How was ionet able to establish a decentralized computing power platform

Source: Trustless Labs
Background
With the launch of GPT-4 LLM by OpenAI, the potential of various AI text-to-image models has been witnessed. The increasing demand for GPU and other computing resources is due to the growing applications based on mature AI models.
According to an article by GPU Utils in 2023 discussing the supply and demand situation of Nvidia H100 GPU, large enterprises entering the AI business have a strong demand for GPU. Tech giants such as Meta, Tesla, and Google have purchased large quantities of Nvidia GPUs to build AI-focused data centers. Meta has about 21,000 A100 GPUs, Tesla has about 7,000 A100 GPUs, and Google’s data centers also have a significant investment in GPUs, although specific numbers were not provided. The demand for GPUs, especially H100, continues to grow driven by the demand for trained large language models (LLM) and other AI applications.
At the same time, according to Statista, the AI market size has grown from $134.8 billion in 2022 to $241.8 billion in 2023 and is expected to reach $738.7 billion in 2030. The market value of cloud services has also increased by about 14% from $633 billion, with a significant portion of this growth attributed to the rapid increase in demand for GPU computing power in the AI market.
For a rapidly growing and highly potential AI market, how can we deconstruct and explore investment entry points related to it? According to a report by IBM, we have summarized the infrastructure needed to create and deploy AI applications and solutions. It can be said that AI infrastructure mainly exists to handle and optimize the large datasets and computing resources relied upon by training models, solving data processing efficiency, model reliability, and application scalability issues from both hardware and software perspectives.
AI training models and applications require a large amount of computing resources, preferring low-latency cloud environments and GPU computing power. In terms of software stack, it also includes distributed computing platforms such as Apache Spark/Hadoop. Spark distributes the workflow that needs to be processed to various large computing clusters and has built-in parallel mechanisms and fault-tolerant design. The naturally decentralized design philosophy of blockchain makes distributed nodes the norm, and the POW consensus mechanism established by BTC establishes that miners need to compete for block results through computing power (workload), which has a similar workflow to AI, which also requires computing power to generate models/inferencing problems. Traditional cloud server vendors are beginning to expand new business models, such as renting out graphics cards like renting out servers and selling computing power. Mimicking the approach of blockchain, AI computing power adopts a distributed system design and can utilize idle GPU resources to reduce the computing power costs for startups.
Io.net Project Overview
Io.net is a distributed computing power provider that integrates the Solana blockchain, aiming to utilize distributed computing resources (GPU & CPU) to address the computing needs in the AI and machine learning fields. Io integrates idle graphics cards from independent data centers and cryptocurrency miners, and collaborates with crypto projects such as Filecoin/Render to aggregate the resources of over 1 million GPUs to address the shortage of AI computing resources.
On a technical level, io.net is built on the ray.io distributed computing framework to provide distributed computing resources for AI applications, including reinforcement learning, deep learning, model optimization, and model execution that require computing power. Anyone can join the io network as a worker or developer without additional permission. The network will adjust the computing power price based on the complexity of the computing task, urgency, and the supply of computing resources according to market dynamics. Based on the distributed nature of computing power, io’s backend also pairs GPU providers with developers based on the type of GPU demand, current availability, the location of the requester, and reputation.
$IO is the native token of the io.net system, which serves as a medium of exchange between computing power providers and purchasers. Using $IO instead of $USDC can reduce the order processing fee by 2%. $IO also plays an important incentive role in ensuring the normal operation of the network: $IO token holders can stake a certain amount of $IO to nodes, and node operation requires a corresponding income during the idle period with $IO token staking.
The current market value of the $IO token is approximately $360 million, with a fully diluted valuation of $3 billion.
$IO Token Economics
The maximum total supply of $IO is 800 million, with 500 million tokens allocated to various parties at the token generation event, and the remaining 300 million tokens will be gradually released over a period of 20 years (decreasing by 1.02% per month, approximately 12% per year). The current circulating supply of IO is 95 million, consisting of 75 million unlocked for ecological research and community development at the time of TGE, and 20 million for the Binance Launchpool mining reward.
During the IO testnet period, the rewards for computing power providers are as follows:
Season 1 (until April 25) – 17,500,000 IO
Season 2 (May 1 to May 31) – 7,500,000 IO
Season 3 (June 1 to June 30) – 5,000,000 IO
In addition to the testnet computing power rewards, IO also allocated a portion of the rewards to creators who participated in building the community. The first round of community/content creators/Galxe/Discord received 7,500,000 IO, while participants from Discord and Galxe in Season 3 (June 1 to June 30) received 2,500,000 IO. The rewards for the first season of the testnet and the first round of community creation/Galxe have already been airdropped at TGE.
