Data center chip architecture used for AI training phase 2017-2025

Full access: To this and over 1 million additional datasets
Save Time: Downloads allow integration with your project
Valid data: Access to all sources and background information

Published by Thomas Alsop, Nov 30, 2020

As of November 2019, application-specific integrated circuits (ASIC) are forecast to have a growing share of the training phase artificial intelligence (AI) applications in data centers, making up for a projected 50 percent of it by 2025. Comparatively, graphics processing units (GPUs) will lose their presence by that time, dropping from 97 percent down to 40 percent.

AI chips

In order to provide greater security and efficiency, many data centers are overseeing the widespread implementation of artificial intelligence (AI) in their processes and systems. AI technologies and tasks require specialized AI chips that are more powerful and optimized for advanced machine learning (ML) algorithms, owning to an overall growth in data center chip revenues.

The edge

An interesting development for the data center industry is the rise of the edge computing. IT infrastructure is moved into edge data centers, specialized facilities that are located nearer to end-users. The global edge data center market size is expected to reach 13.5 billion U.S. dollars in 2024, twice the size of the market in 2020, with experts suggesting that the growth of emerging technologies like 5G and IoT will contribute to this growth.