sluke

sluke

twitter
HomeArchives

Building Highways for AI Agents (2)

#AI48
AI Translation
This post is translated from Chinese into English through AI.View Original
AI-generated summary
AI highway infrastructure materials are ready for use. Clients can access services from different AI power centers, such as data centers or personal devices. Various AI models and workflows can be assembled and deployed, with AI power boxes available from different manufacturers. Different players in the AI highway have different strategies, with small developers focusing on user numbers and larger players on infrastructure resources. Payment models vary, with developers of large models focusing on advertising and edge computing costs, while GPU manufacturers focus on chip sales and telecom operators on monetizing idle computing power. Ultimately, individual users may bear the cost in the long run.

AI Highway Infrastructure Materials Already Prepared#

1. Simplified AI Highway Architecture Diagram#

simple-highway-structure.png

  • Different clients can obtain services from AI routes and different AI computing centers.
  • AI computing centers can be data centers or devices deployed in the network, such as AI computing boxes.

2. Everything on the Internet#

2.0 Slightly More Complex AI Highway Architecture Diagram#

simple-plus-highway-structure.png

  • AI capabilities are assembled into workflows.
  • AI API routes to different computing centers, including routing to different models.
  • Complete billing and payment, and let AI models help you make money.

If it is difficult to understand, you can refer to the downloaded models for BT/PT, use your own computer as a seed, upload files, earn points, and get faster download speeds.

2.1 AI Computing Box#

Various manufacturers have launched AI computing boxes, including major players such as NVIDIA and Qualcomm, as well as maker players and various server manufacturers.
highway-for-AI-NVIDIA-Jetson.png
highway-for-AI-ZimaCube.png
highway-for-AI-ai-in-box.png

In China, there will be a market for domestic AI computing boxes, which are mainly for domestic chips, especially domestic graphics cards, and the main customers should be government and enterprise customers with high data compliance requirements.

  • The requirements for model inference are not as high as for training.
  • The limitations of high-end processes have not yet spread to inference chips.

2.2 Model Deployment#

For example: https://ollama.com/download
There are also various tools for local deployment of AI models.

2.3 AI Workflow#

For example:

2.4 AI Clients#

  • Various local clients, such as chatbox, chatx, opencat, etc.
  • Various web services, such as poe.com
  • Browser plugins, such as chathub, monica, etc.

Strategies of Different Players on the AI Highway#

1. Players#

  • Small developers: Enter the market from the C-end, and the number of users is important.
  • Major players: Enter the market from the computing center side, and infrastructure resources are important.

2. Commercialization, Who Pays#

  • Developers of large models (big companies)

    • During the market launch period, there is a demand for promoting their own models: advertising fees
    • During the prosperous market period, there is a demand for reducing usage costs: purchasing edge computing power

  • Graphics card manufacturers

    • Mainly the demand for selling chips on a large scale: advertising fees
    • They can also be beneficiaries of operating computing networks.

  • Telecommunication operators

    • Monetize idle computing power, package new product lines, and increase stock prices: advertising fees

  • Individual users, in the long run, are the ones who will bear the costs.

Loading...
Ownership of this post data is guaranteed by blockchain and smart contracts to the creator alone.