Businesses have, for decades, changed their infrastructure to support new applications and workloads. That process continues as the use of artificial intelligence becomes mainstream in more organizations. In fact, what many companies find is that even if they have upgraded or recently installed infrastructure for high-performance computing, they still need to do more.
That point was evident in a recently released IDC Worldwide Semiannual Artificial Intelligence Tracker. It found that hardware spending was the smallest of all AI segments (which also includes services and software) but was poised for great growth.
“Of all the spending in the various AI market segments, AI Hardware is by far the smallest,” said Peter Rutten, research vice president, Performance Intensive Computing at IDC. “What this should tell organizations is that nickel-and-diming purpose-built hardware for AI is absolutely counterproductive, especially given the fast-growing compute demand from increasing AI model sizes and complexities.”
That situation will change rapidly. The AI hardware category grew the most in terms of market share in the first half of 2021, with a jump of 0.5% share. It is forecast to have a year-over-year growth of 24.9%.
A familiar pattern with a twist
For years, businesses have upgraded compute infrastructure by moving to faster processors, higher performance storage, and higher speed interconnect technology. Many enterprises now routinely use technologies originally developed for government labs and academic supercomputing centers. For example, it is quite common to see businesses accelerate workloads using GPUs and co-processors, boost storage performance using parallel distributed file systems, or speed data between storage and compute systems using InfiniBand networks.
There has always been this evolution where new technologies were adopted to meet the increasingly more demanding workload. The same thing is now happening as AI becomes more pervasive. However, there is a twist.
In the past, only large enterprises needed to build up infrastructure (including compute, storage, and networking). They were the ones pushing the envelope with new applications and larger datasets.
But two things are different now. First, AI is being used by companies of all sizes and even in organizations that previously were not compute-centric before. And second, some AI applications are based on the collection and analysis of vast amounts of data from scattered IoT devices. In many cases, the organizations have no infrastructure in place to move the data from the devices to compute facilities.
AI infrastructure for agriculture
A recent use case highlights the AI infrastructure issues organizations can face and the need for out-of-the-box thinking.
One non-traditional market that is rapidly embracing AI is agriculture. Global spending on smart, connected agricultural technologies and systems, including AI and machine learning, is projected to triple in revenue by 2025, reaching $15.3 billion, according to BI Intelligence Research. And IoT-enabled Agricultural (IoTAg) monitoring is smart, connected agriculture's fastest-growing technology segment projected to reach $4.5 billion by 2025, according to PwC.
In many parts of the country, the use of this type of technology is relatively new. Many farms have no experience with the technology. And most lack the infrastructure to benefit from its use. Noting these factors, Sway AI, a developer of no-code AI technologies and services, partnered in February with Trilogy Networks, an edge computing and distributed cloud provider, to join the Rural Cloud Initiative (RCI) to enable AI technology for what is called precision farming.
RCI is a coalition of network and edge partners focused on accelerating the digital transformation of rural America. Its mission is to build such a coalition that includes rural telecom operators and AI technology providers.
As part of its work, Trilogy Networks and Inland Cellular teamed up to deploy a private wireless network for farming. It focused on a problem unique to agriculture. The issue was put into perspective by Trilogy's COO Nancy Shemwell in a press release about the work. "Even where I have connectivity to the farm, you have to get communications across the farm,” observed Shemwell. This will become increasingly important, she said, as precision agriculture increasingly requires “real-time decision making at the edge.”
The Trilogy Inland Cellular deal aims to provide that ubiquitous farm-wide coverage. It calls for Inland Cellular to offer private wireless to farms in its service area in the northwestern U.S. to blanket the farms with wireless coverage. In the Inland Cellular Trilogy deployments, data gathered on a farm’s private network using precision agriculture will be sent to edge data centers operated by Trilogy.
This is unique. Until now, private wireless networks have been the domain of large corporations or public entities that either built their own networks or purchased those networks as a service from a large carrier such as AT&T or Verizon. And it shows how AI requires special infrastructure that previously was not considered.