){kind=logo}
){kind=logo}
FULL-ACCESS & VIP TICKETS ONLY
Main Stage
-
Main Stage
-
Main Stage
-
Main StageAgentic AI is transforming the demands placed on computing infrastructure. As AI systems reason, plan, use tools, and coordinate across increasingly complex workflows, the industry must rethink how co …
-
Main StageSpeakers
-
Main Stage
-
Main StageTwo realities have become increasingly clear. First, AI has the potential to deliver a step change in business outcomes for every enterprise. Second, translating that potential into real-world busines …
-
Main Stage
-
Main Stage
-
Lunch
-
Main StageFor two decades, AI infrastructure has been built on the assumption that memory is cheap and plentiful. That era is over. HBM is sold out through 2026, supply constraints are projected to persist into …
-
Main Stage
-
Main Stage
-
Main Stage
-
Main Stage
-
Main Stage
-
Main StageBefore Positron, Mitesh Agrawal ran one of the world's largest GPU clouds—which put him across the table from nearly every AI chip startup pitch of the past decade. The technology was often novel. He …Speakers
Data & Models
-
Data & Models
-
Data & Models
-
Data & Models Track
-
Data & Models TrackThe teams pulling ahead with AI aren't just buying access to someone else's model — they own their intelligence: models that run on their schedule, use their weights, and are trained on their data, at …
-
Data & Models Track
-
Data & Models Track
-
Lunch
-
Data & Models Track
-
Data & Models TrackChairperson
-
Data & Models Track
-
Data & Models Track
-
Data & Models TrackYour GPUs got faster this year. Your cost per token didn't budge. That gap is the misdiagnosis at the center of inference economics: the bottleneck isn't compute—it's memory and data movement, and age …Speakers
-
Data & Models Track
-
Data & Models TrackHistorically, resilience defined system limits. Now, agentic AI is shifting that constraint to scale. Always-on agents create continuous, machine-driven demand that compounds across APIs, services, an …
-
Data & Models TrackWhat happens when intelligence becomes a utility? As AI rapidly moves from experimentation to mission-critical infrastructure, enterprises face a fundamental strategic choice: should they build and ow …
Compute
-
Compute Track
-
Compute
-
Compute
-
Compute Track
-
Compute Track
-
Compute Track
-
Compute Track
-
Lunch
-
Compute TrackAs AI moves into the inference era, scaling real-world deployments becomes the core challenge. This session explores how infrastructure built for production-scale inference is key to enabling sovereig …
-
Compute Track
-
Compute Track
-
Compute Track
-
Compute TrackAs AI data centers evolve to serve diverse models and workload patterns, the era of “one GPU does everything” is ending. Achieving optimal tokens per dollar now depends on effectively combining hetero …Speakers
-
Compute Track
-
Compute TrackModern AI clusters require far more than high-speed switching. Large GPU deployments combine edge networking, Ethernet fabrics, InfiniBand, NVLink, and host/DPU networking into a tightly coupled syste …
-
Compute TrackAgentic AI is becoming the new operational baseline, but scaling on closed stacks can increase cost and limit flexibility. Learn how you can enable open, production-ready AI. See how the build an arch …
Data Movement
-
Data Movement Track
-
Data Movement
-
Data Movement
-
Data Movement TrackAs AI shifts from static models to swarms of agents, much of today’s infrastructure is still optimized for yesterday’s workloads. The real bottleneck is no longer just GPU capacity; it is the intercon …Speakers
-
Data Movement Track
-
Data Movement TrackThe “network wall” has become one of the primary bottlenecks in large-scale AI training and inference. While compute FLOPS have doubled every 18 months, switch ASIC capacity and radix have historicall …
-
Lunch
-
Data Movement Track
-
Data Movement Track
-
Data Movement TrackSpeakers
-
Data Movement Track
-
Data Movement Track
-
Data Movement Track
-
Data Movement TrackDiscover how Xinnor, Celestica, and Megware eliminated storage bottlenecks on two of Germany’s largest University supercomputers, scaling up to 1500 compute nodes and reaching #3 on the IO500 list. We …
-
Data Movement Track
AI Data Center
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center TrackAs demand for AI accelerates, infrastructure leaders face unprecedented uncertainty and opportunity. Join Kevin Janda, Corporate Vice President of Cloud Operations + Innovation at Microsoft, for a can …
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center TrackSpeakers
-
AI Data Center Track
-
AI Data Center TrackRack power is projected to increase from roughly 30 kW to 1 MW in just four years, and the industry has converged on 800 VDC at the rack because it is the only practical way to reduce current, minimiz …
