Original author: David
原文来源：Foresight Research
With the long-term rapid development of the world's science and technology, the market value of giant companies such as OpenAI and Nvidia has increased several times in the past two years. Crypto x AI has become the central narrative of this cycle, with high market sentiment and a steady stream of money pouring out to prove that a strong consensus has been formed. In the context of AI as the goal, decentralization as a powerful tool for AI development is indeed extremely attractive and imaginative. Although there is still a huge gap between the actual business implementation and the centralized model, it has become a common goal of web3 participants to expand the four core aspects of AI with the advantages of web3 and exert greater potential through continuous optimization.
- data
- model
- training
- inference
Currently, decentralization can be supported by technology in the four areas mentioned above. First of all, data must be the core, model, training and inference are all ways to process data, so it can be said that data is the raw material of AI technology, while everything else is processing. Whether it's data annotation or data storage, decentralization has a great role and value here.
If data is the raw material, then computing power is the tool used to process the raw material to maximize the efficiency of the output. Next, going straight to the topic of our article, this article will focus on "computing power" to analyze the ecological framework of Crypto x AI x DePIN and the economic model in it.
In this article, I will mainly explain the ecological framework and market situation of "Crypto x AI x DePIN" to help readers understand the value and potential ⬇️ of decentralized computing power
1. DePIN > Decentralized Computing Ecosystem Framework
Pain points: High-quality computing power, as a necessary product for AI research and development, has been monopolized by traditional giants, making it difficult for startups and individual users to buy cost-effective computing power, which is difficult for most buyers to accept.
Decentralized solutions: At present, most projects in the DePIN track adopt the P2P economic model to provide high-quality resources for the resource demanders, allowing each user to be a physical facility resource provider and obtain token rewards at the same time.
With the surge in demand for decentralized AI computing power, in order to better meet customer needs, the development of decentralized AI computing power supply ecology has formed a balanced and comprehensive framework. Among the top projects, Io.net, Exabit, and PingPong play different important roles in the ecosystem, and the technical barriers of these three projects and the pattern of the future development of decentralized computing power are quite shocking.
The decentralized AI computing ecosystem is mainly composed of three parts, which act as resource agents, resource providers, and channel providers in the ecosystem.
Resource Agents - Io.net
Io.net is a decentralized computing network that acts as a computing power agent to provide high-quality AI computing power to customers at a cheap price. On the supply side, they have globally distributed GPUs, and the client is currently in the seed round to the B round, focusing on startups focused on AI inference.
近期这个基于Solana链的DePIN项目完成3000万美元的A轮融资，由Hack VC领投，Multicoin Capital，Foresight Ventures，Solana Labs等参投。
As the top AI computing resource agent, Io.net is aggregating 1,000,000 GPUs to form a huge DePIN computing network, with the aim of providing customers with computing power at a lower price. Users can manually contribute their idle GPU &CPU computing power to io.net's platform to get $IO token incentives. The core goal is to help AI startups reduce costs by providing high-quality AI computing power while controlling prices through decentralization.
IO Cloud, a computing service provided by Io.net. IO Cloud takes the building blocks of a cluster to keep all GPUs connected to each other, which allows GPUs to coordinate their work at scale during training and inference. When GPUs work in harmony, computing power can be pooled to access larger databases and compute more complex models, and AI startups can deploy computing hardware at one-tenth of the centralized price by using io.net products while getting what they need. Even more strikingly, io.net focuses on the computing power of aggregate machine learning. Io.net can help DePIN giants such as Render Network and FileCoin format GPUs to supply machine learning, and achieve the most fundamental and direct resource support for the underlying technology.
Currently, the number of GPU clusters in the io.net collection is currently the largest in the industry. There are more than 200,000 GPUs available online io.net, with the GeForce RTX 4090 having nearly 50,000 available, followed by the GeForce RTX 3090 Ti with over 30,000.
