1/ Recently, AI compute storytelling has become popular again on Solana. Three names: ZERO / c0mpute ICPX / ICPX Protocol CPX / CapIX Protocol At first glance, they all talk about GPU, AI, computing power, and reasoning. But in fact, the three of them don't run the same store. Like an AI network. A GPU rental platform. A computing power exchange+router. 2/ Let's first use the most naive but understandable metaphor: ZERO: Opened a 'decentralized AI restaurant' You order, it's made by chefs from all over the world and served to you. ICPX: Opened a GPU rental car dealership Do you need a car? Choose a car model, pay, get the key, and drive it yourself. CPX: Opened a "computing version of Ctrip/Amap" Wherever you go, it helps you find the cheapest, fastest, and most suitable route. 3/ The keywords for ZERO/c0 compute are: AI powered by people, not data centers. What it wants to solve is: Why do all AI have to go through big companies like OpenAI, Anthropic, and Google? Why do you have to hand over your prompt to someone else's data center? Why can a model answer? It depends on others' mood? So what it does is a decentralized AI inference network. 4/ The product logic of c0 mpute is very straightforward: User: I want to chat/tune API/generate images/write code Worker: I have a GPU, I will run the model Network: I am responsible for matching, billing, and distributing tasks ZERO: After making money online, repurchase, destroy, and distribute to vendors It's not asking you to rent a machine. It allows you to directly use AI. Users don't need to worry about which GPU is behind it. Just like when you order takeout, you don't care whether the rider is riding Yadi or No. 9. 5/ The biggest catch for ZERO recently is Shard. Simply put, a single consumer grade GPU cannot run large models. Shard's idea is: Then cut a large model into several pieces and distribute them to different GPUs to run together. The biggest problem with this matter before was slow. Transferring across public networks is as slow as using 2G to flash short videos. Shard's narrative is: We ran it to be usable. This is the source of ZERO's imaginative space. 6/ So ZERO has the highest positioning limit. If it becomes, it is not 'an AI app'. It will become an AI inference network. In the future, others can also build on this network: Third party apps, models agent、coding tool， You can all come in. At this point, ZERO is not buying individual product revenue, But rather the inference flow of the entire network. This is the point of FOMO in the market. 7/ Take another look at ICPX. ICPX does not have the grand narrative of ZERO. It's more like practicality. It says: Don't talk too much, developers are currently lacking GPUs. What should I do if I want to run agent, fine tune, eval, batch inference? The answer from ICPX is: Come to me to select GPU, model, create job, pay with wallet, and open terminal. This is very similar to the RunPod/Vast/Akash style on the chain. 8/ Product keywords for ICPX: GPU rentals one-click deploys NVIDIA catalog SDK MCP server wallet payment terminal access pay-per-job compute It does' workflow '. You're not here to chat with AI. You are here to work. People like those who have model errors at 3am and the boss wants a demo in the morning, ICPX is very useful. 9/ The advantage of ICPX is its clear path. Real demand: Developers do indeed need GPUs. Realistic scenario: Agent, reasoning, training, and evaluation all require computing power. The product form is also easy to understand: pay by job, calculate by completion. But its problem is that the ceiling is not as sexy as ZERO's. Because the GPU rental market is too saturated. The question you need to answer is: Why not use RunPod? Why not use Vast? Why not use Akash? 10/ Take another look at CPX/CapIX. CapIX is the most "financial engineering flavor" among the three. It's not just about renting GPUs. It wants to do compute capacity and inference exchange. Translated into adult language: Transform GPU computing power and reasoning ability into something that can be bought, sold, routed, and settled. You buy CPX, You can rent direct capacity, Smart Route can also help you automatically find cheaper inference paths. 11/ The keywords for CapIX are: liquid capacity Smart Route route evidence minimum savings target direct machine time auto-routed inference CPX settlement What does it sound like? Like 'computing power version 1 inch+Ctrip+exchange'. You don't need to know which machine is where, Just say: I want to run this workload. It says: I'll help you find the route and give you a receipt. 12/ The narrative of CapIX is quite interesting. If computing power really becomes increasingly fragmented, Some GPUs are at home, some are in the computer room, some are on small platforms, and some are in idle inference services, That market does need a routing layer. Whoever can find cheap, fast, and available capacity has value. So the imagination of CPX is not 'I have a GPU', But 'I know where to find the GPU'. This is more like selling navigation than simply selling shovels. 13/ But the current problem with CapIX is also evident: The public docs are still locked, The white paper/technical details are not transparent enough, More validation is needed to determine how intelligent Smart Route is and how reliable its route evidence is. So its positioning is very good at telling stories, But the proof of arrival still requires additional classes. in a word: The concept is not bad, but the details still need to be submitted for homework. 14/ Let's take a look at the three together: ZERO is the entrance to the AI network. ICPX is a GPU task platform. CapIX is the computational routing and transaction layer. They all eat AI compute this bowl of rice, But chopsticks are different. ZERO is competing for 'who will serve the final AI request'. ICPX is competing for 'who will take on developer GPU jobs'. CapIX is competing for 'who will price and route computing power'. 15/ If we follow the narrative ceiling: ZERO is the first. Because it talks about a decentralized AI network. CapIX is second. Because capacity exchange has a financialization network effect. ICPX third. Because it is more like a tool type product, clear but less imaginative. 16/ If the clarity is based on the product path: ICPX is the first. Users want GPUs, platforms give GPUs, it's simple and crude. ZERO second. Chat/API/worker is easy to understand, but Shard network is difficult. CapIX is third. The concept is interesting, but Smart Route and Exchange require more public evidence. 17/ So how do you understand these three projects? If you are buying the protocol layer for future AI networks, you will pay more attention to ZERO. If you are buying 'Developers really need to use GPUs', you will pay more attention to ICPX. If you are buying 'computing power will be financialized and routed', you will pay more attention to CPX. All three belong to AI compute narrative, But not in the same betting direction. 18/ Finally, to summarize: ZERO is like building an AI city. ICPX is like opening a GPU rental company in the city. CapIX seems to be building a computing power exchange and navigation system for this city. Which one is the largest? See who really has users, income, developers, and repeat purchases. Narrative is responsible for igniting, The product is responsible for extending its lifespan, Revenue responsibility keeps everyone quiet. Not investment advice, just a project positioning review.