-
Lunch
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
Physical AI
-
Physical AI Track
-
Physical AI TrackPhysical AI is moving intelligence out of the data center and into the real world, powering machines that can percieve, reason, and act. From robotics and autonomous systems to edge compute and next-g …
-
Physical AI TrackBringing AI into the physical world is one of the most important — and most difficult — frontiers in the field today. In this keynote conversation, Archetype AI Founder and CEO Ivan Poupyrev sits down …
-
Physical AI TrackAs AI accelerates demand for data-center capacity, infrastructure must evolve just as fast. This session from Tamas Foldi explores how Physical AI and robotics can compress data-center build and commi …
-
Physical AI Track
-
Physical AI TrackPhysical AI is no longer a concept on a roadmap — it is actively reshaping manufacturing floors, hospital systems, logistics networks, and energy infrastructure around the world. The industry has reac …
-
Physical AI TrackThis panel will explore how engineers, platform developers, and model builders are working to close the sim-to-real gap and engineer robots that can act with genuine autonomy when deployed in the real …
-
Physical AI TrackRecent advances in foundation models have transformed what is possible in robotics and autonomous systems. Yet, one of the biggest challenges remains generalization: how can an intelligent system that …
-
Lunch
-
Physical AI TrackSurgical robotics is not a new field, but, as robotic systems become more autonomous and begin operating in safety-critical, human-centric environments, trust is paramount. This intimate discussion wi …
-
Physical AI TrackThis session gathers leaders in robotics, IIoT, and construction to discuss the critical infrastructure—the rugged hardware, edge connectivity, and real-time sensory systems—required to bring AI into …
-
Physical AI TrackThe next generation of AI won’t live on screens – it will operate robots, autonomous systems, factories, vehicles, medical devices, and critical infrastructure. To succeed, these Physical AI systems m …
-
Physical AI Track
-
Physical AI TrackAs Physical AI demonstrates its value across industries, this presentation will explore the challenges Boeing are facing when applying Physical AI to their aerospace operations. Bala will outline the …
-
Physical AI TrackIndustrial robots are moving beyond rigid, waypoint-based programming towards true autonomy. This session will explore the technologies and architectures that are emerging in both research and industr …
-
Physical AI TrackPhysical AI systems (robots, autonomous machines, industrial inspection platforms, AI PCs) cannot fall back on the data center. Control loops demand deterministic latency, connectivity is intermittent …
-
Physical AI TrackManufacturing is emerging as a key proving ground for Physical AI, where intelligent systems can drive efficiency, adaptability, and resilience on the factory floor. In this session, hear how Novelis …
Main Stage
-
Main Stage
-
Main StageThe cost of AI is growing exponentially, but so is the opportunity. When the cost of intelligence drops by orders of magnitude, customers will build things nobody is even attempting today. Getting the …
-
Main StageBuilding AI superclusters requires tightly integrated networking, storage, and architectures which deliver predictable performance across hundreds of thousands of GPUs and CPUs. Join Oracle and NVIDIA …
-
Main Stage
-
Main StageMLCommons unveils the results of MLPerf Inference v6.1, the newest round of the industry's most trusted benchmark for AI system performance. We will start with a quick word on who MLCommons is and why …
-
Main Stage
-
Main Stage
-
Main Stage
-
Lunch
-
Main Stage
-
Main StageAs enterprises move from isolated model interactions to agentic systems, infrastructure is becoming the defining challenge. A single request may trigger dozens of model interactions, retrieval operati …
-
Main StageAs AI evolves from chatbots and copilots to autonomous agents, the demands placed on data centers are changing dramatically. Agentic AI is driving an explosion in inference workloads, memory requireme …
-
Main Stage
-
Main Stage
-
Main Stage
-
Main Stage
-
Main StageThe world faces insatiable demand for AI inference at the very moment compute is becoming constrained by cost, latency, and energy consumption. The infrastructure that sparked the AI revolution was no …
-
Main StageAI factories and neoclouds are pushing storage architectures to their breaking point—massive multimodal datasets, tens of thousands of GPUs demanding constant feeding, and metadata overhead that grind …
