Author: Iris Chen, Dr. Nick
1. Demands and challenges coexist
The "2022-2023 Global Computing Power Index Assessment Report" points out that against the backdrop of a global GDP slowdown, the digital economy still maintains a strong growth momentum. The proportion of digital economy in GDP in major countries around the world is increasing year by year. It is expected that the overall proportion of sample countries will increase from 50.2% in 2022 to 54.0% in 2026. Computing power has gradually become the main force driving economic growth. For every 1 point increase in the computing power index, the country's digital economy and GDP will increase by 3.6‰ and 1.7‰ respectively. More importantly, when the computing power index exceeds 40 points, the driving force for GDP growth for each 1 point increase will be 1.3 times higher than that below 40 points, and even 3 times higher when it exceeds 60 points. The advantages of the first-mover regions in computing power will be strengthened as the proportion of computing power investment increases, further widening the gap with the latecomer regions, which shows the importance of developing computing power.
1. The AIGC wave is coming, and the computing power industry has huge demand
With the application and development of key technologies such as artificial intelligence, blockchain, Internet of Things, and AR/VR, the demand for computing power will increase in the future. It is expected that by 2030:
AI: Deep penetration into all industries will require 16,000 EFLOPS of computing power (equivalent to embedding 160 billion Qualcomm Snapdragon 855 NPU2s in smartphones)
Blockchain: Supporting areas such as cryptocurrencies will require 5,500 EFLOPS of computing power (equivalent to 1.3 billion AntMiner V9s)
IoT: Connecting all the devices in factories and homes will require 8,500 EFLOPS of computing power (equivalent to using 7.9 billion chips in high-end IoT edge devices)
Space computing/AR/VR/Metaverse: At full potential, 3,900 EFLOPS of computing power will be required (equivalent to 2.1 billion Sony PS4 consoles)
At the same time, the popularity of ChatGPT in 2022 triggered the AIGC wave, and the demand for computing power increased further. In the GPT series released by Open AI, GPT3 is a language model composed of 175 billion parameters, and the parameters of GPT4 are at the trillion level. With the increasing number of parameters in large models, the computing power required to train an AI model will increase 275 times every two years. This will push the growth rate of the global AI computing market to a new height. IDC predicts that the global AI computing market will reach US$34.66 billion in 2026, of which the generative AI computing market will grow from US$820 million in 2022 to US$10.99 billion in 2026, accounting for 4.2% of the AI computing market share to 31.7%. Under this development trend, the demand for computing power in the future will be huge.
2. Security threat costs are difficult to reduce, and the computing industry faces challenges
(1) Security: Flexible access to computing power network and multiple distributed resource nodes
The computing network consists of five main parts: computing network service layer, computing network scheduling layer, computing center, edge computing/user center, and computing network. However, while this architecture provides efficient and flexible computing services, it also brings a series of security challenges:
The computing power network has the characteristics of ubiquitous computing power and flexible access. Frequent resource links will increase the attack exposure of resources.
A huge amount of confidential and private data is circulated in the computing network. If it is tampered with or leaked during transmission, it will cause serious consequences.
Computing power services are end-to-end services with a large user base, a large number of distributed resource nodes, complex data information management, and difficulty in evidence traceability.
The new computing network architecture adds computing network perception units, computing network control units and other network elements, which increases the complexity of management and control
(2) Cost: GPUs are in short supply and computing power is seriously idle
With the prosperity of AI, the demand for GPUs has surged. Currently, most of the GPU market is occupied by NVIDIA, but the supply of NVIDIA chips is tight and the price has also risen. Taking the A100 GPU as an example, its market unit price has reached 150,000 yuan, an increase of more than 50% in two months. At the same time, the application of large models will further increase the computing power cost. It has been calculated that 10,000 NVIDIA A100 chips are the computing power threshold for making a good AI large model. The single training cost of GPT3 is more than 12 million US dollars.
At the same time, GPUs have the problem of idle computing power. GPT3 needs to store more than 1TB of data in memory to train a model with 175 billion parameters, which exceeds any GPU currently available. Due to memory limitations, more GPUs are required for parallel computing and storage, resulting in low GPU utilization and idle computing power. Also due to memory limitations, the model complexity does not increase linearly with the number of GPUs required, which will aggravate the problem of low GPU utilization. GPT4 was trained on approximately 25,000 A100 GPUs for 90 to 100 days, and its computing power utilization was only 32% to 36%. There is also a lot of computing power in independent data centers, crypto miners, and consumer devices of users such as MacBooks and gaming PCs. These resources are difficult to aggregate and utilize.
With the booming development of computing power, electricity demand will also grow rapidly. It is estimated that the global data center electricity demand will be 430-748 terawatt hours from 2023 to 2027, equivalent to 2-4% of the global electricity demand from 2024 to 2027, which brings challenges to the power infrastructure. Morgan Stanley predicts that under the baseline scenario where GPU utilization increases from 60% to 70%, the total power capacity of global data centers will reach 70-122 gigawatts from 2023 to 2027, with a compound annual growth rate of 20%. Specifically:
Bull market scenario (90% chip utilization): Global data center power demand is expected to be 446-820 TWh in 2023-2027
Bear market scenario (50% chip utilization): Global data center power demand is expected to be 415-677 TWh in 2023-2027
Therefore, companies that can meet the rapidly growing power demands of computing power will benefit from this trend, especially those power solution providers that can reduce power supply delays in data centers.
