How can we fundamentally transform the deeply entrenched internet?

Author: Sending Network

In the past year, AI has stepped into the spotlight of the internet stage, sparking an unprecedented wave of excitement.

However, behind the booming concept of generative AI, venture capital activities in the entire AI field have been rapidly cooling down. According to a report from Stanford University's Artificial Intelligence Institute, the peak of merger and acquisition investments in the AI sector occurred in 2021. In the past three years, the amount of related investment transactions has actually been declining year by year.

The reason is not hard to understand: AI has become a game for the giants. Whether it's Microsoft, Google, Meta, or Apple, they are all procuring chips and collecting data for training on their own, and only these companies can gain advantages in computing power and data. For startups, starting an AI project from scratch is becoming increasingly difficult.

In the eyes of most people, this may just be a problem within the vertical field of AI. But is there a possibility that the "oligopolization" trend presented in the AI venture capital field hides a larger, underlying issue that has existed since the invention of the internet?

For example, when we connect to the internet, why is it so difficult to obtain a public IP? Why are IPv4 address resources becoming increasingly scarce? Why do various data breaches and privacy violations keep emerging, reminding us to change various passwords every so often? Why are the various internet services we need, from video sites and music memberships to cloud storage, becoming more and more burdensome?

These fragmented yet specific issues are actually all related to the design of the internet architecture. It can be said that the internet, which is closely related to our lives, is not sufficiently adapted to today's network scale and application forms, requiring a complete overhaul from top to bottom.

The Underlying Problems of Traditional Internet

To understand the underlying issues of the traditional internet, the AI field is a good entry point.

AI has high demands for data collection, storage, and computation, which come at a significant cost. This is a core factor in the rapid oligopolization of the AI field in its early development.

The gap in computing power is already well-known; AI training requires a large amount of GPU computing power, and this demand continues to grow with model iterations and parameter increases. The rush to purchase NVIDIA's most advanced GPUs has lasted for a year, driving NVIDIA's stock price to triple in the past 12 months. Many people lament that computing power is the most important resource in the AI era.

However, in addition to computing power, data is equally important. Currently, one of the biggest obstacles to the iteration of large AI models is data. In the past, large model vendors obtained data from the open internet to build their corpora, but as many social media sites have begun to block AI companies from scraping data, it has become increasingly difficult to construct a high-quality corpus. Moreover, the amount of data required for training the next generation of models is still increasing.

Take GPT as an example. According to the Wall Street Journal, the corpus used in the training of GPT-4 is approximately 12 trillion tokens, while GPT-5's data requirement is between 60-100 trillion tokens, an increase of 5-8 times. Such a scale of data is difficult for a single company to obtain. Furthermore, 99% of the data on the internet is user-generated, but the ownership of this data does not belong to the users, which is undoubtedly unfair.

Not only AI, but with the improvement of internet infrastructure, the volume of all data is rapidly expanding, from images, audio, and video to spatial models. The increase in data scale leads to high costs for servers and CDNs, and users' internet fees are also rising.

This cost is evident everywhere, from the continuously rising membership fees of video sites to the costs of 5G packages, and to the increasingly larger and more expensive mobile storage and cloud storage services, as well as the extremely high costs and thresholds for AI training today… The internet is not only becoming centralized and oligopolized, but it is also becoming increasingly expensive, and users have lost their voice in the entire process.

With such high costs for data exchange and low efficiency, many people may ask: Why can't we reuse the concepts from the past P2P download era to allow more people to participate in building the network, contributing their idle computing power, storage, and communication resources?

The core of the problem lies here.

The internet we use today has a foundational design that has always been a radiating tree structure, with servers at the top, CDN nodes in the middle layer for data distribution, and terminals at the bottom. Although two terminals can establish a direct connection through the network, such connections remain rare, and the vast majority of data still only flows between terminals and servers.

