Recently, while chatting with a friend in Silicon Valley, one point that impressed me was when he told me in an almost incredulous tone that AI can now do things beyond your imagination. After hearing this, I casually asked my question: If that's the case, why does it seem like life hasn’t changed much with the advent of AI?
A friend explained that AI is still in the stage of rolling out infrastructure. On one hand, the valuations are higher, and on the other hand, sufficient mature infrastructure is needed for the changes brought by AI to ultimately manifest at the application layer.
Turning to the Crypto field, Helium, which once made the market question its usefulness, has proven at the end of last year and the beginning of this year that Crypto is not just about tokens; the products themselves can also become part of many people's daily lives.
Similar to the long-tail market in communication, Crypto is also attempting to address some small-scale training needs and AI training markets involving certain on-chain data through decentralized computing, storage, and token incentives.
Focusing on Telegram, utilizing decentralized computing power networks and cloud storage networks to provide infrastructure for AI, targeting the continuously hot Meme in this market cycle, and using AI to assist in creation, PinGo has carefully selected an easily understandable yet not readily considered niche track.
Building decentralized AI infrastructure
Currently, the publicly available information about PinGo is relatively limited, but through the project's documentation, a rough outline of PinGo's vision for a decentralized AI infrastructure can be sketched.
Aggregating idle CPU computing power
In terms of computing power, PinGo has established a network that can activate idle CPUs. Although this initially left the author a bit puzzled, upon reflection, it makes sense.
Currently, there should be no idle GPUs in the market; they are either rented out for AI training or need to be used for professional work such as rendering images, making 'idleness' unlikely. Only CPUs, being multi-threaded processors, have some computing power that can be considered 'idle'.
Although CPUs are less efficient in computation compared to chips specifically designed for AI training, PinGo itself is not aimed at overly complex models. As a supplement to centralized GPU computing power leasing, as long as enough idle CPUs can be mobilized, it can meet the vast majority of demands.
Fog storage and fog computing based on CDN
In the document, PinGo mentioned plans to quickly concentrate CPU resources using CDN. CDN stands for Content Delivery Network, which is a network set up by traditional cloud services to store popular data in various locations to improve data transmission efficiency, allowing users not to retrieve data from the main data center every time, thus alleviating pressure on the data center and greatly enhancing user experience.
Although PinGo did not explain in detail how its so-called CDN network operates, from the overall concept and architecture, this CDN network should be a peer-to-peer network designed for quicker responses to training tasks.
In simple terms, PinGo does not have a 'data center' like cloud services; it must have a large number of 'data relay stations' to handle the global storage and computing resources.
These CDN nodes scattered around will utilize algorithms to aggregate the most recent idle computing resources to process data upon receiving training tasks, thus achieving training objectives. After completing the training, these CDNs will promptly feedback the results to the corresponding users.
Essentially, PinGo utilizes a distributed network architecture to meet the demands for data storage and computation during the training process, which is also the 'Decloud' concept it mentions. I prefer to use the 'fog' concept, which is more decentralized than cloud, to elaborate on the specific implementation process.
In terms of security, PinGo employs passive measures including NGFW firewalls and role-based access control, as well as active means such as penetration testing and database audits to ensure data security within the CDN network.
Low cost and customization
By utilizing idle computing power to build a decentralized data network to meet the demands of the AI long-tail market, PinGo hopes to achieve lower computing costs and more personalized customization needs with this solution.
Perhaps many people will wonder whether the demand for personalized customization can really be met in a situation where several major cloud services dominate the market. Is there really market space?
This has actually been a question I have pondered for a long time, but the fact is that the market space is still larger than expected. A report by Canalys shows that as of the fourth quarter of 2023, excluding Alibaba Cloud, Huawei Cloud, Tencent Cloud, Baidu Cloud, AWS, and Tianyi Cloud, there is still a 4.4% market share, and based on nearly 10 billion USD in domestic cloud spending in 2023, this share amounts to several billion RMB.
In this regard, I also consulted professionals in the industry, who stated that in fact, the customization level of large cloud service providers is not high, perhaps to a certain extent due to 'big companies bullying customers', where users can only choose services based on set tiers. Additionally, the prices of large cloud services are also relatively high, of course, because they can largely guarantee stability, but many demands do not require such high stability. These ordinary demands have given rise to the existence of small cloud service providers.
