SC26 Preview: What Should Enterprise Buyers Watch in Chicago?

SC26 runs November 15-20, 2026 at McCormick Place in Chicago, and enterprise buyers should watch four things: cooling and power hardware, the November TOP500 list, AI rack availability claims, and storage and networking for GPU clusters. This is a preview written on November 6, before the show opens. It covers only what organizers and exhibitors have already published, and it contains no results from the show. Any product or ranking news from the week itself is not reflected here.

What is on the SC26 schedule?

SC26 is the International Conference for High Performance Computing, Networking, Storage, and Analysis. The published program gives these fixed points, and it is worth confirming times on the SC26 site before you travel.

• Monday, November 16, 5:30-6:45 pm: the HPC Unites Plenary, an international panel on how people can harness AI across science.
• Tuesday, November 17: the keynote by Anima Anandkumar of Caltech, titled "The Next Scaling Law after LLMs: AI that Understands the Physical World."
• Tuesday through Thursday: Invited Talks, with the Awards Ceremony on Thursday, according to the program page.
• Exhibit floor: the SC26 exhibits page lists the Grand Opening Gala Reception on Monday, November 16, 7-9 pm, then exhibits 10 am-6 pm on November 17 and 18 and 10 am-3 pm on November 19.

What will the November TOP500 list show?

We do not know yet, and this preview makes no prediction. TOP500's submission rules set an October 24 deadline for new sites in the November release, with systems installed by November 1. The current June 2026 list, the 67th edition, has China's LineShine at No. 1 with 2.198 exaflops on HPL, ahead of El Capitan at 1.809, Frontier at 1.353, Aurora at 1.012 and JUPITER Booster at 1.000. It was the first China-based system to lead the list since 2017.

For an enterprise buyer, the ranking itself matters less than the efficiency figures and the system designs behind them. TOP500 ranks systems on the Linpack benchmark, which is a scientific computing measure and does not tell you how a server will handle your inference workload. El Capitan's 60.94 gigaflops per watt, as reported in June, is a useful reminder that performance per watt now shapes system design. Use your own benchmark and your own power limit when you compare hardware.

Which cooling and power products are being previewed?

Physical infrastructure is a clear theme in pre-show announcements. Panduit said on November 5 that booth 4000 will demonstrate direct-to-chip cooling, high-density fiber connectivity, power distribution and infrastructure management. Gamma Technologies lists booth 6300 with a focus on data center cooling and power simulation. For most enterprises, this is the most practical part of the floor, because colocation and on-premises limits now shape what you can deploy. Our colocation shortage guide explains why, and our liquid cooling comparison will help you ask sharper questions.

• What facility water temperature and flow does this cooling system need?
• Can it be retrofitted to an existing air-cooled room, or does it need new plumbing?
• Who services it, and does the server warranty stay intact?
• What maximum power per rack has been tested, not just designed?
• Which server models are validated with it today?

What about networking and storage for GPU clusters?

The GPU often gets the headlines, but cabling, storage and scheduling decide whether it stays busy. CommScope lists booth 4654 with Rapid Fiber Connect, which it describes as a plug-and-play platform for dense GPU racks, pods and clusters. Hammerspace is at booth 4238 with its data platform. NVM Express will highlight NVMe drives at booth 1020 in the Open Standards Pavilion. Parallel Works is at booth 5405, offering meetings on running Slurm and Kubernetes side by side, bursting from on-premises clusters to cloud GPUs, and governing AI and HPC spend. Those are vendor descriptions, so ask for customer references and test results.

Will NVIDIA or AMD news at SC26 change what I should buy?

As of November 6 we found no SC26 press release from NVIDIA or AMD, so anything said here about their show plans would be a guess. Two dated facts are still useful. NVIDIA will discuss its fiscal third-quarter results on a call on Tuesday, November 17, at 2 pm PT, the same day as the SC26 keynote. Listen for what it says about supply and Vera Rubin timing. AMD said in its third-quarter release that Anthropic, Meta, Microsoft, OpenAI and Oracle Cloud Infrastructure are running models and workloads on the first Helios rack-scale systems, and that it expects to scale Helios and Venice production in the fourth quarter.

Watch the status words. GamesBeat reported in May that Vera Rubin was ramping into full production, and NVIDIA said in August that production shipments had begun that month (call transcript). "Ramping" and "shipping" do not mean an enterprise can order one for delivery next month. HPE's March release said its Vera Rubin NVL72 system would be available in December 2026, so check whether that date still holds. Our guides to Vera Rubin timing and HBM4 supply cover the wider picture.

Should I wait for SC26 before buying AI hardware?

If a purchase is funded and needed, no. The first named users of new rack-scale platforms are large AI and cloud companies, and most enterprise workloads do not need them. Holding a quote for a conference can also cost you, because quote windows have shortened and memory prices are the bigger variable, as we explain in why server prices went up. Waiting makes sense if your project is still unfunded, if your facility cannot yet power the system, or if you are specifically evaluating cooling for a build next year. In those cases, use the show to gather facts rather than to time a purchase.

What should buyers ask vendors at SC26?

• Is this shipping now, sampling, or only announced? What is the first customer ship date for an enterprise?
• What is the maximum power per system and per rack, and what cooling does it require?
• Which configurations are validated, and which are only planned?
• How does the warranty work, and who provides support in my region?
• What is the memory configuration in the benchmark, and can I buy that configuration at a realistic price?
• How does CPU-to-GPU balance change for my inference workload?

How will we know what actually happened at SC26?

Treat the show week as a source of claims and the following weeks as the test. Check the TOP500 site for the November list, read vendor data sheets rather than slides, and ask for written availability. Independent coverage from outlets such as ServeTheHome and The Register can help separate shipping products from roadmap items. Then compare anything that interests you against the hardware you can order today.

Nexus Compute can give you that comparison. We supply new GPU servers, workstations, storage and networking through authorized distribution channels with the full manufacturer warranty, and we check compatibility against your platform before dispatch. Quotes come within 48 business hours, with purchase orders and net terms for qualified organizations. Request a quote or email sales@nexus-compute.com with what you saw and what you need to run.