystem.
Simply put, in the current situation of global computing power shortage and GPU supply exceeding demand, how to make good use of idle GPU resources is the key. Through the DePin operation model, Aethir encourages users or nodes to contribute their idle GPU computing power and achieve scale, providing computing power support for enterprises in need and meeting the demand for computing power.
Technical Architecture & Token Economy
We all know that AI large-scale model training requires the use of "a whole block of high-performance computing power", and the current solution is to use multiple GPUs with high-performance cards connected to each other to form the required computing power. However, in contrast, civilian computing power is often scattered, and network conditions are generally poor, which makes it a difficult problem to make scattered computing power efficiently serve AI training.
In this environment, Aethir needs to solve two problems to meet the needs: how to ensure the quality of computing power and how to solve network problems. With the help of Aethir's own H100 computing power cluster, as well as excellent architecture and token economic model, the two problems have been effectively solved.
NVIDIA's H100 GPU is a core element of Aethir's decentralized cloud infrastructure. More than 4,000 H100s are available for on-demand use by AI enterprise customers, and it is expected that the platform will add thousands more NVIDIA H100s and deploy a large number of NVIDIA H100s in the next six months.
Each H100 added to the Aethir network has undergone rigorous testing and screening, including performance parameter configuration, model usability, bandwidth throughput, stability, etc., to ensure its performance in high-speed training and inference tasks.
And Aethir’s advantage goes beyond the number of available H100s. Traditional cloud computing services concentrate GPU resources in centralized server centers, making it impossible to effectively transmit GPU power to clients far away from data centers. On the other hand, thanks to Aethir’s distributed network infrastructure, customers at the edge of the network in most parts of the world can be effectively covered. Each client is served by the closest available H100 chip, eliminating latency issues.
In addition to the distributed deployment of H100 to ensure the stability of computing power, Aethir's architecture and economic model ensure the quality of scattered computing power and the stability of the network.
Aethir's architecture design includes five basic roles: miners, developers, users, token holders, and Aethir DAO. The three core parts are Container, Indexer, and Checker:
Container is the core computing unit of Aethir, responsible for executing and rendering applications. Each task is encapsulated in an independent Container. Each Container runs the user's tasks as a relatively isolated environment to avoid interference between tasks. If there are computing resource-intensive applications such as big data processing or machine learning, they can be successfully executed in the Container and finally get the results. That is, the required high-performance computing power is achieved.
Indexer is mainly used to match and schedule user task requirements and available resources in real time. Real-time matching and scheduling are to ensure that user needs can be met in the shortest possible time, and the fault tolerance and redundancy design is to deal with possible service failures. Other backup nodes can be selected for task scheduling to prevent task progress from being interrupted. At the same time, dynamic resource adjustment can dynamically allocate resources to different tasks according to the system load to achieve the goal of optimizing overall performance.
Checker is responsible for real-time monitoring and evaluation of the performance of the container. It can monitor and evaluate the status of the entire system in real time and respond promptly to possible problems. If you need to respond to security incidents such as network attacks, you can issue warnings and initiate protective measures after detecting abnormal behavior. Similarly, when bottlenecks or other problems occur in system performance, Checker can also issue reminders in a timely manner so that the problem can be solved in a timely manner, ensuring service quality and network security.
At the same time, Aethir has established a strict node reward and punishment mechanism, which rewards high-quality nodes that meet the quality standards and imposes economic penalties on nodes that do not meet the service quality standards, ensuring the stability and availability of the entire network. This mechanism effectively protects customer rights and interests and improves the service awareness of nodes.
Aethir's token economy is centered on the ATH token, which is used for transactions, platform governance, incentives and development of cloud computing services. The total number of tokens is 42 billion. By purchasing GPU computing power and staking mechanisms, the ATH token promotes the decentralized management and growth of the ecosystem, while providing a commitment and economic security for new node operators and users.
The key to the token economic model lies in its consensus mechanism (Proof of Rendering), which consists of two main components: Proof of Rendering Capacity and Proof of Rendering Work.
Proof of Rendering Capacity: This strategy conducts a targeted evaluation for each Container, where its token investment (the number of ATH tokens staked by the node), computing power level, and online time are taken into account to estimate the effective computing power of the Container. This mechanism enables all nodes to participate in the network fairly and encourages node operators to stake more ATH tokens.