Thus, the core logic of Web 1.0 is "distribution," where the flow of information on the internet is unidirectional, more akin to television or radio, except that the distributed content has become a blog post, a piece of music, or a game, with the server hosting the information and the terminal requesting it forming a one-to-many relationship.

With Web 2.0, UGC emerged, and the flow of data began to become bidirectional, but it still only involved the flow between terminals and servers, without truly becoming networked. Because until the Web 2.0 era, the number of terminals accessing the internet globally was not sufficient, and the early terminals that could remain online for long periods were not the majority, requiring servers as the foundation of the network.

Friends with a bit more "internet age" experience may easily understand where the problem lies. Just like when we used P2P download software in the early days, we needed someone to "seed" to achieve high download speeds. P2P downloading is a completely decentralized network structure, but at that time, we often hesitated to let the download software run in the background on our computers to "seed" for others because there was no benefit, and our computers might often be turned off.

The Web3 revolution that everyone has been talking about in recent years aims to break the past "tree-like" centralized network structure, promoting interconnection between terminals and ultimately forming a decentralized network. One of the biggest changes in the Web3 era is that there are more terminals that can remain online and provide resources, such as our home routers and smartphones, which no longer "turn off." On the other hand, various tokens under Web3 provide an incentive mechanism for all "shared resource" behaviors.

To some extent, Web3 has already provided a complete theoretical foundation to address the internet's "low efficiency" problem and has the background conditions for change.

In fact, "inefficiency" is just one aspect of the traditional internet's problems. The basic structural framework of the internet we use today was established in the 1990s, and the engineers of that time could hardly have imagined that the internet would become as widespread as it is today. A series of network issues we face today, such as insufficient IPv4 address resources and the vulnerability of DNS servers to hijacking and pollution, can all be traced back to the underlying design flaws of the traditional internet.

For many years, there have been voices in the industry calling for the "updating and iteration of the internet's underlying framework," but due to the high switching costs, which require the cooperation of countries and vertical fields around the world, it has never been fundamentally resolved. Even IPv6 has never been fully adopted.

It can be said that the internet ship we are on today is already a "patched-up ship" that is increasingly unable to meet the demands of the times. Ultimately, someone must take on this mission to promote the bottom-up renewal of the internet.

Transforming the Internet with Decentralization

More and more people are realizing that transforming the internet means changing its overly centralized state, which is monopolized by a few giant companies and organizations. Entering the Web3 era, a large number of developers and enterprises have begun to focus on this issue.

From Bitcoin to NFTs, the entire Web3 industry seems to have been focused on applications—transactions, contracts, authentication… These applications constitute the first wave of the Web3 ecosystem.

However, the idea of decentralization is not just about solving specific application problems; it has the potential to thoroughly transform the underlying technology stack of the internet, overturning the efficiency and security issues caused by the current overly centralized internet.

DePIN has emerged. Although it may seem as unfamiliar as all the acronyms in the Web3 field, this term hides the greatest ambition in the Web3 domain. This group of developers is attempting to create a "decentralized physical network infrastructure."

Many DePIN projects have been designed from the outset to address a series of legacy issues in the traditional internet at the physical, network, and application layers. For example, the leading startup in this field in China, SendingNetworks, has proposed a top-down solution.

At the physical layer, they have designed a decentralized router hardware to solve the problem of excessive concentration of computing power and data communication resources under the traditional internet framework. Through a large number of decentralized routers in the network, data can be stored in a distributed manner, utilizing each user's idle bandwidth for transmission and access, thereby improving circulation speed and efficiency.

At the network layer, Sending Networks has addressed the most challenging IP address allocation issue of the traditional internet's TCP/IP protocol. Under the new network framework, users can directly access the internet through their wallet addresses, which serve as a natural "ID" in the Web3 era. Moreover, based on wallet addresses, communication protocols will become more secure and trustworthy, avoiding common attacks that exploit IP vulnerabilities.

Based on the decentralized physical and network layers, a variety of decentralized application layers can emerge.

