Pantera、Multicoin、 Sequoia betting on Gradient In the cryptocurrency industry, there are not many truly technologically advanced decentralized A1 infrastructure projects. Gradient is restructuring Al's underlying operating system to enable everyone to participate in large model training and inference, rather than relying on giant data centers. Pantera、Multicoin、 The investment from top institutions such as Sequoia also confirms its potential. What does Gradient do? Gradient is a next-generation network designed for decentralized AI infrastructure, dedicated to addressing two core issues: 1. How to further reduce the threshold for fine-tuning model training, so that all industries can train their own models, and the ability to train models is no longer monopolized by a few companies. Gradient, through the distributed Echo training framework, enables fine-tuning of large models to occur in distributed network environments and devices, further reducing model training costs; 2. How to lower the threshold for model inference and enable every household to use their own large model. Gradient uses the distributed inference framework Parallax to enable Macbook/Windows PCs to assemble supercomputers that can run models like DeepSeek R1. This not only enables local deployment of large models with strong privacy protection, but also enables the world to form a "world computer" and achieve efficient and scalable inference services. I also took a long time to understand what they are doing. Gradient is building a "highway" for AI, not another ChatGPT, but the underlying operating system for AI. It aims to enable anyone and anywhere to access the network and run intelligent models, allowing the models to run on globally distributed nodes. Why is Gradient capable? Top team endorsement Investors: Pantera Capital, Multicoin Capital, Sequoia China (HSG) Financing scale: $10 million seed round Team size: about 40 people, mostly with technical backgrounds Founder background: Eric Yang - Former Sequoia Capital investor, knowledgeable in strategy and ecology, skilled in combining technology and capital; Yuan Gao - Master's degree from Columbia University, former Head of Growth at Helium Foundation, led the global expansion of centralized networks from 0 to millions of nodes in the past; Upon closer inspection, the capital side is very top-notch. The founding team's genes are typical of a "finance+technology hybrid", understanding both capital operations and writing underlying protocols. Among them are ACM gold medals, alumni of the Yao class, and top university researchers. The overall temperament is similar to the early DeepSeek, low-key but ambitious. Personally, I feel that the team style is extremely down-to-earth: do research first, and then produce products. In addition, their technical papers have been submitted to multiple top conferences, and currently 8 research results are being submitted to the top conferences. The integration of top tier funds, hardcore technology, and mature models, in the direction of AI, is personally optimistic, and there should be another round of major financing in the future. How to do Gradient? After carefully reading Gradient's blog document, my biggest impression is: Gradient is moving AI from the data center to every computer, not making an application, but rebuilding the way AI operates. The traditional big model is like a huge "super brain". To make it work, it needs to be placed in the data center of a large company, using hundreds or thousands of expensive GPUs. Gradient wants to break down this "super brain" and distribute it. Anyone can use their own devices, computers GPU、 Even your Mac and laptop can be connected to run large models, contribute computing power, and receive rewards. This is like a decentralized AI cloud, working together to accomplish tasks that were originally only possible in supercomputing centers. A new paradigm of intelligent inclusiveness and shared benefits for all The key to implementing this intelligent network relies on the collaborative work of Gradient's black technology components, which are like the "nervous system" of the entire distributed AI network: -Echo disassembly training -Parallax disassembly reasoning -Lattica Efficient Data Transmission -Protocol Coordination Incentives Echo: It is the "training engine" of Gradient, which transforms the process of fine-tuning model reinforcement learning into a continuous process of absorbing nutrients from the world's computers. Massive devices propagate knowledge to Echo by outputting model inference results and self-evaluation results, while Echo continuously acquires knowledge, iterates the model, and further sends the iterated model to a massive number of devices for the next round of model iteration, repeating the cycle; Parallax: It is the "inference engine" of Gradient, equivalent to the brain division system of AI. It can automatically cut a large model into small pieces and allocate them to different devices to run together. Whether it is a MacBook, PC, or GPU server, they can all participate in the calculation, just like multiple people playing music together. Each node is only responsible for a part, but together they form a complete melody; Lattica: a data neural network equivalent to Gradient, responsible for securely transmitting data, model parameters, and logs between different nodes. Based on point-to-point communication, it does not rely on centralized servers, has fast speed, and saves bandwidth. The daily transmission volume has now exceeded 10TB, like a decentralized AI exclusive highway. In summary, it is to enable data to flow smoothly; Gradient Protocol: It is equivalent to the incentive and governance system of Gradient, used to record computing power, tasks, and rewards. When users contribute GPU, data, etc., they can receive token incentives, allowing the entire system to not only run, but also operate and motivate itself, allowing participants to "receive rewards". Among them, Gradient Cloud is built to provide AI inference services that charge by volume. Currently, it supports large models such as Qwen3-235B and GPT-OSS-120B, with a price range of 0.09-1.50 per million tokens. Request traffic will be automatically allocated to Parallax nodes worldwide, achieving decentralized inference. The combination of the three enables the entire system to run, spin, and motivate itself. Participation opportunities 1. Previously promoted EXP points (AFK), S1 ended on August 28th 2. The community will hold some activities around product promotion, and the community is considered active. Active community builders may receive rewards such as airdrops. Everyone can pay attention to it 3. Recently, ICOs have been very popular and there may be opportunities to get into the market. 4. Perhaps more user activities will be launched later. Daily Chat usage: https://chat.gradient.network/