Data & Models
-
Data & Models Track
-
Data & Models
-
Data & Models
-
Data & Models Track
-
Data & Models Track
-
Data & Models Track
-
Data & Models TrackBy 2028, Gartner expects AI coding costs to overtake the average developer’s salary as token consumption surges. That is a symptom of a bigger question the AI infrastructure community is already askin …
-
Lunch
-
Data & Models Track
-
Data & Models Track
-
-
Data & Models Track
-
Data & Models Track
-
Data & Models Track
-
Data & Models TrackA trillion-dollar IPO throws off more data than any desk can read unaided: an options tape measured in terabytes a day, thousands of pages of filings, an endless stream of news. Using the SpaceX debut …
-
Data & Models Track
-
Data & Models Track
-
Data & Models TrackCompliance activities in financial services and insurance remain dominated by manual checks and siloed rule engines, despite increasing transaction volumes and regulatory complexity. This keynote expl …
Compute
-
Compute Track
-
Compute
-
Compute
-
Compute Track
-
Compute TrackSpeakers
-
Compute TrackJoin chip design startup founders, VCs and leaders from the semiconductor ecosystem for an engaging discussion on what it takes to build a chip design startup from the ground up. From idea incubation …
-
Compute TrackAs AI continues to move to the edge, system architects face growing demands for higher performance, software portability, and development efficiency. At the same time, increasingly complex workloads a …Speakers
-
Lunch
-
Compute Track
-
Compute TrackThe underlying hardware of AI infrastructure contains severe vulnerabilities, is increasingly targeted by nation-states and is putting the AI Economy at risk that’s estimated to be $15T in additional …
-
Compute TrackEvery keynote this week has a capacity number in it: gigawatts, hundreds of thousands of chips, ten-year contracts. This panel covers the gap nobody talks about, the distance between the announcement …
-
Compute Track
-
Compute TrackFor NeoCloud operators, acquiring GPU capacity is only the beginning. Once the infrastructure is online, the commercial clock starts: the immediate priority is turning that capacity into a service cus …
-
Compute Track
-
Compute TrackSpeakers
-
Compute Track
Data Movement
-
Data Movement Track
-
Data Movement
-
Data Movement
-
Data Movement TrackAI's data gravity is reshaping the storage stack. As context windows expand, agentic workflows multiply, and KV-cache and CMX tiers grow beyond what DRAM and HBM can economically hold, NVMe SSDs are e …
-
Data Movement TrackAs frontier models scale to trillions of parameters and employ increasingly advanced architectures, memory remains a critical determinant of system performance, efficiency and reliability. In this pan …
-
Data Movement TrackThe rapid evolution of agentic AI is fundamentally reshaping AI infrastructure. As AI systems become capable of planning, reasoning, retaining memory, and orchestrating multiple agents, the demand for …
-
Lunch
-
Data Movement TrackSpeakers
-
Data Movement Track
-
Data Movement TrackAs LLM services diversify from ultra-low-latency interactions to cost-sensitive, long-context workloads, a homogeneous GPU-HBM architecture is no longer sufficient. This presentation introduces a work …Speakers
-
Data Movement Track
-
Data Movement TrackAgentic AI workloads—with long prompt contexts, multi-turn tool calls, and execution loops—are moving memory, not just compute, onto the critical path. As memory demands grow, memory stranding compoun …
-
Data Movement Track
-
Data Movement Track
-
Data Movement Track
AI Data Center
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center TrackCompute is becoming a strategic asset in the AI economy. As organizations move from AI experimentation to scaled deployment, access to the right compute infrastructure will determine their ability to …
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
Lunch
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center TrackLegacy infrastructure management can't keep pace with rapid AI data center scaling, complex and expensive high-density racks, ballooning power demands, and increasing threats from sophisticated attack …
-
AI Data Center Track
-
Future-Proofing AI Data Centers: Designing for Rapid Hardware Evolution & Infrastructure FlexibilityAI Data Center Track
-
AI Data Center TrackThe innovation cycles for AI data centers as well as co-design specialization can boost performance but trade off supply chain strength, increasing costs and reducing efficiency. To solve this challen …
-
AI Data Center TrackEverything you've learned about data centers is now obsolete in the AI age. Rack densities are increasing by as much as 100X, electricity management presents serious new challenges and risks, and netw …
