AI Sweden: Driving AI Innovation with OpenNebula

BACKGROUND: THE NEED FOR A SMARTER CLOUD SOLUTION

AI Sweden, the national center for applied artificial intelligence in Sweden, supports AI research and innovation through collaborations with various industries, including healthcare and automotive sectors. Given its mission to foster AI adoption, AI Sweden required a cloud infrastructure that was both flexible and efficient, enabling seamless collaboration among partners without unnecessary constraints.

Before adopting OpenNebula, AI Sweden faced challenges with proprietary cloud solutions. “Some competitors had restrictive terms of service. They required us to send five engineers for training and two salespeople, which didn’t make sense for our needs – we’re not selling anything,” recalls Ted Henriksson, Systems Engineer at AI Sweden. “With only two people running our lab, such requirements were impractical.” The team needed an alternative that would support GPU passthrough while allowing them to maintain control over their infrastructure.

WHY OPENNEBULA? SEAMLESS MULTI-TENANT FLEXIBILITY

AI Sweden explored several solutions but found OpenNebula to be the best fit. “We needed a platform that could handle GPU passthrough effectively,” says Henriksson. “Managing KVM through OpenNebula gave us the flexibility we required without the rigid constraints of larger proprietary systems. It was simple, practical and aligned with our needs.”

Another key advantage was OpenNebula’s openness to customization. “We liked that we could insert custom KVM configurations if necessary,” Henriksson explains. “Even though we don’t always need to, having that freedom is invaluable. We also had strong storage partners who helped us fine-tune the setup, making the transition smooth.”

IMPLEMENTATION: SCALING WITH SIMPLICITY

AI Sweden initially deployed OpenNebula in 2019 and has since scaled its infrastructure. “We have a large deployment within the system, with approximately half of the hosts equipped with GPUs,” Henriksson notes. “The primary use case is AI research, so the GPU-enabled hosts are critical. The other half are included simply because OpenNebula makes it easy to manage VMs.”

The deployment process evolved over time. “Our second production run involved automating deployments with Ansible, which took about a month to refine. Now, provisioning new hosts is quick,” says Henriksson. “Once set up, it just works. There are always minor tweaks, but nothing significant enough to disrupt our workflow.”

RESULTS: A RELIABLE, NO-HEADACHE INFRASTRUCTURE

OpenNebula has streamlined day-to-day operations at AI Sweden, making cloud management more predictable and efficient.

“Honestly, it’s been refreshingly uneventful,” Henriksson remarks. “We used a proprietary provider before, but OpenNebula has been far more aligned with our needs. Automation is working well, and I appreciate that everything remains stable and steady. That’s exactly what we want in infrastructure.”