Proof of Rendering Work: When the container is providing computing services, the Checker will monitor the work status and submit the specific service conditions (such as latency, resolution, frame rate, etc.) to the chain. Proof of Rendering Work rewards are allocated based on the quality of work and working hours.
Aethir's service fee is paid in ATH tokens, and the price is anchored to the fiat currency to ensure the stability of the service fee. If the service is interrupted due to a Container failure, the user will receive a corresponding refund, and the Container that failed will be fined.
In order to meet the needs of customers of different sizes and needs, Aethir has also designed two operating models: Retail and Wholesale. The wholesale model sacrifices a certain degree of flexibility to provide lower service fees, a preference for long-term service guarantees and transaction settlement convenience. The retail model is more flexible and can provide services at any time according to demand without reservation or commitment.
Aethir has introduced a fiat currency pricing mechanism, allowing customers to pay for computing power services in fiat currency, greatly reducing its entry threshold and financial risks, and enhancing customer stickiness and compliance.
In general, Aethir, through its excellent architecture and token economic model, has the potential to fully mobilize high-quality scattered computing power from the perspective of technical architecture and system, solving the problem of high-performance computing power required for large-model training, which is the largest AI demand scenario in the current business environment. This makes Aethir extremely commercially available in large-model training scenarios.
Application in the field of games
In the past few years, we have witnessed the widespread application of AI in the gaming field, such as novice guidance and NPC text generation. With the development of technology, the application of AI in the gaming industry will be more in-depth in the future, requiring a lot of computing power for training and learning. In addition to the sharing of AI computing power, cloud gaming is another area where Aethir excels.
In the 20 years of rapid evolution of GPU and storage technology, as the computing power of home PC devices has increased, developers have been pursuing more and more game graphics and game space. For example, GTA uses computing power to generate cities, and Need for Speed uses computing power to improve car collisions and image quality, which has led to the birth of exquisite game graphics and super-large and realistic cities. The final result is that the capacity of hundreds of GB and the requirements for graphics card performance are difficult for a large number of players' computers to load.
Cloud gaming refers to a gaming method based on cloud computing. In the cloud gaming mode, all games are run on the server side, and the rendered game screens are compressed and transmitted to users through the network. On the client side, the user's gaming device does not require any high-end processors and graphics cards, only basic video decompression capabilities are required.
Aethir brings significant value to the cloud gaming field through its distributed network, including providing low-cost high-end GPU computing power, optimizing the gaming experience to achieve instant access to devices and low latency, serving game developers and providing new publishing methods and Game transplantation services expand the player base, improve game upgrade efficiency, and improve game security.
Ecological Development
Aethir and IO.net have reached a strategic partnership. The two have jointly conducted technical research and development and docking to open up the connection between Aethir's H100 and the IO network. In this way, Aethir's H100 can automatically connect to the IO network and provide stable enterprise-level services to IO network customers. Through joint technical research and development and docking, Aethir's H100 can automatically join the IO network and provide stable enterprise-level services to IO customers. Aethir's H100 providers will be able to receive token rewards from the IO network while receiving Aethir rewards.
At the same time, Aethir is promoting the integration of cluster and edge computing. Simply put, through Aethir's edge computing, customers can match the nearest node to ensure the smooth flow of computing power and network.
Aethir Edge is a hardware device designed for Aethir edge computing services. It will break the single centralized GPU cluster deployment mode far away from users and deploy computing power to the edge. In this way, customers far away from centralized server clusters can also enjoy stable and seamless GPU cloud computing services.
APhone is a Web 3.0 cloud phone launched by Aethir. It uses Aethir's distributed cloud architecture to achieve a secure, device-independent Web3 communication experience that transcends the geographical boundaries of communication service providers. As part of the Aethir network, APhone not only supports seamless dApp access, but also integrates the Web 3.0 application store. As of now, the number of Aphone users exceeds 36,000.
Summarize
As AI becomes an important technological innovation, the shortage of AI computing power has become a major obstacle to the development of the industry. As an emerging cloud computing solution provider, Aethir aims to solve this problem by establishing a distributed computing network to provide efficient and scalable computing power for global enterprises and individuals.
Aethir's architecture design and token economic model, as well as its applications in the gaming field and edge computing, all demonstrate its potential in terms of technical architecture and business layout. Aethir ensures the stability and availability of the network through an incentive mechanism, providing commercial-level solutions for fields such as AI and gaming.
Overall, Aethir has performed well in terms of project fundamentals, technical architecture, products and ecology. Its token ATH is about to go TGE, which is worth our attention.
