<p>我对过去几年 DePIN 加密生态系统的感觉是，大多数项目描述的应用场景和商业模式都是伪需求。</p>
<p>这是一个方面。</p>
<p>另一方面，我发现非常牵强的是，许多所谓的 DePIN 项目最初旨在利用闲置资源，但随着它们的发展，这些闲置资源不仅变得无用，项目也完全变成了被大玩家垄断的游戏。最典型的例子就是 Filecoin——从最初描述的允许家庭闲置的小硬盘参与存储挖矿，后来变成了存储服务器之间的竞争。不仅是 Filecoin，绝大多数后续的 DePIN 项目也是如此。一旦一个 DePIN 项目变成这样，它不仅未能充分利用闲置资源，反而消耗了更多的资源。为什么会这样？我认为一个重要原因是，在许多 DePIN 项目中，将闲置资源连接到 DePIN 网络的想法往往是，这些闲置资源在现实中根本无法完成网络分配给它们的任务，或者说，实际执行这些网络任务的硬件要求非常高。具有如此高门槛硬件的节点在现实中不太可能连接到这些类型的 DePIN 网络（除非专门为 DePIN 项目）。因此，在这些 DePIN 项目中，硬件要求低的资源节点几乎没有机会获得网络奖励。在这一点上，比特币挖矿与这些 DePIN 项目有显著区别。比特币的计算任务可以被分解和并行化，从而出现了矿池策略。这一策略允许高性能和低性能设备连接到矿池并参与挖矿。唯一的区别是，计算能力更强的设备获得更多奖励，而计算能力较低的设备获得较少奖励。因此，在比特币网络中，一旦满足某些硬件门槛，计算能力高低的节点都有机会获得奖励。在我看来，这是这两个系统之间的一个关键区别：不同资源持有的节点是否都有公平的机会获得网络奖励。在前者中，资源较少的节点几乎没有机会；而在后者中，资源较少的节点仍然有机会，只是获得的奖励较少。最近，一个名为 DePIN 的项目在 AI + Crypto 领域似乎走上了与比特币类似的道路，克服了之前 DePIN 项目面临的困境。</p>
<p>这个项目是 dphn.ai。它目前是一个专注于 AI 推理的去中心化网络，旨在激励运行各种大型模型（目前主要是各种开源大型模型）的系统（服务器、云等）加入网络进行 AI 推理工作。以运行 DeepSeek 的系统为例。这些系统作为节点加入这个网络。它们在计算能力、工作效率和生成代币的速度上有所不同。但这并不重要。在这些节点加入网络后，当网络收到处理使用 DeepSeek 的推理任务请求时，它会随机选择其中一个节点来处理该任务。任务完成后，系统运行验证机制来检查结果。如果一切顺利，该节点将获得奖励；否则，将受到惩罚。在分配任务时，网络随机选择一个节点。这个节点可能处理任务的速度快或慢，但所有节点都有平等的机会。然而，这种方法要求用户对性能反馈有一定的容忍度。因为如果选择了计算能力较低的节点来处理任务，用户将收到更长时间的反馈；如果选择了计算能力较高的节点，用户将收到更短时间的反馈。这个项目已经上线一段时间，目前的发展势头似乎良好。这似乎在验证 DePIN 在这个细分市场的可行性。然而，关于项目本身，它有几个明显的问题：在我看来，最大的问题是进入门槛不够高，尤其是在技术上。目前，它的突出之处主要是由于市场相对较小且缺乏关注。但如果这个模型在未来点燃推理领域的需求，必然会有大量竞争者进入市场。那么它的竞争优势在哪里呢？此外，这个系统目前主要依赖大型开源模型。它无法支持真正强大和高效的闭源模型，如 Anthropic 和 OpenAI。即使在它目前支持的开源模型中，越来越多的模型正在减少其开源范围和程度。如果这种趋势持续下去，当开源模型的能力变得过于薄弱而闭源模型又不再支持它们时，这个模型还能继续运营吗？所以让我对这个项目印象最深刻的并不是它的质量或投资价值，而是它似乎证明了 DePIN 在某些 AI + Crypto 领域是可行的。</p>

BTCS

ABTC

ETHE

<p>DePIN 加密生态系统面临挑战，许多项目未能有效利用闲置资源，并被大型参与者垄断。一个新项目 dphn.ai 旨在激励多样化的节点进行 AI 推理任务，使得无论计算能力如何，获得奖励的机会都是平等的。然而，它面临潜在的竞争和对闭源模型支持的限制。如果需求增加且竞争者进入市场，它当前的成功可能是短暂的</p>