For instance, at the most basic level, deploying a web site on a wallet address-based network can lead to faster access, and because this network is decentralized, it is also more secure and harder to attack; developers can also build a CDN with a larger node network based on the decentralized network, utilizing the idle bandwidth resources of each terminal to reduce content distribution costs and increase efficiency; and they can also build streaming media, IM software, and other applications that can leverage DePIN's core advantages: efficiency, privacy, and security.

Such a foundational revolution could ultimately drive AI applications to break free from oligopolistic control and return to the hands of every internet participant.

Today, large companies training AI first collect user corpus data from various UGC platforms like Reddit to form a corpus, and then conduct centralized model training. However, in a decentralized foundational network, the generation, storage, model construction, training, validation, and final application of data can all occur within a single network, with everyone participating in the process, supervising it, and sharing the results.

In such an environment, we need not fear that AI technology will be controlled by a few oligarchs, used to exploit users or even commit evil; all data and models will be accommodated in a more transparent and secure environment.

Currently, the DePIN track has the most decentralized projects related to AI, as the gaps in computing power and data during the AI training process have brought the issues of excessive centralization in the internet into the spotlight. However, a group of developers, including SendingNetworks, still believes that only by fundamentally transforming the entire chain of the internet from the physical, communication to application layers can we push it into a new future.

The First Step Towards the Future

Of course, this is not the first time someone has proposed to "disrupt the internet." While painting a beautiful future for DePIN, a bigger question still looms overhead: How can we turn it into reality?

The development and popularization of the traditional internet has been a top-down, unidirectional process. A very small number of computer scientists designed the underlying framework of the internet in its early days, and only after more and more enterprises, developers, and users participated in building the ecosystem did the internet achieve widespread adoption.

Because of this, despite the outdated and obsolete nature of the traditional internet's entire operating mechanism, it remains difficult to shake, becoming increasingly entrenched.

Therefore, to promote the entire internet world towards the future of DePIN, the most urgent need is to create new "incentives."

Once the internet has covered the vast majority of people on Earth, the revolution of the Web3 era cannot be completed overnight in a top-down manner as it was in the past; instead, it must be a bottom-up process, where the first batch of seed users drives developers and enterprises to change, gradually expanding into a transformation for the entire industry. Thus, the core of all protocols and applications in Web3 revolves around "incentive mechanisms."

SendingNetworks has chosen router products as the cornerstone for the cold start of the entire decentralized network, building a decentralized physical network and leveraging the financial attributes and incentive mechanisms of Web3 to construct the future of DePIN.

The reason for choosing router products as the breakthrough point is largely because routers are products that truly penetrate all levels of the internet. In this envisioned framework for the future internet, routers serve as both the information exchange terminals for users accessing the internet and as edge nodes that can remain online for long periods in a decentralized network, while also integrating and utilizing the storage and computing resources within the local area network.

Binding all of this together is the "consensus layer" designed by SendingNetworks. If users can ensure that their routers, as edge nodes, remain online and available for a long time while providing more communication, computing, and storage resources, they can convert this into quantifiable "contribution proof" on the blockchain and ultimately receive rewards.

The early DePIN network will be fully compatible with the traditional internet, ensuring that users' basic experiences are not compromised, but the spark of the decentralized network will be ignited, and the data flowing between decentralized terminals will increase, with value becoming increasingly evident, until it ignites the revolutionary fire of the entire internet infrastructure.

Many people may also be curious about how SendingNetworks, as a startup, can drive such a fundamental transformation. In fact, the SendingNetworks team originated from the early Microsoft Windows development team, and during the mobile internet era, they developed the Dolphin browser, which had 200 million active users.

From operating systems to browsers, for the past 20 years, the core team of SendingNetworks has been dealing with the most fundamental technologies of computing and the internet, addressing the issue of "how information flows, is distributed, and is presented." It is only in the Web3 era that they have gradually realized that to change the way information is distributed and presented, the internet itself must be transformed.

And now, they have taken the first step towards this ideal.