Cloud Or Colo? The Point When AI Startups Need To Make The Switch

Whitepaper

Published September 2021

X

An AI startup, like any other startup, builds a business plan that evaluates a market and delivers what it needs. There are simple and relatively inexpensive ways to get started on the public cloud, with IT infrastructure loaded up with GPU or FPGA accelerators, or hefty CPUs that can do a respectable amount of floating point and integer math that machine learning training and inference workloads require.

Today, any startup can go to one of the big public clouds, which sell AI as-a-Service (AIaaS) for training and inference, in some cases using their homegrown ASICs or AI engines fashioned from FPGA logic. The cloud giants have plenty of raw CPU compute on hand in every imaginable configuration of processor and memory, flash, and block storage, and with fast Ethernet and sometimes InfiniBand network interconnects.

Biting the hand that feeds IT © 1998–2022