Industry Trends & Technology Innovation - Semiconductor industry scales AI infrastructure to meet growing performance, reliability, and flexibility demands for modern AI training and inference workloads [1] - Networking technology adapts Ethernet-based AI clusters through Multipath Reliable Connection (MRC) to enhance bandwidth utilization, maintain data integrity, and optimize traffic flow [1] Core Technology Features - Advanced network solutions leverage intelligent packet spray, out-of-order packet handling, and path-aware congestion control to support large-scale AI deployment [1]

System Architecture & Hardware Specifications - AMD Helios architecture integrates 72 MI455 GPUs per rack, supported by 18 compute nodes with each node containing 1 CPU and 4 GPUs[8][64] - Scale-up interconnect utilizes 12 Broadcom Tomahawk 6 switch ASICs (A6) connected to each GPU via three 400-gigabit Ethernet links, creating 36 parallel network planes[11][12] - Tomahawk 6 chip series includes models with 1024 100-gigabit SerDes achieving 102 terabits per second, and 200-gigabit SerDes variants built on 3-nanometer technology[71][72][73] - Each MI455 GPU provides 23 terabytes per second of high-bandwidth memory (HBM) bandwidth and 3.6% terabytes per second of scale-up bandwidth[54][55] Network Technology & Performance - Scale-up network operates on Ethernet technology using UAL-OE (UALink over Ethernet) protocol, delivering 1.8 terabytes per second of bandwidth per GPU with single-layer switching topology[11][15][25] - Reliability and resilience mechanisms allow the system to survive the failure of any individual link, individual switch, or entire switch tray by redirecting traffic across parallel network planes[20][21] - Advanced network features include compressed headers for optimizing small message delivery, link layer retry for error handling without transport layer loss, and a fully unified packet buffer to absorb traffic bursts[76][79][81]

Industry Trends - Enterprise computing diversifies across general-purpose, artificial intelligence, and emerging agentic workloads, rendering single central processing unit designs insufficient [1] - The industry experiences evolving requirements for differentiated central processing unit profiles tailored to distinct workload classes [1] Product Strategy - Advanced Micro Devices introduces the Venice portfolio as a purpose-built hardware lineup aligned with diverse enterprise demands [1] - Original equipment manufacturers develop scalable and workload-optimized platforms mapped to specific processing requirements [1]

Market Trends and Industry Dynamics - Artificial intelligence is driving unprecedented demand for compute, shifting the industry focus from deployment to building enduring artificial intelligence capabilities and sovereign infrastructure [2][4][5] - The high-performance computing and artificial intelligence industries are experiencing a major transition from traditional modeling and simulation to artificial intelligence factories and science applications [38][41] - The European Union has established **19** artificial intelligence factories to provide startups and industries with centralized access to compute, data, models, and expertise [42] Investment Opportunities and Potential Risks - France initiated a national artificial intelligence strategy backed by **2.5** billion euro to deploy foundational high-performance computing and artificial intelligence systems [39] - Industry enterprises face escalating financial constraints and supply chain challenges, particularly regarding the high cost and scarcity of high-bandwidth memory [65][88] - China's emergence as a competitive force with isolated high-performance computing systems highlights global market dependencies on United States chips and the necessity for strategic international collaboration [31][33] Technological Innovation and Strategic Infrastructure - Heterogeneous architectures integrating central processing units, graphics processing units, quantum computing, and diverse accelerators will dominate future infrastructure developments [3][58][102] - High-performance computing procurements are increasingly shifting from benchmark-driven evaluations like LINPACK toward outcome-focused frameworks and dedicated investments in application experts and centers of excellence [63][68][72] - Industry stakeholders project that fault-tolerant quantum computing and agentic artificial intelligence will become scientifically relevant and radically accelerate research productivity within the next **2** to **3** years [106][113]

Enterprise agentic AI is built for real work—coordinating workflows, calling tools, applying guardrails, enforcing policy, and maintaining compliance at scale. This session shows why CPUs are critical to agentic AI, serving multiple distinct roles to manage and execute the entire workflow as well as showcase how AMD EPYC CPUs are the ideal fit to deliver on power, performance and cost at any scale. Discover more: https://www.amd.com/en/corporate/events/advancing-ai/sessions-catalog/building-agentic-ai-with- ...

Market Trends & Industry Dynamics - Enterprises are facing escalating token costs, scaling to 500 to 2,000 USD per employee, with large companies like AMD projecting 400 million USD annually and financial services firms reaching 1 billion USD per year [4][5][42] - Approximately 80% to 90% of enterprises currently rely on cloud-based frontier model APIs for AI inference, creating challenges in cost predictability and budget control [9] - Industry data centers predominantly feature air cooling and under 30 kilowatts rack density, while high-performance liquid-cooled racks supporting 60 to 100 kilowatts or higher are primarily limited to Fortune 50 enterprises[15][16] Investment Opportunities & Technological Solutions - AMD introduced the MI350P PCIe GPU card supporting models up to 260 billion parameters, optimized for air-cooled data centers with 144 gigabytes HBM3E memory and production status starting this week [17][33][35] - AMD, Rackspace, and Unifor formed a strategic partnership to deliver an end-to-end enterprise AI stack, integrating hardware, global infrastructure, and autonomous fine-tuning software [14][22][54] - Enterprises deploying intelligent token routers and local models achieved 56% token offloading to on-premise MI350P infrastructure, resulting in 43% cost savings and 2.9 times faster response times for latency-sensitive security workloads [41][42][44] Company Financial Performance & Operational Impact - Unifor utilized a 31 billion parameter small language model running on two MI350P GPUs to process 65,000 vendor contracts and 2.5 million invoices for an energy company, delivering nearly 1 billion USD in savings and reducing token consumption by 85% [67][68][83] - Optimized enterprise AI solutions achieved 98% to 99% accuracy compared to frontier large language models, delivered 1.4 to 1.5 times faster time-to-first-token, and reduced total cost of ownership by 85% [82][83] - Rackspace will deploy AMD MI350P GPU clusters across global regions including the US, EMEA, and APAC, with full deployment anticipated in early to Q4 2026 [53][85]