Physical AI
-
Physical AI Track
-
Physical AI TrackAI-powered humanoids are rapidly advancing from prototype to real-world deployment, but scaling them remains a complex challenge. This session explores the key technological, operational, and economic …
-
Physical AI TrackHumanoid robots expose the hardest version of the compute-energy problem. High-performance AI for perception, planning and control must be balanced within strict power, thermal & form-factor limits, c …
-
Physical AI TrackThe roads we've paved are already mapped. Yet the ones that matter most, leading to safety, discovery, and survival, are not. In this session, see how a Vision-Language-Navigation enabled autonomous r …
-
Physical AI Track
-
Physical AI TrackRobot performance shows what is possible, but safety determines whether a system can be deployed. This session shows how NVIDIA Halos for Robotics provides a full-stack safety system spanning industri …
-
Physical AI TrackThis panel will address the most limiting factor facing humanoid deployment today – we do not have enough, good data. The discussion will explore what it means for humanoid training data to be “good”, …
-
Physical AI TrackThe next major leap in AI is not only just processing texts, but also predicting and interacting with the physical world. World models and physical AI bridge digital reasoning with real-world executio …
-
Lunch
-
Physical AI TrackRecent advancements in Physical AI are accelerating the deployment of humanoids in people-centric, unstructured environments. This presentation explores the key engineering innovation enabling safe hu …
-
Physical AI TrackPhysical AI systems are increasingly being deployed in a variety of settings, from road to the factory floor, but the model architecture that will unlock the massive scaling potential of autonomous, e …
-
Physical AI TrackQuadric's Chimera(R) general purpose NPU is a Python & C++ programmable processor natively designed for matrix computation at the heart of AI models, not a data center architecture squeezed into a car …
-
Physical AI Track
-
Physical AI TrackPhysical AI has scaled its training stack remarkably fast. Compute, parallel simulation, policy optimization, and synthetic data have all come a long way in just a few years. What has not kept pace is …
-
Physical AI TrackHow do you integrate Physical AI efficiently & effectively into existing manufacturing or construction systems? This panel brings together leaders from the compute & software worlds, alongside the ent …
-
Physical AI TrackPhysical AI won't be won by just models alone, real-world deployment depends on the integration of hardware, software, orchestration and environment constraints. In this session, hear why success in d …
-
Physical AI TrackDeploying Physical AI in real-world environments comes with unique challenges—from integration and data readiness to scalability and operational impact. Yet organizations that get it right are already …
Main Stage
-
Main Stage
-
Main StageAI is entering a new phase. As intelligent systems become more capable and AI workloads expand beyond training to inference, agents, retrieval, orchestration, and physical AI, the demands on infrastru …
-
Main Stage
-
Main Stage
-
Main Stage
-
Main Stage
-
Main Stage
-
Main Stage
-
Lunch
Data & Models
-
Data & Models Track
-
Data & Models Track
-
Data & Models Track
-
Data & Models Track
-
Data & Models TrackLarge synchronous training jobs can turn one unhealthy GPU into a cluster-wide interruption. A PCIe disconnect, Xid error, degraded fabric, or planned maintenance can terminate hundreds of coordinated …
-
Data & Models Track
-
Data & Models Track
-
Lunch
-
Data & Models Track
-
Data & Models Track
-
Data & Models Track
-
Data & Models Track
Compute
-
Compute Track
-
Compute Track
-
Compute Track
-
Compute Track
-
Compute Track
-
Compute Track
-
Compute TrackAs inference scales, performance, power, utilization, and infrastructure efficiency increasingly shape the economics of AI. This keynote explores how FuriosaAI's purpose-built compute and hardware-sof …
-
Lunch
-
Compute TrackDive into the journey of AI workloads on AWS Trainium3 — from distributed training to fine-tuning and deployment. This talk showcases how to optimize each stage using the Neuron SDK. We'll demonstrate …
-
Compute Track
-
Compute Track
AI Data Center
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
Lunch
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center TrackEvery intelligence inherits the conditions it was formed in. When those conditions are extractive, intelligence compounds extraction — grids that burn, water that leaves, value that passes through and …
-
AI Data Center Track
