AMD launches open infrastructure system for agentic AI

AMD launches Helion, an open infrastructure for agentic AI. The system offers flexible configurations with CPUs and GPUs in racks of up to 32,768 cores. Availability is expected for November 2026.

The lack of common standards creates scaling challenges in confidential computing. The IETF RATS is working to establish a common framework for attestations in confidential computing. AI Trust is strengthening cyber resilience in the digital environment. Oracle CISO Brennan Baybeck highlights the importance of humans in AI security governance. Baybeck mentions three common gaps in AI security governance: absence of basic governance, lack of fundamental security and unawareness of the shared responsibility model. The interview with Baybeck took place during the Oracle event in 2026.

AMD launches open infrastructure system for agentic AI — Image: newsroom.arm.com

Confidential computing is becoming a critical area in AI security, with interoperability and governance challenges. AMD, Oracle and other sectors are working to establish standards and infrastructure to ensure trust in AI systems. The need for common attestations and frameworks is intensifying with the diversity of the ecosystem.

The system's flexibility allows organizations to adjust the infrastructure according to their needs, without having to replace all the equipment. This reduces costs and increases efficiency, especially for AI systems that demand high scalability. Helion's architecture also facilitates integration with multiple parties and vendors, which can speed up implementation in complex environments.

But even with these advantages, the system still depends on common attestation standards, which are being developed by the IETF RATS. This may delay adoption in some markets, especially where interoperability is still a challenge. Helion is expected to reach the global market in November 2026, but there is still no information on availability in Brazil.

AMD works with standards for AI security

AMD launches Helion, an open infrastructure for AI agentic computing that redefines the scalability of confidential computing. AMD was already working with the Open Compute Project, which seeks to standardize specifications for server infrastructure. Helion can integrate into the AI security ecosystem, as revealed by siliconangle. The discussion on AI security governance led by Oracle CISO Brennan Baybeck is linked to the IETF RATS's work to establish common frameworks.

AMD reveals that agentic AI is not a single workload, but spans personal agents, enterprise applications, code execution and complex workflows, each with different computing requirements. AMD's Dan McNamara stresses that infrastructure must be optimized to meet these demands. According to HPE, deployment flexibility is essential for agentic AI in production, and Helion reinforces that versatility. Oracle's Brennan Baybeck warns that humans are now the most critical and potentially dangerous link when controlling these 'robots'.