Astera Labs at Supercomputing 2025: Advancing Rack-Scale AI Connectivity with Open Ecosystems

Paroma Sen, VP, Corporate Marketing

As AI models surge in complexity and require unprecedented reasoning performance, the industry is undergoing a transformative shift toward what we call “AI Infrastructure 2.0”—where the rack, not the server, becomes the new unit of compute. The scale and tight coupling required by today’s GPU-dense workloads demand ultra-low-latency, high-bandwidth communication across hundreds of accelerators operating as a unified system, leading to a whole new level of requirements that the infrastructure must now satisfy.

The fastest and most direct path to this rack-scale future is through open standards and multi-vendor collaboration. When companies build on shared specifications, teeing off each other’s unique domain expertise, innovation accelerates in parallel, unlocking larger, faster, more efficient GPU clusters, and ultimately a more sustainable way of building toward a shared future of AI.

Join Astera Labs at Supercomputing 2025, November 17-20 at the America’s Center Convention Complex in St Louis, where we’ll demonstrate how this shift is happening in real time.  

  • Rack-Scale Performance. Supercharged with PCIe 6.
    • See PCIe 6 interoperability in action in PCI-SIG Booth 3118. Astera Labs will showcase a robust PCIe 6 physical and transaction layer integrated with our COSMOS software suite to deliver real-time insights and end-to-end visibility across a multi-vendor ecosystem of PCIe 6 endpoints. The demo also features our Aries Gearbox, seamlessly bridging PCIe 5 and PCIe 6 traffic to enable smooth transitions between PCIe generations.
  • Rack-Scale Memory Innovation. Unlocked with CXL.
    • See how Astera Labs’ Leo CXL Smart Memory Controller boosts AI performance by expanding memory beyond system limits — from 1.5TB per socket to an additional 2TB of available memory. For example, without CXL AI workloads can stall at 400 chatbot instances, with GPUs waiting on storage. With Leo CXL Smart Memory Controllers, throughput can be tripled to 1,200 chatbot instances, slashing latency and boosting GPU utilization. See the demo in CXL Consortium Booth 817.
  • Rack-Scale Connectivity Based on Open Standards. Scale-up with UALink.  
    • Visit the UALink Consortium Kiosk (SNIA Standards Pavilion) Booth 211 to discover how Astera Labs is supporting UALink technology. Learn how UALink is designed from the ground up for scale-up infrastructure and  supported by a multi-vendor ecosystem.

Let’s connect in St. Louis. Reach out to schedule a meeting with our connectivity experts at the show.