FULL-ACCESS & VIP TICKETS ONLY
Main Stage
-
Main Stage
-
Main Stage
-
Main StageAgentic AI is transforming the demands placed on computing infrastructure. As AI systems reason, plan, use tools, and coordinate across increasingly complex workflows, the industry must rethink how co …
-
Main StageSpeakers
-
Main Stage
-
Main StageTwo realities have become increasingly clear. First, AI has the potential to deliver a step change in business outcomes for every enterprise. Second, translating that potential into real-world busines …
-
Main Stage
-
Main Stage
-
Lunch
-
Main StageFor two decades, AI infrastructure has been built on the assumption that memory is cheap and plentiful. That era is over. HBM is sold out through 2026, supply constraints are projected to persist into …
-
Main Stage
-
Main Stage
-
Main Stage
-
Main Stage
-
Main Stage
-
Main StageBefore Positron, Mitesh Agrawal ran one of the world's largest GPU clouds—which put him across the table from nearly every AI chip startup pitch of the past decade. The technology was often novel. He …Speakers
Data & Models
-
Data & Models
-
Data & Models
-
Data & Models Track
-
Data & Models TrackThe teams pulling ahead with AI aren't just buying access to someone else's model — they own their intelligence: models that run on their schedule, use their weights, and are trained on their data, at …
-
Data & Models Track
-
Data & Models Track
-
Lunch
-
Data & Models Track
-
Data & Models TrackChairperson
-
Data & Models Track
-
Data & Models Track
-
Data & Models TrackYour GPUs got faster this year. Your cost per token didn't budge. That gap is the misdiagnosis at the center of inference economics: the bottleneck isn't compute—it's memory and data movement, and age …Speakers
-
Data & Models Track
-
Data & Models TrackHistorically, resilience defined system limits. Now, agentic AI is shifting that constraint to scale. Always-on agents create continuous, machine-driven demand that compounds across APIs, services, an …
-
Data & Models TrackWhat happens when intelligence becomes a utility? As AI rapidly moves from experimentation to mission-critical infrastructure, enterprises face a fundamental strategic choice: should they build and ow …
Compute
-
Compute Track
-
Compute
-
Compute
-
Compute Track
-
Compute Track
-
Compute Track
-
Compute Track
-
Lunch
-
Compute TrackAs AI moves into the inference era, scaling real-world deployments becomes the core challenge. This session explores how infrastructure built for production-scale inference is key to enabling sovereig …
-
Compute Track
-
Compute Track
-
Compute Track
-
Compute TrackAs AI data centers evolve to serve diverse models and workload patterns, the era of “one GPU does everything” is ending. Achieving optimal tokens per dollar now depends on effectively combining hetero …Speakers
-
Compute Track
-
Compute TrackModern AI clusters require far more than high-speed switching. Large GPU deployments combine edge networking, Ethernet fabrics, InfiniBand, NVLink, and host/DPU networking into a tightly coupled syste …
-
Compute TrackAgentic AI is becoming the new operational baseline, but scaling on closed stacks can increase cost and limit flexibility. Learn how you can enable open, production-ready AI. See how the build an arch …
Data Movement
-
Data Movement Track
-
Data Movement
-
Data Movement
-
Data Movement TrackAs AI shifts from static models to swarms of agents, much of today’s infrastructure is still optimized for yesterday’s workloads. The real bottleneck is no longer just GPU capacity; it is the intercon …Speakers
-
Data Movement Track
-
Data Movement TrackThe “network wall” has become one of the primary bottlenecks in large-scale AI training and inference. While compute FLOPS have doubled every 18 months, switch ASIC capacity and radix have historicall …
-
Lunch
-
Data Movement Track
-
Data Movement Track
-
Data Movement TrackSpeakers
-
Data Movement Track
-
Data Movement Track
-
Data Movement Track
-
Data Movement TrackDiscover how Xinnor, Celestica, and Megware eliminated storage bottlenecks on two of Germany’s largest University supercomputers, scaling up to 1500 compute nodes and reaching #3 on the IO500 list. We …
-
Data Movement Track
AI Data Center
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center TrackAs demand for AI accelerates, infrastructure leaders face unprecedented uncertainty and opportunity. Join Kevin Janda, Corporate Vice President of Cloud Operations + Innovation at Microsoft, for a can …
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center TrackSpeakers
-
AI Data Center Track
-
AI Data Center TrackRack power is projected to increase from roughly 30 kW to 1 MW in just four years, and the industry has converged on 800 VDC at the rack because it is the only practical way to reduce current, minimiz …