According to official documents, the overall allocation of $IO is as follows:
$IO Token Burn Mechanism
Io.net executes a fixed preset program to repurchase and burn $IO tokens, and the specific repurchase and burn amount depends on the price of $IO at the time of execution. The funds used to repurchase $IO come from the operating income of IOG (The Internet of GPUs – GPU Internet), which charges a 0.25% order booking fee from both computing power buyers and providers, as well as a 2% fee for computing power purchases using $USDC.
Competitive Analysis
Projects similar to io.net include Akash, Nosana, OctaSpace, Clore.AI, and other decentralized computing power marketplaces focused on addressing the computing needs of AI models.
Akash Network utilizes a decentralized marketplace model to gather and lease surplus computing power by using idle distributed computing resources. It implements efficient, trustless resource allocation based on smart contracts, providing secure, cost-effective, and decentralized cloud computing services. It allows Ethereum miners and other users with underutilized GPU resources to lease these resources, creating a cloud service market. Pricing in this market is done through a reverse auction mechanism, enabling buyers to bid on these resources and drive down prices through competitive bidding.
Nosana is a decentralized computing power marketplace project within the Solana ecosystem, aiming to form a GPU grid using idle computing resources to meet the computational needs of AI inference. The project defines its computing power market operation on Solana and ensures that GPU nodes participating in the network reasonably complete tasks. Currently, it provides computing power services for LLama 2 and Stable Diffusion model inference processes during the second stage of its testnet operation.
OctaSpace is an open-source, scalable distributed computing cloud node infrastructure that allows access to distributed computing, data storage, services, VPNs, and more. It includes CPU and GPU computing power, disk space for ML tasks, AI tools, image processing, and rendering scenes using Blender. It was launched in 2022 and runs on its own Layer 1 EVM-compatible blockchain. This blockchain uses a dual-chain system, combining proof of work (PoW) and proof of authority (PoA) consensus mechanisms.
Clore.AI is a distributed GPU supercomputing platform that allows users to obtain high-end GPU computing resources from nodes providing computing power globally. It supports various use cases such as AI training, cryptocurrency mining, and movie rendering. The platform provides low-cost, high-performance GPU services, and users can earn Clore tokens by leasing GPUs. Clore.ai focuses on security, compliance with European laws, and provides powerful APIs for seamless integration. In terms of project quality, Clore.AI’s website is relatively rough and lacks detailed technical documentation to verify the authenticity and data integrity of the project’s self-introduction. Therefore, we maintain skepticism about the project’s graphics resources and real participation level.
Compared to other products in the decentralized computing power market, io.net is currently the only project that allows anyone to join and contribute computing power resources without requiring permission. Users can participate in the network’s computing power contribution with a minimum of 30 series consumer-grade GPUs, as well as resources such as Apple chips like Macbook M2 and Mac Mini. The more abundant GPU and CPU resources and rich API construction enable IO to support various AI computing needs, such as batch inference, parallel training, hyperparameter tuning, and reinforcement learning. Its backend infrastructure is composed of a series of modular layers, allowing for efficient resource management and automated pricing. Most other distributed computing power market projects cooperate with enterprise-level graphics card resources, and user participation has certain barriers to entry. Therefore, IO may have the ability to leverage the flywheel effect of token economics to access more graphics card resources.
The following is a comparison of the current market value/FDV of io.net and its competitors:
Review and Conclusion
The launch of $IO on Binance can be described as a well-anticipated start, with the testnet being popular across the network and gradually being attacked by the masses and questioned for the opacity of the scoring rules, marking a deserved beginning to a high-profile project. The token was listed during a market correction, starting low and ending up at a relatively rational valuation range. However, for the testers who came for io.net’s strong investment lineup, a mixed bag of joys and sorrows, with most users who leased GPUs but did not persist in participating in each season of the testnet not achieving the desired excess returns, instead facing the reality of “anti-lu”. During the testnet period, io.net divided the prize pool into two pools for GPU and high-performance CPU for each season. Season 1 had a delayed announcement of the point exchange ratio due to a hacking incident, but the final exchange ratio for the GPU pool was determined to be close to 90:1. The cost for users who participated by renting GPUs from major cloud platforms far exceeded the airdrop benefits. During Season 2, the official implementation of the PoW verification mechanism was completed, with nearly 30,000 GPU devices successfully participating and passing PoW verification, resulting in a point exchange ratio of 100:1.
After the highly anticipated start, whether io.net can achieve its goal of providing computing needs at various stages for AI applications and how much real demand remains after the testnet, perhaps only time can provide the best proof.