For PinGo, even if its services only occupy a certain market share in the niche segments of Web3 and AI (the international market size far exceeds 10 billion USD), it is sufficient to generate considerable cash flow, thereby increasing on-chain activity. Additionally, its strategy of aggregating idle computing power not only reduces prices but also, due to the fragmentation of computing power, can meet finer-grained demands, thus attracting user engagement.
Just like Filecoin is already widely used in data and information storage, ordinary users do not have enough awareness.
Starting with Memes, aiming at various AI tools
After the rough era of Web3 entrepreneurship, starting from a specific usage scenario has become a question that needs to be considered from the outset. PinGo seized on two absolute hot spots in the current market: the TON ecosystem and Memes as entry points, initially focusing on providing AI-generated works for Telegram users.
Telegram bots and AI creative factories
The early accumulation of seed users requires a certain degree of incentives. Fortunately, TON-based projects have the enormous user pool of Telegram, and project teams only need to think about how to find their target users among them. To this end, PinGo launched the Telegram bot PUNNY and the AI creative factory PinGo Gallery to help users participate in early incentive programs.
Currently, PUNNY’s plan to attract early users has been completed, with official website data showing a total user count of 500,000 and 280,000 monthly active users, marking a very successful marketing campaign.
PinGo Gallery provides AI-based creative tools, allowing users to create based on keywords and share in Telegram.
I believe this tool can even serve as a foundation for creating Meme tokens. When certain popular memes emerge, the tool can quickly be used to create Meme images, issue Meme tokens based on them, and spread them directly on Telegram.
Whether it's simply creating memes in communication or ultimately pushing them to the market in the form of tokens, this tool has its uses and is considered one of the early tools to attract users.
「Only contributors have the right to use」
In PinGo's design, not everyone can use the decentralized computing power network; only those who contribute computing power to the network and become participants have the right to use it.
Participants can freely choose the size of computing power they contribute, and after providing computing power to the network, they will receive an NFT as a certificate. Only by holding this certificate can they use the computing power for AI training and utilize storage services for data and algorithm models.
This design largely avoids network imbalance caused by too many 'freeloaders' due to a lack of computing power contributors. In past experiences, such imbalances can lead to chronic network decline.
Although the mechanism design of PinGo may prevent a form of 'sybil attack' from causing a short-term surge in users, it essentially plays a positive role in promoting the project's long-term development.
Ultimate goal
Naturally, PinGo's goal is not just to act as a small tool on Telegram. In the future, as the network scales, PinGo can provide developers with the ability to create smart contract agents, AI investment advisors, AI sentiment trading tools, and other AI products based on information such as chat records on Telegram.
Apart from the product capabilities themselves, PinGo's roadmap shows that it plans to develop a blockchain based on its CDN network next year and further build it into a second-layer network of TON.
Developing an independent network will give the token PINGO and the project stronger value capture capabilities. As the frequency of network use increases, the demand for PINGO as the network currency will also continue to rise.
Once everything is completed, PinGo will become a large-scale AI infrastructure that can call upon computing and storage capabilities based on demand, providing users in need of training with low-cost resources, allowing them to avoid purchasing expensive centralized computing power and storage services. Moreover, for a considerable period, tokens can serve as a source of incentives for resource providers, relieving the network from the financial pressures of traditional data centers and enabling rapid expansion.
Since we mentioned tokens, it's worth noting here that the PINGO team and investors hold less than 20%, with 80% of the share used to incentivize computing power contributors, AI users, and CDN nodes, making the builders and users of the network the ones who can truly decide the project's final direction.
Decentralization is becoming an important complement to technological infrastructure.
If one were to say that decentralized infrastructure could replace large-scale data centers in a short time, it might be wishful thinking. However, many DePIN projects have already made the market aware of their important complementary role, and for some not overly complex demands, the services provided by Web3 projects are gradually being accepted by the mainstream market.
The reason why users in the domestic market lack sufficient awareness may be that the infrastructure for communication, payment, and other areas is already mature. However, the reality is that in many capitalist countries, due to insufficient interests, the level of infrastructure construction varies greatly across different regions within the same country.
This is why DePIN, stablecoin payments, etc., are seen as core areas of Web3 that seem somewhat unbelievable to us.
With billions of people worldwide, the egalitarianism brought by Web3 can bring the latest technologies to thousands of households and extend economic benefits to corners that capital is indifferent to.
Perhaps in the future, in a corner of some island nation, the infrastructure of Web3 can help the people there connect to the internet and use computing power that they originally couldn't afford to fulfill their small aspirations for technology.
