The race to build infrastructure for agentic AI is extending below GPUs, CPUs and networking into a less visible but essential layer: firmware. Insyde Software says its InsydeH2O UEFI firmware has been qualified on Arm AGI CPU reference platforms, while its Supervyse OpenBMC firmware is ready for Arm AGI CPU-based AI server racks, giving system builders a production-oriented foundation for bringing next-generation AI servers online.

The infrastructure behind agentic AI is often described in terms of accelerators, processors and high-speed networking. But before an AI server can coordinate any of those components, it needs firmware capable of initializing hardware, enforcing security policies and exposing the system to higher-level software.

That makes firmware an increasingly strategic layer as data-center architectures become more heterogeneous.

At the OCP APAC Summit in Taipei, Insyde Software announced that its InsydeH2O UEFI firmware has been qualified on Arm AGI CPU reference platforms. The company has also prepared production-verified Supervyse OpenBMC firmware for AI server racks based on Arm AGI CPUs.

The announcement expands the firmware ecosystem around Arm’s AGI CPU platform and targets original design manufacturers, original equipment manufacturers, system integrators and cloud service providers building AI infrastructure.

The underlying proposition is relatively simple: system manufacturers can begin platform development with firmware that has already been validated for the target architecture instead of building the entire firmware foundation from scratch.

That can become important as AI server development shifts toward increasingly customized rack-scale systems.

Firmware becomes part of the AI infrastructure stack

UEFI and BMC firmware are not normally the technologies associated with generative AI. Yet they sit at critical points in the server architecture.

UEFI, or Unified Extensible Firmware Interface, initializes hardware during system boot and provides the interface between platform firmware and the operating environment. It can also participate in secure-boot and platform-security mechanisms.

OpenBMC, meanwhile, provides the management software layer for the baseboard management controller, allowing administrators and data-center operators to monitor and manage servers independently of the host operating system.

For conventional enterprise servers, those functions are already essential. AI infrastructure makes them more complicated.

A modern AI server may combine CPUs, GPUs or other accelerators, high-capacity memory, high-speed networking, storage and power-management systems. At rack scale, the number of components that must be initialized, monitored and managed increases considerably.

Agentic AI adds another dimension because workloads can be persistent, dynamic and highly distributed. An infrastructure platform must coordinate resources reliably rather than simply provide peak compute performance.

Insyde’s positioning is that firmware should be ready before those systems reach production.

Arm is building an alternative path into AI servers

The collaboration also highlights the growing importance of Arm-based CPUs in data-center infrastructure.

For years, x86 processors from Intel and AMD dominated enterprise servers. Arm has increasingly challenged that model as hyperscalers and infrastructure companies look for alternatives that can deliver power efficiency, customization and tighter integration with specialized workloads.

Arm’s Neoverse ecosystem has already established a presence in cloud infrastructure, while the company’s newer AI-focused initiatives are aimed at connecting Arm CPUs with accelerators and other components used in modern AI systems.

The Arm AGI CPU reference platform is therefore part of a larger movement toward heterogeneous computing.

Rather than relying on one processor architecture to perform every task, AI servers increasingly divide workloads across CPUs, accelerators, memory and networking devices. CPUs remain important for orchestration, data preparation and general-purpose workloads even when accelerators perform the bulk of AI computation.

That creates a requirement for firmware ecosystems that understand increasingly complex platform configurations.

The OCP connection matters

Insyde’s announcement at the Open Compute Project (OCP) APAC 2026 event is also significant because OCP has become an important forum for open data-center hardware standards and modular infrastructure design.

The industry’s move toward rack-scale AI systems is encouraging greater standardization around components that historically varied considerably between vendors.

Open approaches can reduce development friction for system manufacturers, but they also create an ecosystem challenge: hardware standards are useful only when software and firmware can support them reliably.

Insyde’s support for both UEFI and OpenBMC addresses two different parts of that equation.

The company says its InsydeH2O firmware supports the full Arm AGI CPU reference board and provides UEFI compliance, platform security and manageability capabilities tuned for high-density AI computing workloads.

It also says its firmware has been integrated with the Arm Fixed Virtual Platform, giving customers an environment for early evaluation before physical hardware is available.

That kind of early software access can shorten development cycles. System teams can begin testing firmware, boot flows and platform configurations before final silicon or complete server hardware is deployed.

Security and manageability are becoming AI infrastructure requirements

AI data centers introduce enormous amounts of compute, but scale also magnifies operational risk.

A security vulnerability in firmware can affect a system before the operating system or AI software even starts. At the same time, large fleets of servers cannot realistically be managed through manual intervention.

This is why BMC and firmware capabilities increasingly intersect with infrastructure security and operational efficiency.

For enterprise and cloud operators, the question is not simply whether an Arm-based server can run an AI model. It is whether thousands of those servers can be securely provisioned, monitored, updated and recovered at scale.

Insyde’s Supervyse OpenBMC positioning targets that operational problem.

The broader competitive environment includes firmware and platform-management ecosystems from companies such as AMI, Intel, Microsoft and HPE, alongside open-source projects and vendor-specific BMC implementations. The differentiator for firmware providers is increasingly their ability to support new processor architectures and rapidly evolving AI server designs without becoming a bottleneck in the hardware development process.

That is where Insyde’s Arm AGI qualification could become commercially relevant.

The hidden race beneath AI compute

The AI infrastructure market is becoming a stack of interdependent technologies. GPUs and AI accelerators attract most of the attention, but CPUs, memory, networking, power systems, firmware and management software all determine whether those components can operate reliably as a system.

Insyde’s announcement represents a small but important piece of that larger transition.

If agentic AI drives demand for continuously operating, high-density compute infrastructure, system builders will need faster paths from processor reference designs to production-ready servers. Firmware that is already qualified for those platforms can remove one layer of development risk.

The strategic implication is clear: the race to scale agentic AI is also a race to industrialize the infrastructure underneath it.

Market Landscape

The AI server stack is evolving toward increasingly heterogeneous, rack-scale architectures.

Key layers include:

• AI accelerators: GPUs and specialized processors handle computationally intensive model workloads.
• Data-center CPUs: Arm, AMD and Intel processors manage general-purpose computation and system orchestration.
• High-bandwidth memory: Provides the fast memory required by large AI models.
• Networking: High-speed interconnects connect accelerators, CPUs and storage across servers and racks.
• UEFI firmware: Initializes hardware and establishes the platform environment before the operating system loads.
• BMC/OpenBMC: Enables out-of-band monitoring, provisioning, diagnostics and server management.
• Cloud-native software: Orchestrates workloads across increasingly distributed infrastructure.

The emerging competitive advantage is therefore moving from individual components toward full-stack interoperability.

For ODMs and OEMs, firmware readiness can directly affect time to market. For cloud service providers, standardized management and security capabilities become increasingly important as AI clusters scale.

Top Insights

• Insyde’s UEFI qualification on Arm AGI CPU platforms gives AI server manufacturers a validated firmware foundation for next-generation agentic computing systems.
• Supervyse OpenBMC adds a management layer for Arm-based AI racks, addressing monitoring, provisioning and operational control across high-density infrastructure.
• Arm’s expanding data-center ecosystem reflects the industry’s move toward heterogeneous architectures combining CPUs, accelerators, memory and networking.
• Firmware qualification could become a competitive infrastructure advantage as AI server manufacturers seek faster transitions from reference silicon to production deployments.

Power Tomorrow’s Intelligence — Build It with TechEdgeAI