At the 2025 OCP Global Summit, Mohamed Awad, senior vice president and general manager of Arm Infrastructure Business, delivered a keynote speech.callThe data center industry is rethinking system architecture in the AI era and has announced the launch of the open chiplet collaboration project "Foundation Chiplet System Architecture" (FCSA), hoping to promote AI infrastructure innovation through modular design.
New Thinking on Infrastructure in the AI Era
In his speech, Mohamed Awad pointed out that the traditional one-size-fits-all design approach can no longer meet the diverse demands of today's AI workloads. As AI applications shift from training to large-scale inference, infrastructure requires more elastic, efficient, and scalable solutions.
To this end, Arm has joined forces with its ecosystem partners to launch an open chiplet architecture. Through standardized interfaces and modular design, it allows companies to freely combine different functional units like assembling building blocks, and optimize the configuration for specific workloads.
Advantages and Applications of Small Chip Technology
This modular design allows customers to mix and match chip units with different process nodes and functional characteristics, significantly increasing design flexibility while effectively controlling cost and power consumption. For example, a high-performance computing core and a dedicated AI accelerator can be integrated into a single package to achieve the best performance-to-power ratio.
Mohamed Awad emphasized, "The future of AI infrastructure lies not in the ultimate performance of a single chip, but in leveraging the strengths of diverse technologies through an open ecosystem and modular design."
Promote industrial cooperation and standardization
Arm is collaborating with multiple chip design, manufacturing, and system integration partners to establish chiplet interconnect standards and verification systems, aiming to lower technical barriers and accelerate innovation cycles. This open collaboration model is expected to create new market opportunities for small and medium-sized chip design companies.
The author's opinion
Arm's push for an open chiplet ecosystem holds far-reaching strategic significance. Amidst the exploding demand for AI computing power, it's becoming increasingly difficult for a single vendor to fully master all technical aspects. By establishing open standards, Arm is attempting to replicate its successful model in the mobile computing market in the hardware sector.
This modular design approach is particularly well-suited to the diverse nature of current AI applications. From cloud-based training to edge inference, different scenarios have varying requirements for computing power, power consumption, and cost, and the chiplet architecture provides tailored solutions.
With major companies like Intel and AMD also deploying chiplet technology, the battle for dominance in AI infrastructure architecture has begun. Whether Arm can leverage its ecosystem advantages to win this competition will depend on whether its open strategy can truly create value for its partners.
