This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Read our privacy policy
Andrew Buss, Sr. Research Director, IDC EMEA
This piece was excerpted from remarks delivered at the Mobile World Congress Barcelona 2025.
Our research shows that the most digitally transformed companies view IT as a means of differentiation and competitive advantage. They're investing heavily in IT infrastructure, and are increasingly profitable, competitive, and growing.
These digital leaders are embracing AI. Companies that move beyond the “proof of concept” phase into production at scale have been transformed: 30% of the Digital Leaders have moved a GenAI workload into production. The digital mainstream is only at 5% – that’s the middle 50% of the market – and Digital Followers are only at 2%. So, there is a part of the market that is ready and able to take advantage of GenAI today.
One of the keys in enabling successful AI workload deployment is the availability of real-time data, whatever the type of data, wherever it's located. We can't store absolutely everything, but we need to store most of it.
We find that over half of companies prefer to deploy their workloads onto their private IT. But when it comes to GenAI, they have a slight preference for the public cloud, often because that’s where the actual infrastructure is today. It’s been hard to buy and deploy AI systems because of the huge demand that we see. So, companies have often stood up their initial proof of concept in the public cloud. But increasingly, as we go forwards, we see companies looking to move their AI from the public cloud into private AI when they decide to move at scale.
Our key advice is to really think about where you want your main workloads to be running. If your data is in the public cloud, you will have to consider whether that’s the best place to run those workloads. If you have compliance and regulation needs that say, “Actually, you can't do that,” then you'll have to consider how to manage your data, so you run the AI workloads where you need to.
AI is memory- and data-intensive. It will depend on incredibly high-performance networks and storage to drive those GPUs and AI accelerators to perform at their best.
Today, companies want to keep running their workloads in that private IT mainly because of regulatory compliance and sovereignty concerns. But most companies’ internal IT infrastructure is siloed, and it’s not manageable at scale. It’s lots of independent systems, each with their own automation paradigms. If we try to pull them all together, it's more a “single glass of pain than a single pane of glass.”
So, going forward, we need to make sure our internal private IT works well together and can be automated at scale: advanced automation and orchestration features are absolutely critical.
AI is so intensive for power and compute that, if you don't have control of the infrastructure yourself, your costs can ramp up quickly. This is another reason why companies are keen to move workloads to private IT. We find AI workloads tend to use the hardware at 100% capacity a lot of the time. It drives the hardware incredibly hard, and that can rapidly increase costs in the public cloud. Bringing workloads to internal data centers is key for keeping costs under control.
We also find that a lot of public cloud providers do not have AI services in the region that's required. And, again, regulatory compliance and data sovereignty concerns come through. So, AI is a new workload, so companies are still getting to grips with how to manage it and move forwards with it.
When taking internal private IT and turning it into a cloud-native infrastructure, we need data storage that can move with the times. We need the data itself to be accessible and high-performance, and, ideally, largely independent of the storage technologies underneath. Accessibility of data will be key. For that, we need a unified platform that's integrated, manageable, and ideally, has a lot of intelligent operations built into it.
Resilience is absolutely key, because any downtime results in millions of dollars of brand damage and lost sales and revenue. Even more important is making sure energy consumption is under control. We are seeing energy consumption go up by a factor of 10 with an AI data center compared to traditional enterprise workloads. So, rather than 15 kilowatts per rack, we're seeing 150s kilowatt per rack.
How do we get to where we want to be? Although traditional forms of storage (such as spinning disks) remain relevant, the growth is either in hybrid arrays, where we have an increasing element of flash, or all-flash arrays. With the engineering that’s going into it, flash is decreasing in price and increasing in performance, at such a rate that we see this as the future of data center storage.
Companies should plan for a data center built on Flash: Flash for mission-critical applications; Flash for business-critical applications; Flash for data analytics. All of these need to perform at the highest levels to be agile and responsive, but particularly to be able to respond at the rate that AI can give us advice or generate autonomous, agentic decision-making going forwards. If we're not automated, if we're not orchestrated, this will be almost impossible to achieve.
We also need all flash for our Large Language Models. These are getting to be in the multi-Terabyte, heading towards Exabytes stage. We are going to be flooded with models that have billions, or even trillions of parameters. The large super-clusters we see today have a memory space of many hundreds of Terabytes. In a world that needs resilience, we need All-Flash for backup.
For more on this topic, please see IDC’s new white paper, sponsored by Huawei, entitled Elevating All-Flash Datacenters to Accelerate Intelligence-Digitalization.
