MIIX Capital：io.net 專案研究報告
- WBOY原創
- 2024-06-30 12:00:08794瀏覽
1、專案情況
1.1 商業化
透過組裝來自獨立資料中心、加密貨幣礦工和 Filecoin 或 Render 等專案的 100 多萬個 GPU 來獲取運算能力。
它的目標是將100 萬個GPU 組合到DePIN（去中心化實體基礎設施網絡）中，打造一個企業級、去中心化的分散式運算網絡，透過匯聚全球閒置的網絡運算資源（目前主要是GPU），為人工智慧工程師提供價格更低、更容易取得、更靈活適配的網路運算資源服務。
對於使用者來說，它就相當於一個去中心全球閒散GPU資源的集市，讓人工智慧工程師或團隊可以在這裡按照他們的需求客製化搭配和購買所需的GPU計算服務。
1.2 團隊背景
Ahmad Shadid 是創始人兼首席執行官，此前是 WhalesTrader 量化系統工程師。
Garrison Yang 是首席策略長兼首席行銷官，此前是 Ava Labs 成長與策略副總裁。
Tory Green 是營運官，先前是 Hum Capital 營運長、Fox Mobile Group 企業發展與策略總監。
Angela Yi 是商務拓展副總裁，畢業於美國哈佛大學，負責規劃並執行銷售、夥伴關係和供應商管理等關鍵策略。
2020 年Ahmad Shadid 為機器學習量化交易公司 Dark Tick 建立 GPU 運算網路時，因為交易策略接近高頻交易，所以需要大量的算力，雲端服務廠商高昂的GPU服務費用成為了他們的難題。
對算力的巨大需求以及所面臨的高昂成本促使他們他們決定去做去中心化分散式運算資源這件事，隨後又在 Austin Solana Hacker House 獲得關注度。因此，io.net屬於團隊從自身面臨的痛點出發，提出解決方案並進行業務落地和拓展。
1.3 產品/技術
市場用戶面臨的問題：
可用性有限，使用 AWS、GCP 或 Azure 等雲端服務存取硬體通常需要數週時間，而且市場上流行的 GPU 模型通常不可用。
選擇餘地很少，如在 GPU 硬體、位置、安全等級、延遲等方面用戶幾乎沒有選擇餘地。
成本較高：獲得優質 GPU 非常昂貴，每月很花費數十萬美元用於訓練和推理。
解決方案：
透過聚合未充分利用（例如獨立資料中心、加密礦工以及Filecoin、Render 等加密專案）的GPU ，將這些資源整合到DePIN 中，使工程師能夠在系統中獲得大量計算能力。它允許 ML 團隊跨分散式 GPU 網路建立推理和模型服務工作流程，並利用分散式運算庫，來編排和批次訓練作業，以便可以使用資料和模型並行性在許多分散式裝置上並行化。
此外，io.net 利用具有高級超參數調整的分散式計算庫來檢查最佳結果、最佳化調度並簡單地指定搜尋模式。它還使用開源強化學習庫，該庫支援生產級、高度分散式的 RL （強化學習）工作負載以及簡單的 API。
產品組成：
IO Cloud，目的是部署和管理按需來分配去中心化的GPU 集群，與IO-SDK無縫集成，提供擴展人工智能和Python應用程序的全面解決方案。可提供無限的運算能力，同時簡化了GPU/CPU資源的部署與管理。
IO Worker，為用戶提供一個全面且用戶友好的介面，透過直覺的網路應用程式高效管理他們的GPU節點操作。該產品的範圍包括與用戶帳戶管理、計算活動監控、即時數據顯示、溫度和功耗追蹤、安裝輔助、錢包管理、安全措施和盈利能力計算相關的功能。
IO Explorer，主要為用戶提供全面統計數據和GPU 雲各個方面的可視化圖，讓用戶輕鬆即時監控、分析和了解io.net網絡的複雜細節，提供對網絡活動、重要統計數據、數據點和獎勵交易的全面可見性。
產品特點：
去中心化運算網路：io.net 採用去中心化的運算模式，將運算資源分佈在全球各地，從而提高了運算效率和穩定性。
低成本存取：相較於傳統的集中式服務，io.net Cloud 提供了更低的存取成本，使更多的機器學習工程師和研究人員能夠獲得運算資源。
分散式雲端集群：平台提供了一個分散式的雲端集群，使用者可以根據自己的需求選擇合適的運算資源，並將任務分配到不同的節點上進行處理。
支援機器學習任務：io.net Cloud專注於為機器學習工程師提供運算資源，使他們能夠更輕鬆地進行模型訓練、資料處理等任務。
1.4 發展路線圖
根據io.net白皮書公佈的信息，項目產品的路線圖是：2024年1月-4月，V1. 0全面發布，致力於io.net生態系統的去中心化，使其能夠實現自我託管和自我複製。
1.5 融資資訊
根據公開新聞資訊顯示，2024年3月5日，io.net對外宣布完成3000 萬美元A 輪融資，Hack VC 領投，Multicoin Capital、6th Man Ventures、M13、Delphi Digital、Solana Labs、Aptos Labs 、Foresight Ventures、Longhash、SevenX、ArkStream、Animoca Brands、Continue Capital、MH Ventures、Sandbox Games等參與。 【1】值得注意的是，這輪融資後，io.net整體估值10億美元。
2、市場數據
2.1 官方網站
