Skip to content

Future Cloud for AI (macro trend)

Fizz edited this page Dec 9, 2021 · 11 revisions

AI Platform (AI Cloud)

  • Unified Data and AI. Bring data loading, preprocessing, feature store together to training and serving, same platform
  • Anyscale (0->Max) & Anywhere (public cloud, private cloud, edge).
  • AutoML. end-2-end process automation, hyper-parameter tuning and input space auto search
  • ModelOps. Extend MLOps, focus on enterprise model operationalization
  • Responsible and Trusted AI. From data, algorithm to prediction results, all should be explainable
  • Highlighted Companies
    • CoreWave, a GPU-first cloud for compute-intensive workloads
    • Run:AI, a platform for AI virtualization and orchestration, with advanced GPU pool management and Kubernetes based scheduling.
    • CNVRG, a full stack machine learning operation system, all-in-one ML platform, optimized for the on premise enterprise.
    • Anyscale, a universal framework for distributed computing, Ray applications, eliminates infrastructure and cluster management.
    • Omniverse, a scalable, multi-GPU real-time reference development platform for 3D simulation and design collaboration.
    • Modzy, a secure ModelOps and MLOps platform for businesses to deploy, manage, and get value from AI at scale.
    • DataRobot, AI cloud, a unified platform for all users, all data types and all environments, convert insights to decision

New Application (driver)

ML Framework

  • Parallelism/resolution/fault tolerance...
  • Data loading
  • Compiler

New Infrastructure

  • Hardware changes: Compute/Storage/Network
  • Architecture changes: Disaggregated/Inline

New Model

  • Cloud of Cloud
  • Distributed Cloud