-
Lunch
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
Physical AI
-
Physical AI Track
-
Physical AI TrackPhysical AI is moving intelligence out of the data center and into the real world, powering machines that can percieve, reason, and act. From robotics and autonomous systems to edge compute and next-g …
-
Physical AI TrackBringing AI into the physical world is one of the most important — and most difficult — frontiers in the field today. In this keynote conversation, Archetype AI Founder and CEO Ivan Poupyrev sits down …
-
Physical AI TrackAs AI accelerates demand for data-center capacity, infrastructure must evolve just as fast. This session from Tamas Foldi explores how Physical AI and robotics can compress data-center build and commi …
-
Physical AI Track
-
Physical AI TrackPhysical AI is no longer a concept on a roadmap — it is actively reshaping manufacturing floors, hospital systems, logistics networks, and energy infrastructure around the world. The industry has reac …
-
Physical AI TrackThis panel will explore how engineers, platform developers, and model builders are working to close the sim-to-real gap and engineer robots that can act with genuine autonomy when deployed in the real …
-
Physical AI TrackRecent advances in foundation models have transformed what is possible in robotics and autonomous systems. Yet, one of the biggest challenges remains generalization: how can an intelligent system that …
-
Lunch
-
Physical AI TrackSurgical robotics is not a new field, but, as robotic systems become more autonomous and begin operating in safety-critical, human-centric environments, trust is paramount. This intimate discussion wi …
-
Physical AI TrackThis session gathers leaders in robotics, IIoT, and construction to discuss the critical infrastructure—the rugged hardware, edge connectivity, and real-time sensory systems—required to bring AI into …
-
Physical AI TrackThe next generation of AI won’t live on screens – it will operate robots, autonomous systems, factories, vehicles, medical devices, and critical infrastructure. To succeed, these Physical AI systems m …
-
Physical AI Track
-
Physical AI TrackAs Physical AI demonstrates its value across industries, this presentation will explore the challenges Boeing are facing when applying Physical AI to their aerospace operations. Bala will outline the …
-
Physical AI TrackIndustrial robots are moving beyond rigid, waypoint-based programming towards true autonomy. This session will explore the technologies and architectures that are emerging in both research and industr …
-
Physical AI TrackPhysical AI systems (robots, autonomous machines, industrial inspection platforms, AI PCs) cannot fall back on the data center. Control loops demand deterministic latency, connectivity is intermittent …
-
Physical AI TrackManufacturing is emerging as a key proving ground for Physical AI, where intelligent systems can drive efficiency, adaptability, and resilience on the factory floor. In this session, hear how Novelis …
Main Stage
-
Main Stage
-
Main StageThe cost of AI is growing exponentially, but so is the opportunity. When the cost of intelligence drops by orders of magnitude, customers will build things nobody is even attempting today. Getting the …
-
Main StageBuilding AI superclusters requires tightly integrated networking, storage, and architectures which deliver predictable performance across hundreds of thousands of GPUs and CPUs. Join Oracle and NVIDIA …
-
Main Stage
-
Main StageMLCommons unveils the results of MLPerf Inference v6.1, the newest round of the industry's most trusted benchmark for AI system performance. We will start with a quick word on who MLCommons is and why …
-
Main Stage
-
Main Stage
-
Main Stage
-
Lunch
-
Main Stage
-
Main StageAs enterprises move from isolated model interactions to agentic systems, infrastructure is becoming the defining challenge. A single request may trigger dozens of model interactions, retrieval operati …
-
Main StageAs AI evolves from chatbots and copilots to autonomous agents, the demands placed on data centers are changing dramatically. Agentic AI is driving an explosion in inference workloads, memory requireme …
-
Main Stage
-
Main Stage
-
Main Stage
-
Main Stage
-
Main StageThe world faces insatiable demand for AI inference at the very moment compute is becoming constrained by cost, latency, and energy consumption. The infrastructure that sparked the AI revolution was no …
-
Main StageAI factories and neoclouds are pushing storage architectures to their breaking point—massive multimodal datasets, tens of thousands of GPUs demanding constant feeding, and metadata overhead that grind …
