This presentation details how Meta is addressing the massive compute and data demands of modern generative AI workloads by co-designing hardware, software, and models. The talk highlights the communications architecture of MTIA 300, Meta's custom-designed AI training and inference accelerator featuring dedicated message engines to efficiently offload communication workloads. By decoupling communication from the primary compute grid, the system minimizes resource contention and substantially drives up both compute and network efficiency. Additionally, the speakers explore how "network fungibility" allows the same network interface cards (NICs) to flexibly adapt to both scale-up and scale-out configurations depending on production needs. Ultimately, the presentation demonstrates how a tight co-design loop across infrastructure layers provides a scalable, future-ready roadmap for Meta’s rapidly evolving AI demand.

To address the unique requirements of AI networking, we leverage our proven Jupiter network for north-south traffic and are introducing Virgo, a new fabric for east-west scale-out communication. Virgo is a dedicated accelerator-to-accelerator fabric optimized for massive horizontal scale. It is engineered for deterministic latency and maximum resilience to provide high goodput for the ML workload. Virgo will enable the new Google Cloud's A5x bare-metal instances, which run the NVIDIA Vera Rubin NVL72 platform. It is also a key building block for next-gen TPUs. This talk will discuss Virgo's design principles for resilience, observability, deployability, and multi-building scale.

Physical network design—deciding how thousands of network devices and millions of fibers are distributed across data centers, racks, and failure domains—has traditionally been a manual, time-consuming process performed by design engineers. Yet we are seeing exponentially increasing demand to fulfill the capacity required by AI workloads. At Meta, we developed Loom, a system to automate and optimize the end-to-end physical network design pipeline. Loom orchestrates two core components: a device placement component (Planogram) which places network devices across the physical floorplan subject to dimensional and failure-domain constraints; and a fiber design component, which generates end-to-end fiber connectivity satisfying the variety of connectivity patterns driven by physical constraints. Together, these components accelerate end-to-end physical network design from months to hours, allowing us to produce more complex and efficient designs with fewer resources.

Live Panel with Coreweave, Meta, OpenAI, Oracle
Moderated by Rajiv Krishnamurthy

Training frontier AI models requires connecting tens of thousands of accelerators into a single, tightly coupled fabric — pushing scale-out network design into uncharted territory. This panel brings together engineers building some of the world's largest AI training networks to discuss the architectures, transport protocols, and operational lessons behind today's supercomputer-scale fabrics. Topics will include topology choices, congestion control at extreme scale, and the co-design between network infrastructure and distributed training frameworks. Expect a practitioner-level conversation on what it takes to keep massive AI clusters connected, fast, and reliable.

Prometheus introduced significant complexity for connectivity. We were faced with significantly higher fiber count requirements, highly dynamic demand location, and compressed timelines for delivery. Tune in to find out the innovations required to meet the demand across multiple data centers in the cluster.

Network capacity planning is high-stakes and unforgiving: capacity is committed years ahead, every long-range forecast is guaranteed to be wrong, and the costliest surprises are correlated demand shocks that defeat traditional risk-pooling. Much of the pipeline is already automated — but the decisions that take real judgment have stayed manual, and those are the ones that decide whether a wrong forecast corners you. This talk is about applying AI to exactly those decisions — not to chase a more accurate forecast, but to position capacity so it keeps our options open no matter how demand breaks. We show how agents are beginning to crack the three that matter most: acting inside narrow buy windows (triaging signals continuously to surface the handful of truly novel sites, weeks down to minutes), trusting data across systems that disagree on even basic definitions (a measure-enrich-regrade loop), and choosing among coupled, timing-sensitive levers (a ranked, costed menu of moves with the reasoning attached). The real frontier is a plan that learns — agents that remember every decision and how it turned out, so human judgment compounds over time instead of resetting with each planner.

Designing a datacenter network is where a capacity requirement becomes a blueprint: assigning device roles, wiring the interconnect, laying out addressing and failure domains, and producing the bill of materials — all while keeping the design constrained, connected, and correct. It is some of the most judgment-heavy work in network engineering, and until recently it resisted automation. This talk walks through how that is starting to change. Today, agents assist with discrete steps — sizing a fabric, proposing a topology, analyzing a change — but the trajectory points toward a single agent that can own the design end-to-end. Drawing on real waypoints from Meta's fabrics, Maaz shows what already works, where the hard problems remain (validation, above all), and how the engineer's role shifts from author to approver. The through-line: humans still make the call — they just no longer have to draw every line.

The network capacity lifecycle, from demand signal to production operations, is a pipeline. Each stage consumes the output of the previous one, and failures compound downstream. A bad forecast leads to a bad design; a bad design leads to a painful deployment; a painful deployment leads to operational fragility.

The AI-native effort asks: what if every stage of this pipeline had an AI agent as a first-class participant? An AI agent can act as a force multiplier, automating the mechanical, surfacing the non-obvious, and closing feedback loops that today take weeks or months.

When operating a large-scale production network serving billions of users, we face daily challenges in running network infrastructure reliably and efficiently. AI4NI @Runtime tackles these challenges across three dimensions: reliability, performance, and efficiency. At Meta, we’ve begun developing AI agents in each area and have seen promising early results. In this talk, we’ll share our solution architecture, example agentic systems, and initial findings from production trials, along with the key technical challenges and our planned directions to address them.

Live Panel with Google, Meta, Microsoft, and Oracle
Moderated by Omar Baldonado

Today's largest AI fleets span multiple generations of technology — from power and buildings to fiber and accelerators — and must be delivered at a pace that pushes the entire supply ecosystem to its limits. This panel brings together leaders from hyperscaler infrastructure planning and technical sourcing organizations to discuss how they navigate this complexity. The conversation will explore long-range fleet strategy, multi-vendor orchestration, and the engineering realities of operating at the roofline of industry capacity. Expect a candid discussion on what it takes to plan and deliver AI infrastructure at gigawatt scale.