Perhaps it can be traced back to the 1940s. With the invention of computers, we began to enter an era driven by data and computing. Data and computing power have become new factors of production, and their value is - undoubtedly - being more widely and firmly recognized. At present, as the trend of artificial intelligence (AI) has swept everything, the value of data and computing power has been subverted again by new technologies, so that we have to re-evaluate them - in a way that will inevitably lead to value-added. We can get a glimpse of the prosperity of the computing power market through NVIDIA's market performance. As the global leader in AI computing power, NVIDIA achieved unexpected growth in the latest fiscal quarter, and its total market value is approaching the trillion US dollar mark. On May 25, the company released its first quarter financial report for fiscal year 2024, showing that the company achieved quarterly revenue of US$7.192 billion, a month-on-month increase of 19%, GAAP net profit of US$2.043 billion, a month-on-month increase of 44%, and gross profit margin of 64.6%. Among them, the quarterly revenue of data centers reached a record high of US$4.28 billion, and the development of generative AI accelerated, driving the exponential growth of global computing demand; a16z said that the recent progress of artificial intelligence is incredible and has the power to save the world (see Golden Finance's previous report "a16z founder's 10,000-word long article: Why AI will save the world"). However, building AI systems requires the deployment of greater computing power to train and reason about today's largest and most powerful models. This means that large technology companies have an advantage over startups in the competition to extract value from artificial intelligence, thanks to privileged access to computing power and economies of scale of large data centers. In order to compete on a fair playing field, startups also need to be able to affordably use their own large-scale computing power. We are trying to find a new technology to achieve - in short - provide high-quality, secure computing services at a relatively lower price. Whether it is a centralized large-scale public cloud platform, such as AWS, Google Cloud, Microsoft Cloud, and Alibaba Cloud, or a decentralized cloud computing service provider, such as Akash, Flux, Gensyn, Theta, etc., as a computing power provider, they have been given higher value with the popularity of AIGC.According to Google Cloud's report, 85% of the surveyed companies choose to integrate their business with artificial intelligence (AI), machine learning (ML) or natural language processing (NLP), and expect to achieve a 25% increase in productivity by 2026. As early as October 2021, Roblox, the first stock in the Metaverse, suffered a large-scale shutdown of its internal system due to a failure in its cloud infrastructure. It took three days to resume operations, resulting in a loss of $15 million in revenue and a market value of $1.5 billion. Whether it is a public cloud computing service provider based on AWS or a private cloud infrastructure like Roblox, there are centralization defects. Therefore, creating a decentralized cloud computing service that guarantees reliability with highly distributed computing nodes and a strong total computing power has huge market potential in today's growing computing power value and in the foreseeable future.
As the Gensyn team (2023) notes, current solutions for providing computing supply are either oligopolistic and expensive, or inadequate given the computational complexity required for large-scale AI. Meeting today’s surging computing needs requires a system that can cost-effectively exploit all available computing resources (as opposed to today’s global processor utilization of about 40%). Computing supply is inherently constrained by the asymptotic performance of microprocessors, a problem exacerbated by supply chain and geopolitical chip shortages.
In 2021, there are already 14.9 billion mobile devices in the world. In 2022, China alone will have nearly 1 billion mobile phones (not including other mobile devices such as tablets), followed by 650 million in India and 270 million in the United States. The idle computing power we need exists in these large numbers of devices, or more precisely, in the chips of these devices. We assume that these devices are equipped with the A14 Bionic chip of the iPhone 12, and based on the 10% computing power of each device - to maximize the user experience of the mobile device - we estimate that if we have about 100 million such mobile devices as nodes at the same runtime, we will be able to train our own GPT-3 language model through the computing power generated by them.
Back to the above question: how to provide high-quality, secure computing power services at a relatively lower price. Blockchain technology is the answer to this. As a new type of computer, blockchain has its own unique features in computing power solutions. First, developers can write code and make firm commitments to how the code will behave in the future. And ensure that the commitment - or in more blockchain-like terms - consensus cannot be maliciously tampered with; at the same time, this permissionless component of blockchain can create a fair and open market for buyers and sellers of computing power - or any other type of digital resources, such as data or algorithms - that is available worldwide without the need for transactions through a real middleman; in addition, the decentralized nature of blockchain can ensure that users of the above digital resources cannot have their data and algorithms involved in the process of use stolen or leaked due to the intervention of middlemen.
In a computing power ecosystem, clients (B-end users) who want to perform their computing work must have a way to find workers (C-end users). This is usually done through a centralized server that maintains a record of all workers and their identities. Obviously, such an approach cannot be used in a decentralized network. Therefore, there must be a way to manage this process in a decentralized manner. In addition, there must be a way to assign computing tasks to C-end users. Generally, this allocation method should strive to evenly distribute all computing tasks throughout the network to maximize cost savings, and the needs of both parties must be considered at the same time. Whenever C-end users receive computing tasks, they must be successfully executed at some point in the future, so there also needs to be a way to send and receive information about the work and its status to clients and workers.
In addition, an effective mechanism for giving back to C-end users is necessary. In a blockchain network, payment can be naturally completed by payment of cryptocurrency, and the trust of such transactions has a consensus basis. Workers in the network should be paid only after completing the work and the validity of the work is confirmed by the network consensus. The payment procedure should be carried out in a way that makes it costly for workers to cheat, so that honest participation is more attractive than cheating. Therefore, it may be necessary to combine the payment logic with deeper validity checks. If honest participation is more profitable than cheating, this will ensure that the network is not vulnerable to internal attacks. On the contrary, if the above profitability cannot be guaranteed, the protocol may not be usable. When performing work on behalf of customers, payments must be kept in some kind of escrow mechanism. This is to ensure that customers cannot cheat by defaulting on payments after the worker completes the work, and to ensure that workers cannot receive payments before the work is completed. #Depin赛道 #算力 #手机挖矿 #dmcn #AI
