In-Depth Analysis of Aethir: A Decentralized Cloud Computing Player with Triple Track Strength

The development and progress of LLM large models and AI is one of the greatest technological advancements in human history. Humanity has now entered the AI era, where "computing power" is the most scarce resource in this new world.

The trend in computing power development is edge computing, which can effectively reduce physical latency and become the cornerstone for the development of industries with low latency requirements, such as the metaverse; decentralized distributed cloud computing has advantages in flexibility, low cost, and resistance to censorship, and its development prospects are very broad.

Aethir is a decentralized real-time rendering platform based on the Arbitrum network, providing enterprise-level computing power services for industries such as gaming and artificial intelligence by pooling high-performance GPUs like the H100.

Aethir has partnered with top cloud computing projects in the industry such as io.net and Theta, as well as several leading gaming studios and telecommunications companies, with an expected annual recurring revenue (ARR) exceeding $20 million in the first quarter of 2024.

Aethir Edge significantly lowers the threshold for ordinary users to sell excess computing power and greatly expands the geographical coverage of the Aethir network.

Aethir has already raised $80 million by selling inspector node NFTs, proving that its project prospects and economic model are highly attractive to a wide range of users.

The hourly usage cost of Aethir network's A100 is significantly lower than that of other competitors, providing a clear competitive advantage.

Background of Aethir

The changes in the development process of human society are often realized by a few extremely great scientific inventions and progress. Each breakthrough in technology directly creates a more efficient and prosperous new era.

The Industrial Revolution, the Electrical Revolution, and the Information Revolution are extremely significant technological advancements in human history. They have completely transformed the face of human society, bringing unprecedented changes in productivity and lifestyle. Now, we can no longer return to the era of kerosene lamps for lighting and horse-drawn carriages for delivering letters. With the birth of GPT, humanity has entered another great new era.

LLM is gradually liberating human intelligence, allowing people to use their limited energy and intelligence for more creative thinking and practice, leading humanity into a more efficient world.

We view GPT as another technological breakthrough that changes the world, not only because of GPT's tremendous advancements in natural language understanding and generation, but also because humanity has grasped the规律 of the growth in capabilities of large language models through the evolution of GPT—that is, by continuously expanding model parameters and training data, the capabilities of LLM models can be increased exponentially, and in the presence of sufficient computing power, this process currently shows no signs of bottleneck.

The use of large language models is not limited to understanding human language and dialogue; rather, this is just the beginning. Once machines have the ability to understand language, it is like opening a Pandora's box, releasing infinite imaginative possibilities. People can leverage this capability of AI to develop various disruptive functions.

Currently, in various intersecting technological fields, LLM models have already been making their mark. From video production and artistic creation in the humanities to drug development and biotechnology in hard technology, it is bound to bring about earth-shaking changes.

In this era, computing power is regarded as a scarce resource, with large technology giants holding abundant resources, while emerging developers face barriers to entry due to insufficient computing power. In the new AI era, computing power is strength; those who control computing power have the ability to change the world. GPUs, as the cornerstone in the fields of deep learning and scientific computing, play a crucial role.

In the rapidly evolving field of artificial intelligence (AI), we must recognize the dual aspects of development: model training and inference. Inference involves the functionality and output of AI models, while training encompasses the complex processes required to build intelligent models, which include machine learning algorithms, datasets, and computational power.

Taking GPT-4 as an example, if developers want to obtain high-quality reasoning, they need to acquire a comprehensive foundational dataset and significant computing power to train effective AI models. These resources are mainly concentrated in the hands of industry giants, including Nvidia, Google, Microsoft, and certain cloud computing platforms.

High computing costs and entry barriers prevent more developers from entering, making the leading players even stronger. They possess large foundational datasets and significant computing power, with the ability to continuously expand their scale and reduce their costs, which leads to even stronger industry barriers.

But we can't help but wonder if there is a way to reduce computing costs and industry entry barriers by adopting blockchain technology? The answer is yes. Decentralization distributed cloud computing provides us with such a solution in this context.

Despite the current situation where computing power is expensive and scarce, GPUs are not being fully utilized. This is mainly because a ready-made way to integrate these decentralized computing powers and operate them in a commercial manner has not yet emerged. Here are typical GPU utilization figures for different workloads:

Most consumer devices with GPUs belong to the first three categories, namely idle (just started into the Windows operating system):

• GPU utilization is: 0-2%;
• General production tasks (writing, simple browsing): 0-15%;
• Video playback: 15 - 35%.

The above data indicates that the utilization of computing resources is extremely low, and there are no effective measures in the Web2 world to collect and integrate these resources. However, Crypto and the blockchain economy may just be the remedy to this challenge. The crypto economy constructs a highly efficient global market, and due to its unique token economy and characteristics of Decentralization, the pricing, circulation, and matching of market supply and demand of resources are extremely efficient.