從2024年1月至2024年3月的官網數據看，總訪問量為5.212M，月均訪問1.737M，跳出率為18.61%（較低），各區域用戶訪問數據較均勻，且直接訪問和搜尋訪問佔比超過80%，可能說明訪問用戶資料中髒數據佔比不高，他們對io.net有基本了解，並且願意進一步了解和在網站進行交互。
2.2 社媒社群
3、競爭分析
3.1 競爭格局
io.net的核心業務是跟去中心AI業務算力有關，它最大的競爭對手就是以AWS、Google Cloud、微軟商業雲（Azure為代表）為代表的傳統雲端服務廠商。根據國際數據公司(IDC)、浪潮資訊與清華大學全球產業研究院共同編製的《2022–2023年全球算力指數評估報告》，全球人工智慧運算市場規模預計將從2022年的195億美元成長到2026年的346.6億美元。 【2】
對比全球主流雲端運算廠商的銷售收入：2023年AWS雲端服務銷售收入90.8億美元，Google Cloud銷售收入33.7億美元，微軟智慧雲端業務銷售收入96.8億美元。 【3】三者市佔率佔到全球66%左右，同時這三家巨無霸公司市值均在兆美元以上。
In sharp contrast with the high income of cloud service vendors, how to improve GPU utilization become a focus issue. According to a survey by AI Infrastructure, most GPU resources are underutilized—about 53% believe that 51~70% of GPU resources are underutilized, 25% believe that the utilization rate reaches 85%, and only 7% Utilization rates are thought to be over 85%. For io.net, the huge demand for cloud computing and the problem of insufficient effective utilization of GPU resources are the market opportunities it faces.
3.2 Advantage Analysis
io.net’s biggest competitive advantage is reflected in the ecological niche advantage or first-mover advantage. According to official data: io.net currently has more than 40K GPU clusters, more than 5600 CPUs, and more than 69K Woker Nodes. The time to deploy 10,000 GPUs is less than 90 seconds. The price is 90% cheaper than competitors, and the valuation is 1 billion. Dollar. io.net not only provides customers with low prices of 1-2% off compared to centralized cloud service providers and instant online services without permission, but also provides computing power providers with additional startup through the upcoming IO token. Incentives to jointly help achieve the goal of connecting 1 million GPUs.
In addition, compared with other DePIN computing projects, io.net focuses on GPU computing capabilities, and the scale of its GPU network is more than 100 times ahead of similar projects. io.net is also the first in the blockchain industry to integrate the most advanced ML technology stack (such as Ray cluster, Kubernetes cluster and giant cluster) into the GPU DePIN project and put it into large-scale practice, which makes it not only the number of GPUs, but also the largest number of GPUs. Leading the way in technology application and model training capabilities.
