What do we think of the DePIN track?

Author: Yolo Shen; Source: Foresight Ventures

Traditional computers have five components: Computer, memory, controller, bus, I/O. From the perspective of the development of blockchain, the development of calculators and memory is relatively complete. If the entire distributed system is compared to humans, then the human brain and memory system are already complete, but the perception and perception system is still in a very primitive state. At this stage, Depin is undoubtedly the most popular buzzworld. How to achieve it? All this undoubtedly needs to start with "trusted touch", and it is well known that "feeling" needs to rely on the spine and nervous system to process.

If the blockchain system is to build consciousness on the iceberg, then the sensor network represented by the bottom layer of Depin is the subconscious under the iceberg. Now the challenge is here, who is the spine and nerves of the distributed system? How do we build the spine and nerves? In this article, we will start with the small lessons of the development of the Internet of Things, build the development ideas of Depin, and help buidlers land better.

1. Historical Review of the Internet of Things

Looking back at the development of the Internet of Things since 2015, there were two main difficulties at the time: First, the hardware devices had a single input and output. Second, the product features were not enhanced after the devices were connected to the network, and the devices did not have the characteristics of scale.

During this period, the core issue was: What changes will occur after the microcontroller of the hardware device is connected to the network? First, from the perspective of the action of connecting to the network, it is possible to upload and download hardware devices. The next question is, why does hardware need to upload and download? Can uploading and downloading increase product competitiveness? During that period, we saw a number of products such as smart curtains and smart air conditioners. Since the I/O dynamic lines of the hardware were relatively determined at the beginning of the design, the space for software development was relatively limited. Therefore, after being connected to the Internet, the product features only had one feature, mobile terminal control, similar to "remotely turning on the air conditioner and remotely pulling the curtains". Most of the functions were remote + traditional controllers. For C-end users, this design is indeed a bit tasteless. Another core question is whether IoT devices have the ability to scale after being connected to the Internet? As we mentioned earlier, hardware has more uploads and downloads when it is connected to the Internet. If downloading is equivalent to upgrading and expanding functions, then uploading is the aggregation and integration of data. The value of the latter data lake was very cumbersome in the early IoT era. There was a huge contradiction between the exponentially rising storage costs and the difficult-to-develop data sales.

In summary, IoT devices cannot improve product strength and service dimensions in both download and upload modes. So is it possible in the Depin era?

What changes has AI brought?

From the characteristics of AI, we see many possibilities:

1. Everything is anthropomorphic, with independent upload and download requirements. If the edge side cannot reason about the large model, the end side needs to be independently connected to the Internet. This will transform the past convergence structure with mobile terminals as stars and devices as satellites into a communication structure with independent device networking.

2. Device sovereignty. From simple product sales, to the dual-wheel drive of purchasing from users + data sales. The device is responsible to the user as a whole, and to the data provider as a collection of sensors.

3. "Data is reliable, privacy is reliable" is the prerequisite for the transformation of ordinary devices into mining machines. If the data is not reliable, then logically opening multiple virtual machines can hack the entire incentive system; if privacy is unreliable, then in the long run, the user's willingness to interact will be suppressed.

Combined with the development of Ai, we see that Depin may have some different possibilities:

1. The emergence of AI has increased the necessity of autonomous networking of AI hardware. The cost of equipment access to the network may drop rapidly in the next three years. Combined with the decline in storage costs and computing power costs, the layout cost of edge computing power/sensors is also likely to drop. So many devices have been laid out, and converting them into mining machines to collect sensor data will appear a tipping point.

2. After the problem of independent connection between devices and the cloud is solved, there will be more scenarios for the interconnection between devices. How to use various low-cost NFC hardware for interactive gameplay will also become a potential innovation point.

3. Commoditizing various types of collected perception data is the core bottleneck of equipment mining. If the standard of abstract information commodities is formulated, it is the main challenge.

2. Depin's investment themes and views:

Based on the development experience of IoT in the past five years and the changes in new AI features, we believe that there are three major investment themes:

• Hardware infrastructure with cellular modules as the core.

• Abstract communication layer services with communication information commodities as bulk products.

• Generalized miners as a kind of dealer service.

Investment theme 1: Depin infrastructure centered on address bus modules

What is a module?

Modules are functional modules that integrate baseband chips, memory, power amplifier devices, etc. on a circuit board and provide standard interfaces. Various terminals can achieve communication functions with the help of wireless modules. With the development of the entire computing network, the definition of modules has been continuously enriched, forming an ecosystem of cellular networking + computing power + terminal-side applications:

• Traditional cellular IoT modules: Basic connection modules, whose main function is to achieve cellular communication. These modules only contain chipsets that support this connection and do not require additional functions.

• Smart Cellular IoT Modules:In addition to providing connectivity like traditional modules, additional computing hardware is integrated in the form of central processing units and graphics processing units (CPU and GPU).

• Artificial Intelligence Cellular IoT Modules:In addition to providing the same functionality as smart cellular IoT modules, dedicated chipsets for AI acceleration, such as neural, tensor or parallel processing units (NPU, TPU or PPU) are also included in the connectivity modules.

From the perspective of the entire industry chain, upstream chip and downstream equipment manufacturers occupy most of the value chain, and the middle module layer is characterized by high market concentration and low gross profit. The devices for traditional services mainly include: PC, mobile phone, POS machine. Due to its huge concentration, once a module middle layer with broad consensus is deployed, it is equivalent to naturally migrating various stock devices into mining machines. If the users of traditional Web3 are in units of people, then the middle layer represented by modules will enable a large number of smart devices to enter web3, and the Tx between these devices will generate a large number of on-chain demands.

