OCP 2026: AMD CEO Lisa Su Highlights Open Infrastructure for the Future of AI
At the 2026 Open Compute Project (OCP) Global Summit in San Jose, California, Dr. Lisa Su, Chair and CEO of AMD, highlighted the importance of open infrastructure in supporting the rapid growth of artificial intelligence (AI). During her opening keynote, Su demonstrated how AMD’s portfolio of computing technologies, software, and integrated systems is designed to deliver the performance and flexibility required for agentic AI, from enterprise deployments to gigawatt-scale AI facilities.
As organizations increasingly adopt AI to automate tasks, analyze information, and support complex decision-making, demand for computing infrastructure continues to rise. AMD is working with industry partners and technology companies to advance open standards and develop systems that can accommodate evolving AI workloads. Through its broad portfolio of processors, graphics processing units (GPUs), networking technologies, and software, the company aims to help customers build scalable infrastructure for the next generation of AI applications.
The four-day OCP Global Summit provides a platform for AMD, its partners, and customers to showcase new computing platforms, technical demonstrations, and practical applications. The event also highlights the industry’s efforts to establish common hardware and software standards that make AI infrastructure easier to develop, deploy, and expand.
The keynote followed AMD’s earlier announcement of the AMD Helion agentic AI rack, an open rack design based on an Open Compute Project specification and powered by AMD computing technologies. Together, these developments demonstrate AMD’s continued commitment to open infrastructure as AI workloads become more sophisticated and resource-intensive.
Open Infrastructure Becomes Essential for Agentic AI
A central theme of Su’s keynote was the growing importance of flexible, full-stack infrastructure for supporting agentic AI. Unlike conventional AI applications that primarily respond to individual prompts, agentic AI systems can retain context, use external tools, perform multiple steps, and repeatedly request responses from AI models to complete complex tasks.
These capabilities create additional demands on computing infrastructure. An AI agent may need to retrieve information, analyze results, interact with software applications, and generate further responses before completing a task. Each step can consume computing resources, increasing requirements for processors, memory, networking, and efficient software execution.
Inference, the process of running trained AI models to generate responses or perform tasks, has become a major component of AI computing demand. As more businesses integrate AI into everyday operations, the need for infrastructure capable of handling large volumes of inference workloads is expected to grow.
AMD’s strategy focuses on providing customers with systems that can support these changing requirements across different deployment sizes. Enterprise customers may need infrastructure for specialized AI applications, while cloud providers and large technology companies require significantly larger systems capable of supporting extensive AI services.
Open standards are important to this approach because they provide a shared foundation for hardware design, system integration, and infrastructure management. Rather than depending entirely on proprietary architectures, customers and partners can work with standardized designs that support greater flexibility and interoperability.
AMD Helios Advances Integrated AI Rack Infrastructure
One of the key developments highlighted in the broader OCP discussion is AMD Helios, the company’s rack-scale AI infrastructure platform. Following its debut at the 2025 OCP Global Summit, Helios is progressing toward production deployment, with customer deliveries, infrastructure installations, and workload validation activities underway, according to AMD’s announcement.
AMD Helios integrates several technologies into a unified rack architecture, including AMD Instinct GPUs, AMD EPYC processors, AMD Pensando networking technologies, and the AMD ROCm software ecosystem.
This integrated design reflects the growing need to coordinate computing, memory, networking, and software rather than treating them as separate infrastructure components. Large AI models require substantial processing power, sufficient memory to hold model parameters and working data, and high-speed communication between computing resources.
When these components are not properly balanced, performance bottlenecks can limit the efficiency of an entire AI system. A rack-scale architecture is intended to address these challenges by bringing the major infrastructure elements together within a coordinated design.
AMD Helios also emphasizes open standards, allowing infrastructure developers and customers to build systems around a common framework. This can help simplify integration, improve deployment consistency, and provide a foundation for expanding capacity as demand increases.
