SC26 Puts AI, HPC and Quantum Integration in Focus
- Cargo
- Machine Learning
- Shipped
- Shipper
- Mark Albertson
- Readers
- 703
SC26 highlights how AI, accelerated computing and quantum systems are reshaping supercomputing, with data movement and system design central to results.

For high-performance computing, the next performance gains may depend as much on moving data and coordinating systems as on adding faster processors. That’s the thread running through SC26, where AI infrastructure, exascale computing and research workloads meet. The event is scheduled for Nov. 17–19 in Chicago, with coverage planned around how these technologies are changing supercomputing and enterprise infrastructure.
The shift has practical consequences. Teams deploying AI and scientific workloads need compute, storage, networking and application tools to work together. A powerful accelerator alone can’t deliver much if processors spend time waiting for data, or if results remain stuck in experiments instead of reaching production. The SC26 discussion is expected to examine those constraints alongside emerging approaches such as quantum-HPC integration and agentic AI in research.
Quantum systems join the HPC stack
Dell Technologies’ work with Nvidia and other organizations illustrates how vendors are combining established HPC infrastructure with newer computing approaches. One project is the Doudna supercomputer at Lawrence Berkeley National Laboratory. The planned system will use Nvidia’s Vera Rubin CPU-GPU platform and Quantum-X800 InfiniBand networking in Dell integrated rack systems and PowerEdge servers. Its target workloads include large-scale scientific computing and research that depends on substantial data processing.
The pairing matters because building a research system involves more than choosing a processor. Compute, interconnects and storage must support workloads that move large volumes of data while keeping the system’s processing units busy. The Doudna design brings those pieces together in a single infrastructure plan, though the source provides no performance results from the completed system.
Dell has also teamed with Diraq, a developer of silicon-spin qubits, on integrating quantum processing units with HPC and AI server infrastructure. The collaboration aims to connect quantum processors to classical systems in commercial data centers. That points toward a hybrid model in which quantum hardware works alongside conventional compute rather than replacing it.
How those systems will be used at scale remains an open question. TheCUBE Research analyst Dave Vellante argues that quantum hardware and accelerated computing may complement one another, particularly for error correction. The practical test will be whether combining them makes useful workloads more feasible, and how much supporting infrastructure those workloads require.
Cloud HPC targets industry workloads
Google Cloud is pursuing HPC services for manufacturing, health care and life sciences, weather forecasting, and electronic design automation. Its H4D virtual machine, introduced last year, is built on fifth-generation AMD EPYC processors. Google reports whole-node performance above 12,000 GFLOPS and memory bandwidth exceeding 950 GB/s. Those figures position H4D as a cloud option for workloads that need substantial processor throughput and fast access to memory.
A reported use case is the Weather and Climate Operational Supercomputing System. Google Cloud said in July that it had been selected as the system’s primary provider of HPC infrastructure. The work shows how cloud-based compute can support forecasting research and the development of future models. The source gives no additional details on deployment scale, costs or measured forecast improvements, so the operational impact remains difficult to assess from the available figures.
HPC’s cloud role isn’t limited to large providers. Vultr is building infrastructure across 33 global regions, using CPUs, GPUs and open, composable architectures. Its stated focus includes AI, risk management, compliance and fraud prevention, with an emphasis on agentic AI workloads. The source describes its direction but offers no benchmark data or cost comparisons for its platform.
Data access and system design shape results
More compute can create a new bottleneck: getting consistent access to data across platforms. Omni Analytics is developing a data collaboration product that connects directly to cloud data warehouses without first loading or duplicating the data. That approach could simplify access for teams working across systems, although the source doesn’t report performance or deployment results.
The underlying issue cuts across these examples. As John Furrier of theCUBE Research puts it, different architectures face the same challenge: keeping processing engines occupied instead of waiting for data. That makes the full system design—data movement, storage, networking and application deployment—a meaningful factor in performance. For organizations comparing infrastructure, chip specifications alone won’t answer whether a system will deliver useful results on their workloads.
Paul Nashawaty, also of theCUBE Research, frames the business challenge as connecting high-performance infrastructure and accessible data with modern application development. His view is that organizations need to turn infrastructure investment into measurable scientific or business outcomes. That means evaluating the full path from data access to deployment, not simply counting compute resources.
SC26’s central technical question is therefore how well the pieces fit together. The event’s planned discussions span accelerated research systems, quantum-classical integration and agentic AI. For teams assessing those approaches, the useful signals will be concrete: which workloads run better, what data and infrastructure they require, and whether integrated systems can translate peak capacity into sustained results.