With the continuous development of io.net, if the GPU capacity can be increased to 500,000 concurrent GPUs across the entire network to compete with centralized cloud service providers, it will be able to provide services similar to Web 2 at a lower cost, and have Opportunity to gradually establish its core position in the field through close cooperative relationships with major DePIN and AI players (including Render Network, Filecoin, Solana, Ritual, etc.) to become the leader and settlement layer of the decentralized GPU network, providing services for the entire Web 3xAI ecosystem brings vitality.
3.3 Risks and Problems
io.net is an emerging computing resource integration and distribution platform that is deeply integrated with Web3, and the business involved is highly overlapping with traditional cloud service vendors, which makes it a leader in technology and Markets are faced with location risks and obstacles.
Technical security risks, As an emerging platform, io.net has not experienced large-scale application testing, nor has it demonstrated the ability to prevent and respond to malicious attacks. Facing the access, distribution and management of huge amounts of computing power resources without corresponding experience or practical verification, problems such as compatibility, robustness and security that are common in technical products are prone to occur. And once a problem occurs, it is likely to be fatal to io.net, because customers care more about their own security and stability and are unwilling to pay for these.
The market is expanding slowly, io.net is highly overlapping with traditional cloud service providers, which means it must directly compete with traditional AWS, Google Cloud, Alicloud, etc., and even directly compete with second- or third-tier service providers. Although io.net .net has a more favorable cost, but its service system and market system for Class B customers have just begun. This is very different from the market operation of the existing Web3 industry. Therefore, at present, it is in terms of market expansion. The progress of the project is not ideal, which is likely to directly affect its project valuation and the market value performance of the token.
Latest security incidents
On April 25, io.net founder and CEO Ahmad Shadid tweeted that the io.net metadata API encountered a security incident, and the attacker exploited the accessible mapping of user ID to device ID, resulting in unauthorized metadata Updated, this vulnerability does not affect GPU access, but does affect the metadata displayed to the user by the front end. io.net does not collect any PII and does not disclose sensitive user or device data.
Shadid said the io.net system design allows for self-healing, constantly updating each device to help restore any erroneously changed metadata. In light of this incident, io.net has accelerated the deployment of OKTA's user-level authentication integration, which will be completed within the next 6 hours. In addition, io.net also launched Auth0 Token for user authentication to prevent unauthorized metadata changes. During database recovery, users will be temporarily unable to log in. All uptime records are unaffected, and this does not impact the vendor's computing rewards.
4. Token Valuation
4.1 Token Model
io.net token economic model will have an initial supply of 500 million IOs at creation, divided into five categories: seed investors (12.5%), Series A investors (10.2%), and core contributors (11.3%), R&D and ecosystem (16%), and community (50%). Will grow to a fixed maximum supply of 800 million over 20 years as IO is issued to incentivize network growth and adoption.
The rewards adopt a deflation model, starting from 8% in the first year and decreasing by 1.02% every month (approximately 12% per year) until reaching the 800 million IO upper limit. The shares of early supporters and core contributors will continue to decrease as rewards are distributed, and the community's share will grow to 50% after all reward distribution is completed. 【4】
The functions of its token include allocating incentives to IO Workers, rewarding AI and ML deployment teams for continued use of the network, balancing partial demand and supply, pricing IO Worker computing units, and community governance.
In order to avoid payment problems caused by the fluctuation of IO currency prices, io.net has specially developed a stable currency IOSD, which is pegged to the US dollar. 1IOSD is always equal to 1 USD. IOSD can only be obtained by destroying IO. Additionally, io.net is considering some mechanisms to improve network functionality. For example, IO Workers might be allowed to increase their probability of being rented by staking native assets. In this case, the more assets they invest, the greater their probability of being selected. Additionally, AI engineers staking native assets can prioritize access to high-demand GPUs.