<p>My feeling about the DePIN crypto ecosystem over the past few years is that the vast majority of projects describe application scenarios and business models that are pseudo-demands.</p>
<div class="lb-trans"><p>我对过去几年 DePIN 加密生态系统的感觉是，大多数项目描述的应用场景和商业模式都是伪需求。</p>
</div><p>This is one aspect.</p>
<div class="lb-trans"><p>这是一个方面。</p>
</div><p>On the other hand, what I find very far-fetched is that many so-called DePIN projects initially aimed to utilize idle resources, but as they developed, not only did these idle resources become useless, but the projects also completely turned into a game monopolized by large players. The most typical example of this is Filecoin—from its initial description of allowing even idle small hard drives at home to participate in storage mining, it later became a competition among storage server players. Not only Filecoin, but the vast majority of subsequent DePIN projects are also like this. Once a DePIN project becomes like this, it not only fails to fully utilize idle resources but also consumes even more resources. Why does this happen? I think one important reason is that in many DePIN projects, the idea of ​​connecting idle resources to the DePIN network is often that, in reality, these idle resources are simply unable to complete the tasks assigned to them by the network, or rather, the hardware requirements for actually performing these network tasks are very high. Nodes with such high-threshold hardware are unlikely to connect to these types of DePIN networks in reality (unless specifically for DePIN projects). Therefore, in these DePIN projects, resource nodes with low hardware requirements have almost no chance of receiving network rewards. In this respect, Bitcoin mining differs significantly from these DePIN projects. Bitcoin's computational tasks can be broken down and parallelized, leading to the emergence of mining pool strategies. This strategy allows both high-powered and low-powered devices to connect to mining pools and participate in mining. The only difference is that devices with higher computing power receive more rewards, and those with lower computing power receive less. Therefore, in the Bitcoin network, once certain hardware thresholds are met, nodes with both high and low computing power have a chance to receive rewards. In my view, this is a key difference between the two systems: whether nodes with different resource holdings all have a fair chance to receive network rewards. In the former, nodes with fewer resources have almost no chance; in the latter, nodes with fewer resources still have a chance, just receiving less reward. Recently, a project called DePIN in the AI ​​+ Crypto sector seems to have followed a similar path to Bitcoin, overcoming the predicament faced by previous DePIN projects.</p>
<div class="lb-trans"><p>另一方面，我发现非常牵强的是，许多所谓的 DePIN 项目最初旨在利用闲置资源，但随着它们的发展，这些闲置资源不仅变得无用，项目也完全变成了被大玩家垄断的游戏。最典型的例子就是 Filecoin——从最初描述的允许家庭闲置的小硬盘参与存储挖矿，后来变成了存储服务器之间的竞争。不仅是 Filecoin，绝大多数后续的 DePIN 项目也是如此。一旦一个 DePIN 项目变成这样，它不仅未能充分利用闲置资源，反而消耗了更多的资源。为什么会这样？我认为一个重要原因是，在许多 DePIN 项目中，将闲置资源连接到 DePIN 网络的想法往往是，这些闲置资源在现实中根本无法完成网络分配给它们的任务，或者说，实际执行这些网络任务的硬件要求非常高。具有如此高门槛硬件的节点在现实中不太可能连接到这些类型的 DePIN 网络（除非专门为 DePIN 项目）。因此，在这些 DePIN 项目中，硬件要求低的资源节点几乎没有机会获得网络奖励。在这一点上，比特币挖矿与这些 DePIN 项目有显著区别。比特币的计算任务可以被分解和并行化，从而出现了矿池策略。这一策略允许高性能和低性能设备连接到矿池并参与挖矿。唯一的区别是，计算能力更强的设备获得更多奖励，而计算能力较低的设备获得较少奖励。因此，在比特币网络中，一旦满足某些硬件门槛，计算能力高低的节点都有机会获得奖励。在我看来，这是这两个系统之间的一个关键区别：不同资源持有的节点是否都有公平的机会获得网络奖励。在前者中，资源较少的节点几乎没有机会；而在后者中，资源较少的节点仍然有机会，只是获得的奖励较少。最近，一个名为 DePIN 的项目在 AI + Crypto 领域似乎走上了与比特币类似的道路，克服了之前 DePIN 项目面临的困境。</p>
