Interview: AI Infrastructure and Independent Data Management Are On Trend

Steve-McDowellSteve McDowell is Principal Analyst and Founding Partner at NAND Research and a contributor to Forbes. He has deep industry experience in engineering, strategic marketing, and strategy roles to help top-tier technology providers deliver the right products and solutions for the next-generation datacenter. He is an expert in IT architectures and infrastructure, storage, data management, hybrid multi-cloud, HCI/CI, edge computing, and AI/ML/DL. He shares his views on the market below, with AI infrastructure solutions top of mind.

What were the most influential stories in your world in 2023?

SM: Well, you have to say generative AI. But I don’t care as much about all the cool applications for generative AI, because I cover infrastructure and this very much changes the way we think about the infrastructure that supports data, right? And AI, I think it has surprised people– including the OEMs– how cloud centric it is already. If you look at the earnings for Dell, HPE, Lenovo, and other big tech vendors, they’re like, we didn’t see the tailwinds we expected from generative AI.

But then you look at Nvidia’s earnings and the cloud guys and that’s where all the money is; AI has turned into a cloud-first play. The year 2023 was the year we pretty much stopped saying digital transformation and started saying data transformation–and that’s due to AI. I think 2023 is also a big year for cyber security, coupled with governance and compliance. All these trends have the same set of implications on your data, which is: I need to know what my data is, where it’s at, how it’s being used, and who’s talking to it.

“AI has turned into a cloud-first play.”

How do you see things shaping up for enterprise tech in 2024? Are you bullish or bearish?

SM: I am bullish: layoffs are a fact of life in tech. I don’t see that as a fundamental problem in the industry; there has been some rebalancing. But also when tech companies were planning for 2023, which usually starts in Q3, the plans didn’t factor in the impact of GenAI. That resulted in a lot of scrambling and reallocating of resources. But it is starting to stabilize a bit now.

We’re moving from phase one of generative AI, which is let’s train it, let’s figure out what these LLC’s can do, to we’re going to enable all these applications with generative AI and that requires a lot of horsepower to put it to work, to do the inference. And again it all comes down to the data.

For generative AI to be useful in an enterprise, I can’t take an off the shelf ChatGPT. I need to supplement the model with my data, which means I need to understand my data.

What are the implications for storage vendors?

SM: The demands on the whole storage infrastructure continue to change and I don’t think we fully understand yet what that means. It’s all unstructured data that’s interesting to AI and analytics. I can’t remember the last time a vendor said I’m selling a storage array. Now everyone is selling “data infrastructure platforms.”

How is the edge and AI market playing out?

SM: The 5G guys promised that 5G is going to explode the edge. We didn’t see that happen, but man moving AI to the edge will. I was at a national retail show a couple of weeks ago. They’re using AI to do things like inventory control and loss prevention. Everybody hates self-checkout, right? Now I can have a virtual checkout person there to help me. Qualcomm showed at CES their car of the future with a generative AI interface that helps with things like, how do I change my tire or check my oil. You can ask your car and it will give you step to step directions. That’s going to be on the market in 18 months.

In the storage industry, we hear lots of talk about all flash environments, StaaS, and AI-ready storage. Is this where customers are right now?

SM: I don’t think I’ve seen a single storage vendor who said their storage is not AI ready. What we learned in 2023 is the number of companies doing training at scale is very small. AI as a service is going to solve the training problem. What companies need is storage that keeps up with my analytics needs, and that’s flash. The storage vendors are figuring that out. Flash price for capacity is very competitive with nearline hard drives. Sustainability is also a factor. Flash has a much smaller footprint. If it’s not reading and writing. I am not drawing on power. Where flash is still failing is in archival storage. If I’m doing big volumes of low performance storage, the hard drive guys are still gonna win. If you go to an Amazon data center, you’ll see a lot of spinning drives.

Storage as a service is also strong; it’s already a healthy percentage of revenues for the vendors now. This comes out of what cloud did, which was reset expectations of how we buy and consume infrastructure. Finance guys love “as a service” and especially when it’s a managed service.

Moving to cloud, do you think AI is going to fuel a resurgence of enterprise cloud spending and cloud data migrations in 2024?

SM: AI is a cloud-first story because AI infrastructure Is expensive and complex. NVIDIA is maximizing its position. If I want to train a large language model, that’s a $30,000 card. The accelerators are expensive and they don’t have the life of a server. You may need to swap them out in eight months. If you look at recent cloud earnings, AI is changing the dynamics. Amazon was a little late to the party with AI infrastructure for a variety of reasons. Looking at the growth last quarter, Amazon’s is the lowest of any cloud provider. AI is helping cloud, but people are going to where there’s availability and there’s still scarcity of the GPUs.

What else do you hear from customers or users on their infrastructure priorities this year? Security is surely a top one.

SM: Enterprises are not buying and building infrastructure for AI. But they do care about security. That market is booming right now, and it’s such a fragmented space in terms of the vendors who service it so it’s confusing and hard to navigate. Observability is another big one. Five years ago, observability meant reading log files but observability now is a core capability for IT. It’s Black Friday and I have to rebalance all my workloads to meet the escalating demand. And again, it comes back to what is the data that supports that!

How does unstructured data management fit into the above trends that we’ve been discussing?

SM: I think it’s central to everything. AI operates almost exclusively on unstructured data. Doing analysis on data in silos and understanding the data you have is business critical. The “data is the new oil” phrase that came out years ago is finally true.

But building intelligence into my unstructured data management doesn’t just feed all these business processes that are data heavy and data dependent right now; it also plays into cost optimization, certainly in a cloud.

There is a very real economic impact to how I manage this data. There’s also a security component. How many data breaches are there going to be which are driven by somebody’s exposed S3 key? Everything I’m doing in an enterprise now is governed by what’s living in my unstructured data.

Do CIOs and IT executives see the need to manage data differently—independently of storage vendors– and be more efficient with storage spending?

SM: Storage infrastructure moves more slowly than everything in your data center, but to do all these things we are talking about, you have to manage data independently of the media. One aspect is getting a grasp of my data spread across at least three if not five vendors. That gets in the way of the unified view. I need a data infrastructure independent of wherever the bits live. It also gives me a common set of controls. I’m a fan of object storage because it gives more flexibility than a traditional NAS.

IT buyers want choice and independent data management improves the negotiating power of the IT buyer. The IT buyers want this both for reducing complexity and cost control. If you can cross cloud boundaries with an unstructured data management solution, it opens up all this flexibility where I can get the promise of cloud native and I can deploy new workloads wherever I have resources. Storage vendors are pushing back against it, but I think unified data management is the future.

Getting Started with Komprise:

Contact | Data Assessment