Cloud Computing Development Trends

The development of AI impacts the future of humanity, and the progress of computing power determines the development of AI. Since the invention of the first computer in the 1940s, computing models have undergone numerous transformations. From bulky mainframe computers to lightweight laptops, from purchasing centralized servers to renting computing power, the threshold for obtaining computing power has been gradually lowered. Before the advent of cloud computing, companies had to procure their own servers and continuously update them with technological innovations, but the emergence of cloud computing completely changed this model.

The basic concept of cloud computing is that the demand side rents servers and accesses them remotely, paying according to the amount used. Traditional enterprises are now being disrupted by cloud computing. In the field of cloud computing, virtualization technology is at the core. Virtualized servers can divide a powerful server into smaller servers for rental and can dynamically allocate various resources.

This model has completely transformed the business landscape of the computing power industry. In the past, people needed to purchase their own computing power facilities to meet their needs; now, they only need to pay rent on a website to enjoy high-quality computing power services. The future development direction of cloud computing is edge computing. Due to the traditional centralized systems being too far from users, this can lead to a certain degree of latency. Although latency can be optimized, it can never be entirely overcome due to the limitation of the speed of light.

However, emerging industries such as the metaverse, autonomous driving, and telemedicine have extremely low latency requirements, which necessitates moving cloud computing servers closer to users. As a result, more and more small data centers are being deployed around users, which is edge computing.

Compared to centralized cloud computing providers, the advantages of decentralized cloud computing mainly lie in:

• Accessibility and Flexibility: Gaining access to computing power chips on platforms like cloud service providers typically takes several weeks, and high-performance GPU models, such as A100 and H100, are often out of stock. Additionally, to obtain computing power, consumers usually need to sign long-term, inflexible contracts with these large companies, which not only results in time losses but also makes business operations rigid, losing a degree of flexibility. In contrast, decentralized computing power platforms allow for on-demand access to computing power and offer flexible hardware options, providing greater accessibility.

• Lower prices: By utilizing idle chips, along with token subsidies from network protocol parties to chip and computing power suppliers, the distributed computing power network may be able to provide cheaper computing power.

• Anti-censorship: Systems like io.net and Aethir do not position themselves as permissionless systems. Compliance issues such as GDPR and HIPAA are addressed during the GPU launch, data loading, data sharing, and result sharing stages.

With the further development of AI and the ongoing supply-demand imbalance of GPUs, more developers will be driven to turn to decentralized cloud computing platforms. At the same time, during a bull market, due to the rise in cryptocurrency prices, GPU suppliers will earn more profits, which will stimulate more GPU providers to enter this market, creating a positive flywheel effect.

technical challenges

1. Parallelization Problem

Distributed computing platforms typically aggregate a long tail of chip suppliers, which means that individual chip suppliers can hardly complete complex AI model training or inference tasks independently in a short time. If cloud computing platforms want to be competitive, they must break down and allocate tasks through parallelization methods to shorten the total completion time and enhance the platform's computing capacity.

However, the parallelization process faces a series of issues, including how to decompose tasks (especially for complex deep learning tasks), data dependencies, and additional communication costs between devices.

2. New Technology Substitution Risk

With a significant amount of capital invested in ASIC (Application-Specific Integrated Circuit) research and new inventions like Tensor Processing Units (TPU), there may be an impact on the GPU clusters of decentralized computing platforms.

If these ASICs can provide good performance and there is a trade-off in terms of cost, the GPU market currently monopolized by large AI organizations may return to the market. This will lead to an increase in GPU supply, thereby affecting the ecosystem of decentralized cloud computing platforms.

3. Regulatory Risk

Due to the decentralized cloud computing system operating in multiple jurisdictions and potentially being subject to different laws and regulations, there may be unique legal and regulatory challenges. Compliance requirements, such as data protection and privacy laws, can also be complex and challenging.

At this stage, the users of cloud computing platforms are mainly professional developers and institutions, who prefer to use one platform for a long time and will not change it casually. Whether to use a decentralized platform or a centralized platform is just one of the considerations; these users place more importance on the stability of the service. Therefore, if a decentralized platform has strong integration capabilities and stable and sufficient computing power, it is more likely to win the favor of these customers, establishing long-term cooperative relationships and stable cash flow income.

Aethir Platform Introduction

Aethir Cloud is a decentralized real-time rendering platform based on the Arbitrum network, which helps gaming and artificial intelligence companies deliver their products directly to consumers by aggregating and intelligently redistributing new and idle GPUs from enterprises, data centers, cryptocurrency mining operations, and consumers.

One of the key innovations of this project is the resource pool, which brings together decentralized computing power contributors under a unified interface to serve global clients. A significant feature of the resource pool is that GPU providers can freely connect or disconnect from the network, allowing businesses or data centers with idle equipment to participate in the network during downtime, enhancing the flexibility of suppliers and the utilization of equipment.

The operation of the Aethir ecosystem relies on three core infrastructures:

• Container: The key function of a container is to provide real-time remote rendering services, delivering a "zero-latency" experience. The container is the actual location of cloud computing, serving as a virtual endpoint that executes and renders applications. This shifts the workload from local devices to the container.

• Checker: The Checker node ensures the integrity and quality of service of the Aethir network by checking the Container and its service processes. After completing its tasks, the Checker Node uses its own private