For businesses investing in large AI clusters, these characteristics are particularly important. Infrastructure must not only deliver high performance but also remain manageable, serviceable, and reliable throughout its operational life.
The Role of AMD Instinct GPUs and EPYC Processors
AMD’s computing portfolio combines specialized accelerators with general-purpose processors to support different stages of AI workloads.
AMD Instinct GPUs are designed to accelerate computationally intensive tasks associated with AI training and inference. These processors help execute the large numbers of mathematical operations required by modern AI models.
AMD EPYC processors provide general-purpose computing capabilities for system operations, data processing, orchestration, and other workloads that support AI applications. Their role is important in infrastructure environments where accelerator performance must be coordinated with the rest of the computing system.
AMD Pensando networking technologies complement these processors and accelerators by supporting communication across the infrastructure. As AI deployments grow, efficient data movement between processors, accelerators, memory, and other systems becomes increasingly important.
Integrating these technologies into a unified rack design can help customers plan infrastructure more consistently and manage the interactions between major system components.
However, performance depends on the complete system, including workload characteristics, software optimization, network configuration, memory capacity, and deployment conditions. Coordinated system design is therefore a central part of AMD’s approach to large-scale AI infrastructure.
ROCm Software Supports Performance and Flexibility
Software is another essential element of It’s open infrastructure strategy. The AMD ROCm software ecosystem provides tools and software components designed to support GPU computing and accelerate AI and high-performance computing applications.
Hardware capabilities alone do not determine the performance of an AI system. Software frameworks, optimized libraries, development tools, and workload-specific improvements also influence how efficiently processors and accelerators execute applications.
It continues to develop ROCm to improve performance, support AI development frameworks, and help developers use its computing hardware more effectively. Software updates can enable performance improvements and new capabilities on existing hardware, potentially allowing customers to gain additional value from infrastructure investments without replacing their entire systems.
For organizations operating large AI clusters, software compatibility and optimization are particularly important because even modest efficiency improvements can have meaningful effects across thousands of computing devices.
By combining hardware development with software improvements, AMD aims to provide a more complete infrastructure platform for AI training, inference, and agentic applications.
Why Open Standards Matter for the AI Industry
The expansion of AI is creating demand for infrastructure that can scale quickly while accommodating changing technical requirements. Meeting this demand requires coordinated advances in processors, accelerators, networking, cooling, power distribution, software, and rack architecture.
Open standards help establish common design principles that manufacturers, system builders, and infrastructure operators can adopt. They can encourage interoperability, broaden supplier participation, and give customers greater flexibility when selecting and integrating equipment.
It’s participation in the OCP ecosystem reflects its effort to contribute to this shared infrastructure foundation. Its work with industry partners also supports the development of open rack designs that can be adapted for different deployment requirements.
The AMD Helion agentic AI rack is an example of this strategy, combining AMD computing technologies with an open design approach intended to support large-scale AI infrastructure.
As agentic AI applications become more widely used, customers will need systems capable of managing increasingly complex workloads while balancing performance, cost, energy consumption, and operational reliability.
AMD’s Vision for the Next Generation of AI Infrastructure
Lisa Su’s keynote at the 2026 OCP Global Summit underscored It’s focus on open, integrated computing infrastructure as a foundation for future AI growth.
Through It Helios, the Helion rack initiative, its Instinct GPUs, EPYC processors, Pensando networking technologies, and ROCm software, the company is working to provide solutions for AI workloads across enterprise systems and large-scale deployments.
The broader objective is to help customers build infrastructure that can adapt as AI applications evolve. By supporting open standards and collaborating with ecosystem partners, It aims to make advanced computing systems more scalable, flexible, and accessible to organizations with different infrastructure requirements.
As AI moves toward more autonomous, multi-step applications, the ability to coordinate computing, networking, memory, and software will become increasingly important. It’s continued participation in the OCP ecosystem highlights the role that open collaboration can play in building the infrastructure needed to support the next phase of artificial intelligence.
Source Link: https://newsroom.amd.com/