4.2 Token Mechanism
IO tokens are mainly used by the demand side and the supply side. For the demand side, each computing job is priced in US dollars, and the network will retain the payment until the job is completed. Once a node operator allocates its reward share in USD and tokens, all USD amounts will be allocated directly to the node operator, while the share allocated to tokens will be used to burn IO coins. All IO Coins minted as Compute Rewards during that period are then distributed to users based on the USD value of their coupon tokens (Compute Points).
For the supply side, it includes availability rewards and computing rewards. Among them, the reward is calculated for jobs submitted to the network. Users can select the time preference "duration of cluster deployment in hours" and receive cost estimates from the io.net pricing oracle. In terms of availability rewards, the network will randomly submit small test jobs to evaluate which nodes run regularly and are well able to accept jobs from the demand side.
It is worth mentioning that both the supply side and the demand side have a reputation system that accumulates points based on computing performance and network participation to obtain rewards or discounts.
In addition, io.net also sets up an ecological growth mechanism, including staking, invitation rewards and network fees. IO coin holders can choose to stake their IO tokens to node operators or users. Once staked, stakers will receive 1–3% of all rewards earned by participants. Users can also invite new network participants to join and share part of their future income. Network fees are set at 5%.
4.3 Valuation Analysis
We are currently unable to obtain accurate revenue data for projects in the track, so we cannot accurately conduct valuations. We mainly conduct this through Render, an AI+DePIN project that is also an AI+DePIN project with io.net. Comparison for your reference.
As shown in the picture, Render Network is currently in the AI+Web3 track The leading project of decentralized GPU rendering solutions has a total GPU resource of 11,946 and a current market value of US$3 billion (FDV US$5 billion); while io.net has a total GPU resource of 461,772, which is 38 times that of Render, and is currently valued at 1 billion. For the io.net and Render projects, the core key capabilities of both are decentralized GPU computing power. Therefore, from the perspective of GPU supply as the core comparison dimension, the market value of io.net will most likely exceed that of render, at least On par.
Render network’s Frames Rendered in 2022 is 9,420,335, and GMV is 2,457,134 US dollars. Currently, Render Network’s Frames Rendered is 31,643,819, which estimates that the entire GMV is approximately 8,253,75 1 Dollar.
Compared with io.net, the 4-month GMV is 400,000. Assuming that io.net grows at an average rate of 4-month GMV of 400,000, the 12-month GMV is 1,200,000. If io.net wants to reach the current GMV of Render Network, There is still room for growth of 6.8 times. Now io.net is valued at US$1 billion. Based on the above analysis, io.net's market value is expected to reach more than US$5 billion during the bull market cycle.
5. Summary
The emergence of io.net has filled the gap in the field of decentralized computing and provided users with a novel and potential computing method. With the continued development of fields such as artificial intelligence and machine learning, the demand for computing resources is also increasing, so io.net has high market potential and value.
On the other hand, although the market has given io.net a high valuation of US$1 billion, its products have not been tested by the market, there are uncertain risks in technology, and whether it can effectively match its supply and demand relationship It is also a key variable that determines whether its subsequent market value can hit a new high. Judging from the current situation, the io.net platform has initially achieved results on the supply side, but it has not fully exerted its efforts on the demand side. As a result, the overall GPU resources of the current platform are not fully utilized. How to mobilize GPU more effectively? The need for resources is a challenge that the team has to face.
If io.net can complete the rapid access to market demand and does not encounter or encounter major risks and technical problems during the operation process, its overall business will start a growth flywheel with its AI+DePIN entity business attributes. , becoming the most eye-catching project product in the Web3 field, which also means that io.net will be a high-quality investment target for the branch. Let us continue to follow up, observe and verify carefully.
Reference resources
【1】https://www.coincarp.com/fundraising/ionet-series-a/
【2】https://medium.com/ybbcapital/promising-sector-preview-the- decentralized-computing-power-market-part-i-368c0621021a
【3】https://www.crn.com/news/cloud/2024/aws-vs-microsoft-vs-google-cloud-earnings-q4- 2023-face-off?page=2
【4】https://www.chaincatcher.com/article/2120813
以上是MIIX Capital：io.net 專案研究報告的詳細內容。更多資訊請關注PHP中文網其他相關文章！