Data & Models
-
Data & Models Track
-
Data & Models
-
Data & Models
-
Data & Models Track
-
Data & Models Track
-
Data & Models Track
-
Data & Models TrackBy 2028, Gartner expects AI coding costs to overtake the average developer’s salary as token consumption surges. That is a symptom of a bigger question the AI infrastructure community is already askin …
-
Lunch
-
Data & Models Track
-
Data & Models Track
-
-
Data & Models Track
-
Data & Models Track
-
Data & Models Track
-
Data & Models TrackA trillion-dollar IPO throws off more data than any desk can read unaided: an options tape measured in terabytes a day, thousands of pages of filings, an endless stream of news. Using the SpaceX debut …
-
Data & Models Track
-
Data & Models Track
-
Data & Models TrackCompliance activities in financial services and insurance remain dominated by manual checks and siloed rule engines, despite increasing transaction volumes and regulatory complexity. This keynote expl …
Compute
-
Compute Track
-
Compute
-
Compute
-
Compute Track
-
Compute TrackSpeakers
-
Compute TrackJoin chip design startup founders, VCs and leaders from the semiconductor ecosystem for an engaging discussion on what it takes to build a chip design startup from the ground up. From idea incubation …
-
Compute TrackAs AI continues to move to the edge, system architects face growing demands for higher performance, software portability, and development efficiency. At the same time, increasingly complex workloads a …Speakers
-
Lunch
-
Compute Track
-
Compute TrackThe underlying hardware of AI infrastructure contains severe vulnerabilities, is increasingly targeted by nation-states and is putting the AI Economy at risk that’s estimated to be $15T in additional …
-
Compute TrackEvery keynote this week has a capacity number in it: gigawatts, hundreds of thousands of chips, ten-year contracts. This panel covers the gap nobody talks about, the distance between the announcement …
-
Compute Track
-
Compute TrackFor NeoCloud operators, acquiring GPU capacity is only the beginning. Once the infrastructure is online, the commercial clock starts: the immediate priority is turning that capacity into a service cus …
-
Compute Track
-
Compute TrackSpeakers
-
Compute Track
Data Movement
-
Data Movement Track
-
Data Movement
-
Data Movement
-
Data Movement TrackAI's data gravity is reshaping the storage stack. As context windows expand, agentic workflows multiply, and KV-cache and CMX tiers grow beyond what DRAM and HBM can economically hold, NVMe SSDs are e …
-
Data Movement TrackAs frontier models scale to trillions of parameters and employ increasingly advanced architectures, memory remains a critical determinant of system performance, efficiency and reliability. In this pan …
-
Data Movement TrackThe rapid evolution of agentic AI is fundamentally reshaping AI infrastructure. As AI systems become capable of planning, reasoning, retaining memory, and orchestrating multiple agents, the demand for …
-
Lunch
-
Data Movement TrackSpeakers
-
Data Movement Track
-
Data Movement TrackAs LLM services diversify from ultra-low-latency interactions to cost-sensitive, long-context workloads, a homogeneous GPU-HBM architecture is no longer sufficient. This presentation introduces a work …Speakers
-
Data Movement Track
-
Data Movement TrackAgentic AI workloads—with long prompt contexts, multi-turn tool calls, and execution loops—are moving memory, not just compute, onto the critical path. As memory demands grow, memory stranding compoun …
-
Data Movement Track
-
Data Movement Track
-
Data Movement Track
AI Data Center
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center TrackCompute is becoming a strategic asset in the AI economy. As organizations move from AI experimentation to scaled deployment, access to the right compute infrastructure will determine their ability to …
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
Lunch
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center TrackLegacy infrastructure management can't keep pace with rapid AI data center scaling, complex and expensive high-density racks, ballooning power demands, and increasing threats from sophisticated attack …
-
AI Data Center Track
-
Future-Proofing AI Data Centers: Designing for Rapid Hardware Evolution & Infrastructure FlexibilityAI Data Center Track
-
AI Data Center TrackThe innovation cycles for AI data centers as well as co-design specialization can boost performance but trade off supply chain strength, increasing costs and reducing efficiency. To solve this challen …
-
AI Data Center TrackEverything you've learned about data centers is now obsolete in the AI age. Rack densities are increasing by as much as 100X, electricity management presents serious new challenges and risks, and netw …
