As AI infrastructure outgrows tightly coupled systems, we’re witnessing a shift toward openness and modularity in designing full-stack solutions for AI. In this session, we’ll examine the rise of Software-Driven Fabrics (SDF) , a programmable, vendor-neutral control plane for modern AI networking. SDF makes real-time data coordination across heterogeneous accelerators and fabrics possible, offering a new, democratized model for GPU scalability and network resiliency.
Prashanth Thinakaran
Prashanth Thinakaran is a Distinguished AI Infrastructure Engineer at Clockwork Systems, a pioneer in nano-second precise network telemetry and software-driven resilience that addresses the unprecedented scale, performance, and reliability Modern AI workloads demand from GPU clusters. In this role, he partners with AI Infrastructure teams at enterprises, hyperscalers and neoclouds to increase their visibility into issues impacting cluster uptime and optimize their availability and utilization with Clockwork’s solution.
Previously, he helped AI-native companies build on cloud-based GPU platforms, providing deep technical guidance on distributed training and inference, multi-node scaling, and performance tuning across complex infrastructure stacks. His Neocloud experience bridged the gap between product engineering and customer enablement, helping fast-moving teams adopt best practices in massive-scale model deployment and operations. Prior to that, Prashanth played a pivotal role at Cerebras Systems - a market leader in high-speed inferencing, - in the design, and deployment of Condor Galaxy 1, a Wafer-scale supercomputer. His work enabled rapid deployment timelines and seamless scaling of AI infrastructure across globally distributed data centers designed for both Inference and Training.
Prashanth also holds a Ph.D. in Computer Science and Engineering from Penn State, where his research focused on high-performance computing and cloud systems. His academic work has been published in top-tier venues including USENIX NSDI, ACM Middleware, ICDCS, and ACM SoCC. He has authored over a dozen peer-reviewed papers and a book chapter, and served as a reviewer for journals such as IEEE TPDS and TCC. During his Ph.D., he held teaching roles in systems programming, and computer architecture, and collaborated with Intel, VMware, and Adobe Research through internships, solving systems challenges at the intersection of academia and industry.
