With the exponential rise of AI causing so much uncertainty for data center provision, how can you be confident that your facility is futureproof? That’s the question DCD’s Zoe Turner asked Wellington Lordelo, global director of AI and Innovation at Digital Realty, in a recent DCD>Talks interview.

Lordelo tells us that we can learn a lot about the future from looking to the past:

“We have seen the same trends that we saw with cloud service providers and the use of clouds within enterprise repeat with AI. What we’ve learnt with the hyperscalers can be a blueprint for anyone to copy and paste.”

Digital Realty recently produced an eBook titled ‘Rewire for data and AI’, a collection of studies from over the past five years with 2,000 global IT leaders from large enterprises and service providers. Lordelo uses this as his starting point:

“We have the privilege of sitting in the front row seat to understand what the implications are right now. With AI, it’s hard to keep up with what is signal and what is noise. The eBook acts as a blueprint that every enterprise or service provider can look at to understand the next steps, without the noise.”

As Lordello indicates, the results can offer an insight for everyone in the industry, with commonality across enterprise and service providers, including legacy, specialist hardware such as high-performance GPUs and liquid cooling, as well as data gravity, pulling towards where data is located:

“Data around the world is costly and sometimes it's inefficient as well,” he explains, “The upshot is that running AI in generic cloud environments becomes really prohibitive, because it’s really expensive at scale.”

But the list doesn’t end there, as Lordelo goes on to point to a lack of expertise in AI in the workforce, and the thorny issues of governance and compliance, with legacy architectures falling prey to the regulatory demands of where and how data must be processed.

“Legacy isn’t just a bottleneck; it’s a silent blocker of AI success. Without addressing these issues, we’ll end up spending more and more and achieving less and less.”

The conversation moves to the question of optimizing the location of data for AI, and looking at trends that it throws up. Lordelo points to the sheer speed of transformation:

“AI adoption is outpacing forecasts. We are seeing years’ worth of AI innovation each quarter. Three forces are driving this – machine learning, generative AI, and agentic AI.”

‘Agentic’ refers to the concept of giving AI agency – that is to say, allowing it to make decisions. Lorelo continues, explaining that “AI first” strategies have become non-negotiable:

“84 percent of the leaders in our study now link data locations to AI road maps, and because enterprises are shifting from just cloud to a hybrid, colocation distributed model, it becomes even more complicated, with questions such as where to deploy systems, who should connect with whom, and the killer quandary: Where’s my data?”

With data being generated at the Edge, also residing in data lakes, and employees and end users consuming data in different locations around the world – having a data location strategy is key. But what effect does data location have on efficiency, cost, and performance?

“Latency is all about physics. Keeping compute close to the data reduces latency and improves inference time. And the other thing is cost – moving petabytes of training data to and from the public cloud is expensive. Colocation helps reduce this transport and egress charge, and improves performance. When we move from training to inference, that latency is even more critical. It’s not just between point A and point B – it’s across the whole alphabet – all the hops within the same network.”

So what does Digital Realty suggest in terms of optimizing that AI infrastructure? What, in Lorelo’s words, does a ‘well-designed infrastructure’ look like?

He suggests that the key points are capacity, to avoid over-provisioning and overspending, and latency, to ensure the compute is close to the data. He reminds us that Digital Realty can address such things in a workshop format with customers:

“You need a reliable colocation partner who is a global player. Compliance is super-important too. Some countries have their own regulations, so you need to understand the laws of the locations you’re using and serving.”

Digital Realty can offer services to make up the much-discussed shortfall in AI expertise within organizations, thanks to a global network of over 200 service solution architects and sales engineers, able to speak in the ‘local language’ of the customer. Lorelo explains:

“They can guide the enterprise through AI infrastructure design, along with our ecosystem of partners, value-added resellers from OEMs. We are the only neutral platform focused on colocation and interconnection. We help the enterprise to see not just from the infrastructure layer, but to other layers too.”

Another way the company can help is with the recent launch of its first Innovation Lab, or DRIL (Digital Realty Innovation Lab), in Ashburn, Virginia, offering the latest cutting-edge solutions within a sandbox environment where enterprises can test and validate IT workloads before scaling and signing contracts. The success of this facility has led to a roadmap to open more in the future. Lorelo enthuses:

“They can experiment with liquid cooling, they can achieve 150kw workloads. They can test high-density GPU racks across different partners and refine their data strategies.”

For more insight on successful colocation deployment in the world of AI, details of the Ashburn DRIL, the plans for new global lab locations, and why the risk factors are like going to the ice cream parlour, watch the full DCD>Talk here.