Physical AI
-
Physical AI Track
-
Physical AI TrackAI-powered humanoids are rapidly advancing from prototype to real-world deployment, but scaling them remains a complex challenge. This session explores the key technological, operational, and economic …
-
Physical AI TrackHumanoid robots expose the hardest version of the compute-energy problem. High-performance AI for perception, planning and control must be balanced within strict power, thermal & form-factor limits, c …
-
Physical AI TrackThe roads we've paved are already mapped. Yet the ones that matter most, leading to safety, discovery, and survival, are not. In this session, see how a Vision-Language-Navigation enabled autonomous r …
-
Physical AI Track
-
Physical AI TrackRobot performance shows what is possible, but safety determines whether a system can be deployed. This session shows how NVIDIA Halos for Robotics provides a full-stack safety system spanning industri …
-
Physical AI TrackThis panel will address the most limiting factor facing humanoid deployment today – we do not have enough, good data. The discussion will explore what it means for humanoid training data to be “good”, …
-
Physical AI TrackThe next major leap in AI is not only just processing texts, but also predicting and interacting with the physical world. World models and physical AI bridge digital reasoning with real-world executio …
-
Lunch
-
Physical AI TrackRecent advancements in Physical AI are accelerating the deployment of humanoids in people-centric, unstructured environments. This presentation explores the key engineering innovation enabling safe hu …
-
Physical AI TrackPhysical AI systems are increasingly being deployed in a variety of settings, from road to the factory floor, but the model architecture that will unlock the massive scaling potential of autonomous, e …
-
Physical AI TrackQuadric's Chimera(R) general purpose NPU is a Python & C++ programmable processor natively designed for matrix computation at the heart of AI models, not a data center architecture squeezed into a car …
-
Physical AI Track
-
Physical AI TrackPhysical AI has scaled its training stack remarkably fast. Compute, parallel simulation, policy optimization, and synthetic data have all come a long way in just a few years. What has not kept pace is …
-
Physical AI TrackHow do you integrate Physical AI efficiently & effectively into existing manufacturing or construction systems? This panel brings together leaders from the compute & software worlds, alongside the ent …
-
Physical AI TrackPhysical AI won't be won by just models alone, real-world deployment depends on the integration of hardware, software, orchestration and environment constraints. In this session, hear why success in d …
-
Physical AI TrackDeploying Physical AI in real-world environments comes with unique challenges—from integration and data readiness to scalability and operational impact. Yet organizations that get it right are already …
Main Stage
-
Main Stage
-
Main StageAI is entering a new phase. As intelligent systems become more capable and AI workloads expand beyond training to inference, agents, retrieval, orchestration, and physical AI, the demands on infrastru …
-
Main Stage
-
Main Stage
-
Main Stage
-
Main Stage
-
Main Stage
-
Main Stage
-
Lunch
Data & Models
-
Data & Models Track
-
Data & Models Track
-
Data & Models Track
-
Data & Models Track
-
Data & Models TrackLarge synchronous training jobs can turn one unhealthy GPU into a cluster-wide interruption. A PCIe disconnect, Xid error, degraded fabric, or planned maintenance can terminate hundreds of coordinated …
-
Data & Models Track
-
Data & Models Track
-
Lunch
-
Data & Models Track
-
Data & Models Track
-
Data & Models Track
-
Data & Models Track
Compute
-
Compute Track
-
Compute Track
-
Compute Track
-
Compute Track
-
Compute Track
-
Compute Track
-
Compute TrackAs inference scales, performance, power, utilization, and infrastructure efficiency increasingly shape the economics of AI. This keynote explores how FuriosaAI's purpose-built compute and hardware-sof …
-
Lunch
-
Compute TrackDive into the journey of AI workloads on AWS Trainium3 — from distributed training to fine-tuning and deployment. This talk showcases how to optimize each stage using the Neuron SDK. We'll demonstrate …
-
Compute Track
-
Compute Track
AI Data Center
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center Track
-
Lunch
-
AI Data Center Track
-
AI Data Center Track
-
AI Data Center TrackEvery intelligence inherits the conditions it was formed in. When those conditions are extractive, intelligence compounds extraction — grids that burn, water that leaves, value that passes through and …
-
AI Data Center Track