</div><p>This project isdphn.ai . It is currently a decentralized network focused on AI inference, with the aim of incentivizing systems (servers, cloud, ...) running various large models (currently mainly various open-source large models) to join the network for AI inference work. Let's take the system running DeepSeek as an example. These systems all join this network as nodes. They differ in computing power, work efficiency, and the speed at which they produce tokens. But that doesn't matter. After these nodes join the network, when the network receives a request to process an inference task using DeepSeek, it randomly selects one of these nodes to handle the task. After the task is completed, the system runs a verification mechanism to check the result. If everything goes smoothly, the node is rewarded; otherwise, it is penalized. When assigning tasks, the network randomly selects a node. This node may process the task slowly or quickly, but all nodes have an equal chance. However, this approach requires users to have a certain tolerance for performance feedback. Because if the node with low computing power is selected to process the task, the user will receive feedback in a longer time; if the node with high computing power is selected, the user will receive feedback in a shorter time. This project has been online for some time, and its current development momentum seems good. This seems to be validating the feasibility of DePIN in this niche market. However, regarding the project itself, it has several obvious problems: The biggest problem, in my opinion, is that the barrier to entry is not high enough, especially technically. Currently, its prominence is mainly due to the relatively small market and lack of attention. But if this model ignites demand in the inference field in the future, a large number of competitors will inevitably enter the market. Where will its competitive advantage lie then? Furthermore, this system currently primarily relies on large open-source models. It cannot support truly powerful and efficient closed-source models such as Anthropic and OpenAI. Even among the open-source models it currently supports, an increasing number are reducing their open-source scope and extent. If this trend continues, will this model be able to continue operating in the future when the capabilities of open-source models become too weak and closed-source models no longer support them? So what impressed me most about this project wasn't its quality or investment value, but rather that it seemed to prove that DePIN could be viable in certain AI + Crypto sectors.</p>
<div class="lb-trans"><p>这个项目是 dphn.ai。它目前是一个专注于 AI 推理的去中心化网络，旨在激励运行各种大型模型（目前主要是各种开源大型模型）的系统（服务器、云等）加入网络进行 AI 推理工作。以运行 DeepSeek 的系统为例。这些系统作为节点加入这个网络。它们在计算能力、工作效率和生成代币的速度上有所不同。但这并不重要。在这些节点加入网络后，当网络收到处理使用 DeepSeek 的推理任务请求时，它会随机选择其中一个节点来处理该任务。任务完成后，系统运行验证机制来检查结果。如果一切顺利，该节点将获得奖励；否则，将受到惩罚。在分配任务时，网络随机选择一个节点。这个节点可能处理任务的速度快或慢，但所有节点都有平等的机会。然而，这种方法要求用户对性能反馈有一定的容忍度。因为如果选择了计算能力较低的节点来处理任务，用户将收到更长时间的反馈；如果选择了计算能力较高的节点，用户将收到更短时间的反馈。这个项目已经上线一段时间，目前的发展势头似乎良好。这似乎在验证 DePIN 在这个细分市场的可行性。然而，关于项目本身，它有几个明显的问题：在我看来，最大的问题是进入门槛不够高，尤其是在技术上。目前，它的突出之处主要是由于市场相对较小且缺乏关注。但如果这个模型在未来点燃推理领域的需求，必然会有大量竞争者进入市场。那么它的竞争优势在哪里呢？此外，这个系统目前主要依赖大型开源模型。它无法支持真正强大和高效的闭源模型，如 Anthropic 和 OpenAI。即使在它目前支持的开源模型中，越来越多的模型正在减少其开源范围和程度。如果这种趋势持续下去，当开源模型的能力变得过于薄弱而闭源模型又不再支持它们时，这个模型还能继续运营吗？所以让我对这个项目印象最深刻的并不是它的质量或投资价值，而是它似乎证明了 DePIN 在某些 AI + Crypto 领域是可行的。</p>
</div>

New Observations on DePIN in the AI + Crypto Field

- DePIN 生态系统面临假需求和资源浪费的问题。  
- Filecoin 案例显示大玩家垄断了资源，难以充分利用闲置资源。  
- 新项目 dphn.ai 随机选择节点处理任务，可能验证 DePIN 的可行性。

在人工智能与加密领域中关于 DePIN 的新观察