Day 2: AI Infrastructure for Training and Inference
April 21, 2026
Location: Computing and Data Science Building, Simonyi Conference Center
AI infrastructure is evolving into a tightly integrated computing substrate spanning GPU clusters, accelerators,memory systems, and distributed software for training and inference. Achieving high Model FLOP Utilization (MFU) and reliability requires co-design across hardware and software layers, along with innovations in interconnection fabrics via software-driven approaches.
This conference explores advances in ML accelerators, compilers, data representations, low-latency inference systems, and large-scale AI platforms. A central theme is reconciling performance with robustness in heterogeneous, failure-prone environments, while leveraging workload and fleet automation to sustain efficiency at scale.
| Time | Agenda | Videos |
|---|---|---|
| 8:00am | Breakfast & Registration | |
| 8:50am | Welcome and Opening Remarks Balaji Prabhakar | VMware Founders Professor of Computer Science, Stanford University | YouTube |
| 9:00am | Keynote 1: The Evolution of ML Accelerators from General Purpose to Task-optimized Nafea Bshara | Vice President and Distinguished Engineer, Amazon | YouTube |
| 9:40am | Voyager: A Compiler and Design-Space Exploration System for AI Accelerators Priyanka Raina | Associate Professor of Electrical Engineering, Stanford University | YouTube |
| 10:10am | Heterogeneous Data and Memory Representations for Efficient AI Thierry Tambe | Assistant Professor of Electrical Engineering, Stanford University | YouTube |
| 10:40am | Break | Jay Borenstein | YouTube |
| 11:00am | SYMI: Efficient Mixture-of-Experts Training via Model and Optimizer State Decoupling Athinagoras Skiadopoulos | Research Scientist, NVIDIA Research | YouTube |
| 11:30am | Michelangelo: Uber’s AI/ML Platform Viv Keswani | Senior Director of Engineering, Uber | YouTube |
| 12:00pm | Lunch Break | |
| 1:00pm | Keynote 2: Connecting GPUs Worldwide into an AI Platform for the World Deepak Bansal | General Manager and Corporate Vice President, Microsoft Azure | YouTube |
| 1:40pm | Are AI Fabrics and Infrastructure Really That Different? Joseph L. White | ISG-CTO Fellow, Dell | YouTube |
| 2:10pm | Déjà Vu: Reconciling Fabric Perfection with Network Reality Murai Sridharan | Senior Vice President or Networking, Oracle Cloud | YouTube |
| 2:40pm | AI Building AI: How AI is Accelerating Model Experimentation and Enabling The Flywheel Animesh Singh | Senior Director, AI Platform and Infrastructure, LinkedIn | YouTube |
| 3:10pm | Break | |
| 3:30pm | Software-Driven Fabrics Using Clocks and Shims Balaji Prabhakar | VMware Founders Professor of Computer Science, Stanford University | YouTube |
| 4:10pm | Fireside Panel: From Packets to Parameters The panel explores how decades of networking and distributed systems innovation underpin modern AI infrastructure, examining evolving system abstractions, scalability challenges, and recurring design patterns. Discussion highlights interconnect design, job scheduling, low-latency response generation and fault tolerance in AI systems. Albert Greenberg | Chief Architect Officer, Uber Sachin Katti | Head of Compute Infrastructure, OpenAI Ion Stoica | Professor of EECS, UC Berkeley | YouTube |
| 5:15pm | Close |