Looking back at the early competition between NVIDIA and Intel, we have gained a lot of historical experience: the computer chip market was dominated by Intel CPU X86 systems in the early years. In some marginal markets such as graphics acceleration, there was competition between the acceleration card ecosystem dominated by Intel and NVIDIA's GPU; and in the broader market (areas with uncertain demand), Intel CPU and NVIDIA GPU cooperated with each other. The two companies coexisted and prospered for a while. The change point appeared in Crypto and AI. A large number of computing tasks are characterized by small tasks and large parallel behaviors, which are compatible with the computing characteristics of GPU machines; and when the wave came, NVIDIA made preparations in several dimensions:

1. Cuda's parallel computing instruction set. Help developers make better use of GPU hardware.

2. Fast iteration capability. The iteration speed that exceeds Moore's Law has won it a living space.

3. Competing with CPU. Effectively leveraged and utilized Intel's stock resources, and quickly seized market opportunities in some decision-sensitive areas.

Back to the module market, there are several similarities with the competition landscape of GPU and CPU in those years:

1. High industry concentration, the head group has a strong pricing power for the entire industry.

2. Development depends on new scenarios, and communication modules + smart chips + standard protocols are very likely to lay a moat on the device side.

3. Rapid iteration has the opportunity to seize new opportunities, traditional players have a long decision-making cycle, emerging scenarios change quickly and have risks, and the environment is suitable for the birth of new species.

In this competition, Crypto Stack is undoubtedly the best technology stack for building protocols and ecosystems, and the migration of stock equipment to cash flow mining machines will generate β-level opportunities. Among them, Dephy is an important player, which realizes the distribution and management responsibilities of the entire Depin network by building an integrated module + Ledger + identity layer.

Investment theme 2: Data bus sensor-represented data collection mining machine

What exactly is a mining machine? We believe that hardware/software that can generate specific information resources and intend to obtain token resources can be called a mining machine. In this understanding, there are several standards for mining machines:

1. Does it generate specific information resources?

2. Can it settle tokens?

Therefore, in the entire process, whether the PoPW (proof of physical work) generated by the device is credible becomes very important, so we believe that every sensor with the ability to produce PoPW needs a trusted (TEE/SE) to ensure the credibility of edge data collection. In the field of sensors, many horizontally scalable networks can be generated. For example, the video resources collected by cameras of different devices will be uniformly measured in one network. Compared with the independent collection of different devices, horizontally expanded sensors + trusted modules can build a larger PoPW resource market. The collected video materials can be better priced according to unified measurement, which is more conducive to the formation of a bulk market for information resources. This is not available in Device-Focus.

Investment Theme 3: Communication Infrastructure of Control Bus Broad Bus

Since some Depin devices exist in the physical world and are related to the traditional business society, and the Crypto world is characterized by PermissionLess, it is particularly important to manage various participating entities in the absence of KYC. We believe that the entire web3 world needs a communication abstraction layer that integrates cellular networks and public IP networks. Users/devices only need to pay Crypto currency to obtain corresponding network services. The specific directions are:

1. Integrate traffic. Connect the traffic resources of global operators, treat traffic as a bulk information product, and trade and price it with tokens.

2. Integrate number segments. Connect global number segment resources, treat numbers as an identity layer, trade and price them with tokens, and use Blockchain as the governance system.

3. Integrate IP resources. Connect public IP resources, integrate public IP pools as a resource, use public IP as a resource pool, support arbitrary access jumps, trade and price them with tokens, and use Blockchain as the governance system.

III. Conclusion

1. Depin should not be based on devices, as devices do not have horizontal scale capabilities; the core of depin is Pin, and the core of Pin is the authorization code. We regard devices as a collection of sensor modules, and the pin code of each sensor module is the permission for data access to the network, and also the authentication permission of PoPW. Only devices with network access rights and recognized contributions can be called mining machines. Therefore, the core of the entire Depin track lies in how to make the contribution of edge devices measurable, and how to make the contribution of the same sensor of different devices have consistent measurement.

2. According to the different ways in which traditional computers transmit information, they can be divided into three categories: Data Bus for transmitting various data information; Address Bus for transmitting various address information; Control Bus for transmitting various control signals. The DePin bus will also have similar ones: as an identity certificate for device access to the network. (Address Bus) as a PoPW certificate for data verification. (Data Bus) as a means of device management. (Control Bus)

3. Since some of the Depin projects have RWA attributes, they exist in the physical world and are related to real economic life. Therefore, more in-process management methods are needed to achieve autonomous risk control. There are two main implementation channels: First, through the management of cellular operator traffic, once a device violates the rules, the device can lose the PoPW mining rights from the traffic end, which is a more real-time management method compared to Slash. The second is to buy out upstream resources through miners + resource pools. For example, a dealer has 100 number segment resources. When 30 of them are at risk, he may be punished with a warning of license revocation. Today, we will mix the 30 resources with the resources of other dealers, buy out the real resources (RWR) through miners, and control the resources through mixed number segments. Under the premise of ensuring the risks of upstream dealers, we can obtain as many resources as possible. Replicate the Liquity model to various RW resources.

Disclaimer: All articles of Foresight Ventures are not intended as investment advice. Investment is risky. Please evaluate your personal risk tolerance and make investment decisions prudently.