Resource Provider - Exabit
As the most potential AI computing power provider, Exabits, as an AI computing power service node, can provide sufficient chips for deep machine learning. The Exabits team can also be called a unique one in terms of traditional AI computing resources. The team used to be a first-tier agent of the AI giant NVIDIA, relying on such technical resource barriers, Exabit has direct access to hundreds of computer rooms on the resource supply side, and has access to A/H100, RTX4090 and A6000 machines.
Exabits provides large-scale machine learning computing power to web3 computing giants on the client side. Compared to Nebula Block, customers spend more than $140,000 per month to acquire cloud services, and after migrating to Exabits, customers will be charged around $40,000 per month for cloud services, reducing overhead by more than 70% and improving efficiency by 30%.
The main purpose of Exabits is to provide customers with the fastest, highest quality and most reliable computing power through a unique computing power supply channel. High-quality computing power can save user costs while providing customers with a full range of service options.
The quality of AI computing power provided by Exabits has been recognized by many AI computing power agencies, and now it has reached cooperation with computing giants such as Renders Network and Io.net, so as to contribute to machine learning through decentralization.
资源渠道商(Uber)- PingPong
PingPong, as a DePIN resource channel, provides services through matching on request. PingPong adopts a platform-based open protocol to provide services after providing underlying aggregate resources. PingPong's goal is to become a service aggregator for DePIN, which can be understood as the 1inch of DePIN, or the aggregator of Uber.
How to provide services: PingPong provides SDKs through the control layer to obtain various networks and policies, resource status, performance, stability, etc., and then provides the SDK to users through routing algorithms.
Pain points: The resources and services in each DePIN network are limited, and the quality of services is not good enough due to the excessive concentration of regions to find resource allocation globally.
Solution: Routing Algorithm - Obtain data, basic network information and machine information, etc., aggregate to generate policies, and provide services through customer requirement matching. The purpose is to improve the quality and service of the application layer of DePIN, and to find the best price computing network when the resources are insufficient.
2. Analyze the decentralized computing power ecosystem
Io.net and Exabits have entered into a strategic partnership, which is committed to improving the speed and stability of the io.net network as a supplier with an extensive GPU machine library. Io.net reseller allows customers to purchase and lease the highest quality computing power provided by Exabits directly on the io.net network. Io.net and Exabits agree that the success of the decentralized computing industry and the combination of web3 and AI can only be achieved through close collaboration with early industry leaders. With the growing demand for computing power, some of the problems faced by traditional cloud computing are as follows:
- Limited availability: Using cloud services such as AWS, GCP, and Azure often takes weeks to gain access to the hardware, and the most commonly used GPU models are often unavailable.
- Choice limitations: Users are limited in their choice of GPU hardware, location, security level, latency, etc.
- High cost: The right GPU is expensive, and the cost of training and inference can easily reach hundreds of thousands of dollars per month.
The vision of decentralized computing is to provide an open, accessible, and affordable alternative that solves the core problems of centralized cloud service providers, including limited availability, limited hardware options, and high training and inference expenses. Judging from the current situation, challenging the position of the major giants in cloud computing still requires innovators to work together to create and support each other to take a revolutionary step.
Asset model
- Asset-heavy model
Exabits, as the supply side, has an absolute barrier backed by NVIDIA. The only machines that are valuable for machine learning computing power are the A100, RTX4090 and H100, and the price of these three machines is about $300,000 each. At the same time, these machines have become highly scarce resources and have been monopolized by traditional AI giants for a long time. In this case, the resources that Exabits can connect with on the supply side are extremely valuable. Since the quality of retail investors sharing their own GPU idle computing power is not enough to support the computation and processing of large-scale AI models, the role of Exabits in the decentralized computing ecosystem is crucial and cannot be easily replaced.
