How Enterprise AI Can Ease The Data Gravity Burden

COMMISSIONED:  Among the many tough decisions IT leaders face is where to best host AI workloads.

Whether you’re deploying traditional AI applications that predict supply chain performance or launching generative AI digital assistants that serve information to customers, the location options for such software are myriad.

Yet the calculus has changed. For the first time in probably a decade, the public cloud is not necessarily the first destination that you IT leaders turn to. In fact, Gartner expects that 50 percent of critical enterprise application workloads will reside outside the public cloud through 2027.

You know, of course, that many factors go into workload placement. Yet it almost always starts with a single consideration: your data. And AI demands a lot of it. Especially the modern GenAI applications that are transforming digital operations worldwide.

Avoiding Cloud Sins Of The Past

GenAI apps create vast volumes of unstructured data such as text, audio and video. The more data such apps generate, the more that data gravity weighs on an organization’s workloads and the harder they become to move.

Ironically, you IT leaders have likely already experienced this challenge during the “cloud-first” phenomenon of a decade ago, as your organizations rushed to capitalize on the agility and innovative prospects promised by the public cloud.

In recent years – and for various reasons – you and your peers have probably removed some application workloads from the public cloud.

Some of you realized your providers couldn’t meet data locality regulations. Others erred in refactoring applications for the public cloud only to watch them break. Still others found the more data they created, the more expensive it became to run their applications.

To put this into a consumer-friendly perspective, consider the platform shift from cable TV to streaming services over the past few years. Tired of feeling locked into contracts that included hundreds of channels they rarely watched, millions of customers “cut the cord” on cable TV, switching to Netflix, Hulu, Apple TV+ and other services.

The public cloud has come to mirror cable TV, with providers launching dozens of services that IT leaders don’t need. Moreover, as with cable TV, IT leaders came to feel locked in to their existing cloud contracts; when the renter’s remorse kicked in, they began to move some workloads.

And with compute and storage needs diversifying today, you IT leaders want more Netflix and Hulu, less cable TV, which is a big reason for some of the repatriation to on-premises systems.

A Different Tack To Support AI

Your organizations require a more prescriptive approach to their AI needs – one that affords them control over their corporate IP and data – while allowing them to maintain performance and meet resiliency requirements.

Also, AI inferencing can tax systems – it requires real-time access to data – so organizations must control the compute and storage that fuel their AI applications. Controlling these resources will also help your IT departments better manage the data gravity that attends AI applications.

How can you, as an IT leader, solve for such challenges?

There is no cookie-cutter approach, but one option includes maintaining a laser-focus on delivering the optimal business outcomes, regardless of where you choose to run your applications, whether this includes your own datacenters, with the ability to extend to AI PCs and other devices at the edge.

This approach, known as helps you meet performance requirements and reduce latency while ensuring that your data is secure and compliant with data locality and sovereignty rules. There are paths to enterprise AI that may be kinder to your budget.

For example, you might consider deploying open-source models on-premises, which helps you bring AI to your data while right-sizing your model(s) to meet your operational requirements. Pre-trained models incorporating retrieval augmented generation help refine results with corporate data and run well on GPU-powered servers behind the corporate firewall.

Deploying Meta’s Llama 2 model on-premises with RAG proved as much as 75 percent more cost-effective than running GenAI workloads in Amazon Web Services’ public cloud, according to a survey conducted by Enterprise Strategy Group. ESG also found that running Mistral’s 7B open-source model with RAG on premises was found to be 38 percent to 48 percent more cost-effective than AWS.

These are key savings at a time when the cost of inferencing rises over the lifetime of a model.

Cultivating enterprise AI isn’t easy; most organizations lack the technical wherewithal to build such an operating model let alone stand up the architecture and infrastructure modernized for AI.

It also requires experts who can help you get your data house in order so you can run your models with the performance and efficiency you need at the right cost. Who you choose as your trusted advisor will help shape the outcomes of your AI strategy.

Learn more about the Dell AI Factory.

Clint Boulton is senior advisor of AI portfolio marketing at Dell Technologies.

Contributed by Dell Technologies.

 

Sign up to our Newsletter

Featuring highlights, analysis, and stories from the week directly from us to your inbox with nothing in between.
Subscribe now

Be the first to comment

Leave a Reply

Your email address will not be published.


*


This site uses Akismet to reduce spam. Learn how your comment data is processed.