Watching Netflix, searching Google, calling an Uber, switching off the lights with Alexa, unlocking your phone, even selecting the right shade of makeup – these are just a few examples of how we interact with artificial intelligence and machine learning (AI/ML) every single day.

At its core, AI/ML is about pattern recognition. The ability to recognize patterns in real time has countless possibilities to improve business processes, enterprise outcomes, and people’s lives. IDC estimates that the worldwide AI market, including software, hardware, and services, will reach the $900 billion mark in 2026, with a compound annual growth rate (CAGR) of 18.6 percent in the 2022-2026 period.

As more and more organizations adopt AI/ML, their IT teams will need to focus on the practicalities of how to cost-effectively build and manage an infrastructure that can support these powerful abilities and scale for future growth. And if there’s one component of that process that remains especially underestimated and misunderstood, it’s the data storage infrastructure that’s required to support these emerging applications.

Here are four common AI/ML storage myths that need to be busted.

1 - AI/ML is all about the GPU

Before the emergence of modern graphic processing units (GPUs) with extreme computational power, the AI/ML applications and neural networks in use today were nothing more than a fascinating concept. The accelerator silicon is without a doubt critical for AI/ML applications, but it is equally worthless without adequate storage and networking.

Storage and networking are the hands that “feed the beast.” They ensure that the next set of data is available to the accelerator before it has finished with the current set. The choice of storage and networking infrastructure therefore must be considered just as carefully as the GPU. In fact, each element must be balanced to achieve the optimal result: too much storage performance or capacity will be wasted, while too little will leave expensive computational silicon idle.

2 - AI/ML requires high-IOPs all-flash storage

To “feed the beast,” the accelerator requires data to be available whenever and wherever it is needed. This means that AI/ML storage is not simply about pure speed. Expensive all-flash storage systems with impressively high IOPs could very well be a waste of budget.

Accelerators have varying levels of performance, as do different AL/ML applications. For example, the computation per image in object recognition applications takes long enough that a hybrid (hard-disk drive and solid-state disk) system would work just as well as an all-NVMe solution – at a much lower price. IT teams need to balance their compute accelerators, AI/ML workloads, with their storage options to find the optimal solution. Independent benchmarks such as MLPerf can help here.

3 - Storage tiering will reduce AI/ML costs

Tiered storage is a common strategy to maximize storage resources and minimize costs. “Hot” mission-critical and frequently accessed data is placed on expensive and fast storage media (i.e., SSDs), while “cold” archival data that is very rarely accessed or updated is kept on the cheapest storage devices (i.e., tape). Although a widely adopted approach to cost-effectively manage storage requirements, this model cannot be applied to AI/ML applications. That’s because there is no such thing as cold data in AI/ML.

All AI/ML training data is used on every training run, and so tiering some training data off to different layers of storage will simply slow down the process. Instead, AI/ML storage solutions must treat all data as “hot” and ensure all data is always available.

At the same time, the accuracy of AI/ML workloads increases with the volume of training data available. This means that the storage infrastructure must be able to scale without disruption as training data volumes expand. Scale-out linear growth, in contrast to storage tiering, is a key storage requirement for these environments.

4 - AI/ML can make effective use of a dedicated single-use storage system

AI/ML is of most value when applied to an organization’s core data. For example, banks are adopting these technologies for fraud detection, and drug manufacturers can better analyze data from experimentation or manufacturing to speed up drug development. In the case of Amazon’s grocery stores or IL Makiage, the AI-powered makeup company, AI technologies are at the core of its technology and business infrastructures. For many businesses, AI/ML are no longer experimental side projects that could be served by a dedicated single-use storage system; rather, they have become an integral part of the business. As such, these applications must be consolidated into the organization’s core IT infrastructure and storage solution.

Choose wisely...

AI/ML innovations are set to drive massive transformations across the enterprise and impact nearly every aspect of an organization. Many technologies are expected to reach mainstream adoption in the next two to five years, such as edge AI, decision intelligence, and deep learning, according to the Gartner hype cycle. As organizations embark on their own individual journeys to apply this powerful new technique, the choice of underlying storage infrastructure will have a major impact on organizations’ ability to maximize the potential of AI/ML applications.