Exabits' asset-heavy model requires a large amount of fixed asset investment, which makes it difficult for startups to copy and imitate this amount of capital and technology investment. Therefore, if Exabits can cooperate with more decentralized computing power agents to provide sufficient supply of computing resources needed by the industry in the case of continuous expansion of the supply side, it will be easy to achieve industry monopoly and scale effect in the B2B decentralized computing power field.
However, the biggest risk is that after investing a lot of capital, it is impossible to continuously provide resources for computing power agents, so whether the supply side can make large-scale profits depends on whether the computing power agents can have continuous customers. No matter who the hashrate agent is, as long as there are customers and demand, the value of Exabits as a supply side will grow with the growth of demand.
- Asset-light model
Io.net, as the most outstanding computing power agent at present, relies on GPUs distributed around the world on the supply side to form a huge decentralized computing network. From a business point of view, io.net adopts an asset-light operation model, and builds a strong brand here in AI computing power agency through community operation and the establishment of a high degree of consensus.
Io.net's core business:
- Aggregate retail GPU computing power and reward tokens
- Obtain high-quality computing power from the supply side and sell it to AI startups
From the perspective of the enterprise:
- Buy low and sell high-quality computing power from the supply side to C-end customers
- Help users earn tokens by sharing idle GPU computing power
- To provide customers with a computing power mining and staking platform, but you need to invest about $4,000 in the early stage to have a better income. Based on this, Exabits also offers H100 machines that can be fragmented for leasing, so as to improve liquidity.
From the customer's perspective:
- Io.net The price of network computing power is about 80% cheaper than other centralized cloud computing services.
- Stake to earn & Share to earn。
- After the customer invests a certain amount of capital, it can roll over the profit.
As a typical asset-light company, the biggest advantage is that the risk is relatively low, and the team does not need to invest a lot of machine costs before starting like the supply side. Due to the less capital investment, it is easier for the company and investors to obtain higher profit margins. At the same time, because the threshold for entering the industry is low, the commercial model is easy to be plagiarized and copied, which is a point that needs to be carefully considered by long-term value investors.
Three, from 10 to 100?
If the cooperation between Exabit and Io.net can help the decentralized computing ecosystem go from 1 to 10, then bringing PingPong with you may have a chance to reach 100.
PingPong aims to become the largest DePIN service aggregator directly against web2 uber. As a channel provider, by aggregating the real-time situation of various resources, customers are connected to the best resources in terms of price and quality. PingPong adopts a B2B2C asset-light business model, the first B-end is the supply side, the second B-end is the resource agent, and the C-end is to provide customers with the best resource selection through information.
As a platform, if the channel provider can develop into a platform that can issue assets as much as possible, the product will be more valuable. PingPong can use the SDK provided by the routing algorithm to create its own AI agent with computing resources, convert new financial assets, and dynamically help customers who use applications to carry out dynamic mining through the SDK, focusing on mining computing power useful for computing resources. This model, understood as Assets on assets, can greatly enhance the liquidity of resources and funds.
For PingPong, they hope to see more suppliers and agents enter the decentralized computing ecosystem, so that they can better highlight their advantages, expand their business lines and have more customers. To understand very simply, the reason why Baidu and Dianping can dominate the information field is because more merchants and information are uploaded to the Internet, so that customers have a high demand for channel providers.
Fourth, the future can be expected
Decentralized cloud computing is still developing step by step, although the ecological framework and model of decentralized cloud computing have become very clear, and the leaders of various roles are also fulfilling their accusations in the ecosystem, but it is still very early to shake the status of traditional cloud computing giants. When compared with traditional centralized cloud computing, decentralization can indeed solve many problems of customers conceptually, but the overall resources and volume of this market are still very small in comparison. In the case that the computing resources to support the promotion of AI are far from enough, the market needs another clear stream, or a model, to solve the dilemma. The decentralized cloud computing that we can see now can indeed meet some of the needs of start-up AI companies, and where to go from here, let's all be witnesses to this road of disruption, and the participants will follow the evolution of the revolution together!