2. Development Trends and Project Introduction
1. Decentralized computing provides a secure and low-cost computing solution for Web 3
The essence of Web 1 is union, web pages are "read-only", and users can only search and browse information; the essence of Web 2 is interaction, websites are "writable and readable", and users are not only the recipients of content, but can also participate in creating content. Web 3 is the era of the Internet of Everything, websites are "readable, writable and holdable", and the ownership and control of digital content created by users belong to themselves, and they can choose to sign agreements with others to distribute them. As the representative of the next generation of the Internet, Web 3 emphasizes decentralization, openness and user sovereignty. Decentralized computing is different from traditional cloud computing, effectively meets the computing needs driven by modern technology, and becomes the core of Web 3 infrastructure. With the development of new Internet technologies and the further expansion of data volume, the decentralized application market has broad prospects for development. Zhiyan Consulting predicts that the global decentralized application market size is expected to reach US$1,185.54 billion in 2025.
Facing the security, cost and power challenges of the computing power industry, building a decentralized distributed computing power network is an important direction for the development of AI infrastructure. Decentralized computing makes comprehensive use of existing computing resources through the leasing, sharing and scheduling of computing power, providing a secure, low-cost, and zero-downtime computing power solution for various applications in the Web 3 ecosystem. Compared with traditional centralized systems, the specific advantages of decentralized computing are as follows:
"Safety
The participants all have processing capabilities. If one participant is threatened, the others can respond.
Allows for distributed control and decision-making. Helps ensure that no single entity can exercise total control over the internet or its users, users are less likely to be monitored or censored, and online privacy and freedom of expression are greater.
》Low cost: Decentralized computing spreads costs and responsibilities across multiple entities, making it more affordable and sustainable in the long run. Currently, the Web 3 decentralized computing platforms on the market can offer prices that are 80-90% lower than centralized computing platforms.
Cheaper computing power. In traditional data centers, the cost structure is servers (30%), housing (12%), network (15%), AC (21%), power supply (17%), and labor (5%), while decentralized computing relies on users to share resources and contribute computing power in a mutually beneficial way, theoretically saving 70% of the cost.
Cheaper training costs. Decentralized computing allows GNN training to be expanded to a billion edge models with the help of thousands of parallel threads of serverless technology. According to UCLA research, for large models, decentralized computing can provide 2.75 times higher performance per dollar than traditional systems. For sparse large models, decentralized computing is 1.22 times faster and 4.83 times cheaper.
Cheaper deployment costs. Traditional AI solutions require heavy investments in software development, infrastructure, and talent. Decentralized computing allows developers to leverage existing resources and infrastructure, making it easier to build and deploy AI applications. It also democratizes AI development, allowing users to share computing resources and collaborate on developing AI solutions.
Infrastructure better suited to AI. By reducing the cost of training and computing, decentralized computing has the potential to enable more organizations and individuals to use AI and drive growth and innovation across many industries.
》No downtime service: The decentralized network nodes are dispersed, theoretically never downtime, and there is no single point of failure.
Project Introduction
Akash Network: A decentralized cloud computing market that allows users to buy and sell computing resources securely and efficiently. Unlike other decentralized platforms, users can run any cloud-native application on Akash managed containers. There is no need to rewrite the entire Internet in a new proprietary language, and there is no vendor lock-in to prevent switching cloud providers.
io.net: A decentralized computing network that allows machine learning engineers to access distributed cloud clusters at a lower cost than centralized services. It has featured products such as IO Worker, IO Cloud, and IO Browser, and is valued at over $1 billion on Solana.
2. AI drives high-performance computing, and high-performance computing empowers AI
High-performance computing refers to computing systems that use supercomputers and parallel computer clusters to solve advanced computing problems. Such systems are typically more than 1 million times faster than the fastest desktop, laptop or server systems and have a wide range of applications in established and emerging fields such as self-driving cars, the Internet of Things and precision agriculture.
High-performance computing accounts for only about 5% of the total available market in data centers, but with the rapid development of AI and the use of large models, the increase in AI and high-performance data analysis workloads is driving changes in HPC system design. HPC also empowers AI, and the two promote each other's development. Global HPC spending will be approximately US$37 billion in 2022. Hyperion predicts that it will reach US$52 billion in 2026. At the same time, the HPC-enabled AI market will have a compound growth rate of 22.7% from 2020 to 2026.
Project Introduction
Arweave: The newly proposed AO protocol uses a non-Ethereum modular architecture to achieve ultra-high performance computing on the public storage chain and even achieve a quasi-Web2 experience, providing a good new infrastructure for Web3 x AI.
iExec: A decentralized cloud computing platform that provides high-performance computing services, allowing users to rent computing resources to perform computationally intensive tasks such as data analysis, simulation, and rendering.
