Dr Thomas Di Giacomo, Chief Technology and Production Officer, SUSE stresses that enterprises must stop waiting for scarce hardware and instead unlock idle compute, converge silos, optimize workload placement, embrace open source, and adopt software‑defined infrastructure to sustain AI momentum despite supply constraints.
AI hardware shortages are slowing enterprise adoption globally. How is SUSE helping organizations identify and overcome the most critical infrastructure bottlenecks without waiting for new hardware?
Honestly, the instinct when hardware is scarce is to just wait for more of it. But that’s not a strategy. Our recent SUSE Customer Reference Panel survey of 110 practitioners found that 71% of IT leaders are frustrated by AI hardware delays, and with memory prices up 95% or more, waiting on the supply chain simply isn’t realistic anymore.
What we tell customers is to look at what’s already sitting in front of them. Most organizations have far more capacity than they realize. The opportunity is unlocking it, not waiting for a bigger box to show up.
Your research highlights that enterprises often underutilize existing compute resources. What practical steps can CIOs take today to unlock hidden performance within their current infrastructure?
My first piece of advice is always the same: converge your silos. So many organizations are running separate virtual machine and container clusters that never talk to each other, and that’s exactly where all that idle capacity is hiding.
Our research shows enterprises are leaving massive amounts of CPU and RAM sitting completely idle. Bring those workloads onto one platform and you can realistically push utilization toward 80%. Then layer in smart tiering for memory-heavy workloads. It buys you real time before you’re forced into an expensive hardware upgrade.
How can Kubernetes-driven orchestration and containerization help enterprises run AI workloads more efficiently, especially when GPU availability is limited?
It comes down to simple maths, really. A GPU sitting idle in one silo does nothing for a workload that’s starved for compute in another silo. Kubernetes gives you one shared pool instead of a handful of fenced-off ones. Once you converge onto a single platform, that 80% utilization target stops being aspirational and starts being achievable, without buying a single new piece of hardware.
With AI moving closer to the edge, how can distributed architectures reduce dependency on centralized high-end hardware and improve overall workload efficiency?
Edge takes real pressure off centralized, high-end hardware that’s particularly hard to get hold of right now. But I don’t want people to think of it purely as a workaround. It also opens up use cases that need to run close to where the data actually lives, independent of connectivity. That’s why I see edge as a permanent part of how enterprises will build going forward.
SUSE emphasizes open source as a strategic enabler. How does open source help enterprises navigate hardware shortages while maintaining flexibility, scalability and cost control?
When hardware is this hard to come by, the last thing you want is software that only plays nicely with one vendor’s stack. We’re seeing real appetite from enterprise leaders for portable, vendor-neutral software rather than locking themselves into one hardware vendor’s roadmap, and that’s precisely what open source protects. It means your plans aren’t being held hostage to someone else’s shipping schedule or pricing decisions.
What role does intelligent workload placement play in ensuring AI models run efficiently across mixed environments, on-prem, cloud, and edge, especially when hardware is constrained?
Smart memory tiering is a good example of what this looks like in practice. Using high-performance storage as a secondary memory tier lets you delay a costly DRAM upgrade while still running efficiently. But placement is bigger than storage. It’s about making sure every workload runs where it actually performs best, whether that’s on-prem, in the cloud, or at the edge, on the right chipsets, too. Get that right and it keeps paying off long after the hardware market eases.
In a period of hardware scarcity, how can enterprises maintain resilience, ensure business continuity, and avoid stalled AI initiatives?
Being candid, hardware delays are a major reason I see AI roadmaps stall. Scarcity and long procurement lead times are stalling regular IT projects across the board. My advice is to flip the delay into a forcing function rather than treating it as a blocker. Start with an honest utilization audit of what you already have. Start with solutions on neocloud that can then be ported across your environment. The enterprises that do that are the ones still moving forward, no matter what’s happening upstream in the supply chain.
Looking ahead, how does SUSE see AI infrastructure evolving over the next couple of years, and what should enterprises prioritize now to stay competitive despite ongoing hardware shortages?
The previous “just buy more hardware” approach to scaling is becoming financially unsustainable, and I don’t think it will reverse even once supply improves. Constraint doesn’t have to mean giving up architectural agility, that’s a mindset shift as much as a technical one. Software-defined infrastructure is how enterprises get more from what they already own. The organizations that build that discipline now are the ones that will still be competitive two years from now, whatever the hardware market looks like.
