Leaseweb: AI Creates a Data Management Problem, Not Just a Compute Problem, Podcast

By Doug Green

“Your opportunity to demonstrate value is to show that you understand what happens to the data.”

In this Technology Reseller News podcast, Eli Lahr, Senior Solutions Engineer at Leaseweb, expands on a recent TR article examining one of the less discussed challenges of AI infrastructure: the data created around the workload itself.

Leaseweb is an infrastructure-as-a-service provider offering bare metal, cloud hosting, object storage and network services. Lahr says much of the AI infrastructure conversation has centered on GPUs, but MSPs should be paying equally close attention to data.

AI workloads do not simply consume training data. They also generate logs, artifacts, backups, metadata and other information that continues to grow as projects scale.

That can quickly expose infrastructure bottlenecks.

“The minimum amount of data for some training sets is now measured in petabytes,” Lahr says. “The question becomes where all of that data goes and how quickly you can get it back.”

For MSPs, the answer is not simply more storage.

Different data has different requirements. Some information needs extremely low latency and fast access, while logs or backups may be better suited to less expensive storage tiers.

The challenge is putting the right data in the right place, at the right speed and at the right cost.

Lahr also points to “data gravity.” As datasets grow, moving them between providers or locations becomes increasingly expensive and time-consuming. That makes decisions about where data is stored—and how close it is to compute—an important part of AI infrastructure planning.

For MSPs, that creates an opportunity to move beyond simply supplying infrastructure.