CETI: Founded by the former CEO of crypto.com, it targets enterprise-level high-performance computing centers.
3. Turning point of human-computer interaction: spatial computing
Spatial computing refers to computers that use AR/VR technology to integrate the user's graphical interface into the real physical world, thereby changing human-computer interaction. The release of Microsoft's MR headset Hololens in 2015 was a milestone in modern spatial computing. Although it was not popular, it proved the potential of spatial computing. This year, Apple's release of Vision Pro brought more accurate spatial perception technology and a deeper user interaction experience, pushing spatial computing to the forefront.
In fact, we are reaching a turning point in human-computer interaction: from the traditional keyboard and mouse configuration to the edge of touch gestures, conversational AI and enhanced visual computing interaction. According to IDC's forecast, global VR device shipments will reach 9.17 million units in 2023, a year-on-year increase of 7%, while AR device shipments will be 440,000 units, a year-on-year increase of 57%. It is expected that in the next four years, the VR market will grow at an annual rate of more than 20%, while the AR market will reach more than 70%. The development of AR/VR technology will greatly increase the importance of spatial computing. Following PCs and smartphones, spatial computing has the potential to drive the next wave of disruptive change - making technology a part of our daily behavior and connecting our physical and digital lives with real-time data and communications.
Project Introduction
Clore.ai: A platform that connects tenants and users who need GPUs, designed to give users access to powerful computing resources at competitive prices and flexible terms. Its powerful GPUs allow users to render movies at a professional level, significantly reducing the time required, and are compatible with a variety of rendering engines, and can also be used for AI training and mining.
Render Network: A decentralized GPU rendering platform designed to advance next-generation rendering and AI technologies, enabling users to scale GPU rendering work to high-performance GPU nodes around the world on demand.
4. Edge computing becomes an important supplement to cloud computing
Edge computing refers to processing data at a location that is physically closer to the end device, where the "edge" is located at a location where the round-trip time to the end user is at most 20 milliseconds. Edge computing deploys computing resources closer to the end device so that data can be processed locally, thereby reducing the delay in transmitting data to the cloud for processing and the pressure on network bandwidth. Therefore, it has more advantages in terms of latency, bandwidth, autonomy and privacy.
Technology giants such as Facebook, Amazon, Microsoft, Google and Apple are investing in edge computing and edge locations (from internal IT and OT to external, remote sites) to get closer to end users and where data is generated. Bank of America predicts that by 2025, 75% of enterprise-generated data will be created and processed at the edge, and by 2028, the market size of edge computing will reach $404 billion, with a compound annual growth rate of 15% from 2022 to 2028.
Project Introduction
Aethir: A cloud computing infrastructure platform, Aethir Edge was launched in April 2024. As the only authorized mining device of Aethir, Aethir Edge is leading the development of decentralized edge computing and leading the future of edge computing towards democratization.
Theta Network: A decentralized video transmission service platform that aims to solve bottleneck problems such as high cost and low efficiency in existing video transmission systems. It is planned to launch Theta EdgeCloud, a hybrid cloud computing platform based on a fully cross-edge architecture, in the second quarter of 2024.
5. AI training is expected to shift to AI reasoning
Under the trend of decentralization, AI training is not the best landing scenario for DePIN at present. The requirements of AI production for computing power mainly revolve around AI reasoning and AI training. AI training refers to training a complex neural network model by feeding a large amount of data, and AI reasoning refers to using the trained model to infer various conclusions using a large amount of data. Therefore, decentralization is combined with computing power, and the difficulty coefficient decreases layer by layer from training to fine-tuning training to reasoning. If a decentralized computing power application is built on Ethereum for GPT to use, a single matrix multiplication operation alone will consume up to $10 billion in gas fees and take 1 month. The training cost of each token (1000 tokens is approximately equal to 750 words) is usually about 6N (N is the number of parameters of the large language model), while the reasoning cost is only about 2N, which means that the reasoning cost is about one-third of the training cost.
At the same time, compared with AI training, AI reasoning is more closely related to the demand of large-scale application terminals such as consumer electronics. Counterpoint Research predicts that the shipment volume of the global PC market will return to the pre-epidemic level in 2024. It is expected that from 2020, AI PC will grow at a compound growth rate of 50% and dominate the PC market after 2026. With the emergence of new AI-integrated consumer electronics products such as AI PCs and AI smartphones in 2024, the trend of large-scale application of end-side AI models and AI software will become increasingly apparent, which also means that the importance of AI reasoning is becoming increasingly prominent and has become the core technology behind the efficient operation of end-side large models and AI software. The development focus of the AI industry is expected to shift from training to reasoning.
Project Introduction
Nosana: A blockchain-based distributed GPU resource sharing platform designed to solve the GPU shortage problem in the market. In 2023, it turned to AI reasoning and pioneered a large-scale GPU computing grid for AI reasoning, which is a move to purposefully integrate blockchain technology into AI, making it an ideal tool for handling AI's demanding computing requirements.
Exabits: A decentralized AI and high-performance computing service platform that aims to build a fair, easy-to-use, and inclusive AI ecosystem, providing affordable accelerated computing for AI model training and reasoning.
