SC25 marked an exciting convergence of supercomputing and artificial intelligence, and Lablup was right at the heart of it. Throughout the supercomputing week, global leaders in HPC and AI gathered to exchange ideas, explore breakthroughs, and shape the future of computing. For Lablup, it was an incredible opportunity to share our vision for unified HPC / AI infrastructure operating system and to connect with the global HPC community driving this transformation at scale.

See our highlights.

Click the image above to launch YouTube.

Lablup at SC25, in numbers

• 72㎡ booth: This year, Lablup went big. We expanded our booth to showcase our global-scale capabilities more effectively.
• 6+ use cases: Seeing is believing. We demonstrated over six real-world use cases, helping SC25 attendees understand Backend.AI's value and customer benefits.
• 4+ integration demos: Backend.AI in action across NVIDIA GPUs, Intel Gaudi 3 AI accelerators, and even consumer devices like Intel ARC GPUs and NVIDIA DGX Spark, from desktop to enterprise scale.
• 12 theater sessions: Featuring Lablup, HyperAccel, WEKA, VAST Data, and Vultr, drawing crowds interested in HPC workload management solutions, AI accelerators, high-performance storage, and cloud GPU services.

Spreading the power of orchestration

Supercomputing clusters and AI clusters have fundamentally different characteristics. Supercomputing prioritizes massive scale with minimal data movement and limited network usage, while AI demands intensive GPU-CPU networking and GPU-to-GPU interconnects to feed tremendous amounts of data for model training.

These workloads present orthogonal needs for vertical integration and horizontal scaling. Various vendors are addressing this challenge by offering solutions to harmonize these opposing factors. The key task today is integrating these divergent elements effectively, and this is where Backend.AI excels through the last decade, with the power of orchestration.

... and the power of integration

IT infrastructure is never built from a single component. To run HPC and AI workloads effectively, CPU and GPU nodes, high‑performance storage systems, carefully designed interconnect networks, and robust cooling for dense systems all need to work in harmony as one environment. On top of this, software plays an even more critical role. Only when there is software that can combine all these elements efficiently, with the right configuration and maximum performance, does the underlying hardware infrastructure truly come together.

At SC25, we highlighted how Backend.AI supports a wide range of datacenter‑class AI accelerators and consumer CPUs from more than 11 vendors, including NVIDIA and Intel, while also integrating with high‑performance storage systems from providers such as Pure Storage, VAST Data, WEKA, and Dell Technologies. Through Backend.AI, customers can unlock the real benefits of end‑to‑end integration across their heterogeneous infrastructure.

The winner!

This year, we hosted an exciting new challenge at our booth where visitors optimized hyperparameters on NVIDIA DGX Spark to achieve the fastest training speed (TPS). The Day 1 winner took home an Intel Arc B580 GPU, while the Day 2 champion received the NVIDIA Jetson Orin Nano Super Developer Kit. Congratulations to our winners!

Looking to the Future Together

Thank you to everyone who visited our booth. Let's keep building scalable, integrated computing solutions together in 2026 and beyond.