Amidst the surge in AI development, the battle for hardware infrastructure standards is intensifying. To prevent ecosystem fragmentation, the Open Compute Project (OCP) recently appointed Arm to its board of directors, placing it on equal footing with tech giants like AMD, NVIDIA, Meta, and Google to jointly lead the development of specifications for next-generation AI datacenters.
This move not only establishes Arm's influence in the high-performance computing field, but also means that "open standards" will become a core strategy for meeting the challenges of converged AI data centers.
AI racks consume as much power as hundreds of homes; open collaboration is the only solution
Mohamed Awad, general manager of Arm's Infrastructure Business Group, highlighted the current predicament: It is estimated that by 2025, the computing power of a single AI rack will rival that of a top-tier supercomputer in 2020, while its staggering power consumption will be equivalent to the combined electricity consumption of 100 American households. This exponential growth is beyond the capacity of traditional general-purpose server architectures.
To this end, Arm has proposed the vision of a "converged AI datacenter," aiming to maximize AI computing density per unit area while simultaneously reducing overall power consumption and cost. Achieving this goal requires collaborative design across computing, acceleration, memory, and networking, which is where OCP open standards play a key role.
Contributing to the "neutral" chiplet architecture and creating the "USB" standard in the hardware industry
In order to fundamentally promote hardware innovation, Arm will alsoBasic chiplet system architecture (FCSA) specifications contributed to OCP.
Think of FCSA as the USB standard for the hardware world. It defines a unified framework that is not tied to a specific CPU architecture, allowing small chips from different suppliers to be "stitched" together through advanced packaging technology and communicate smoothly.
This contribution stems from the "Arm Total Design" ecosystem launched by Arm in 2023. The scale of the ecosystem has tripled in one year, and 10 key partners have been added, including ASE, Astera Labs, Marvell, and Alchip, covering the entire supply chain from packaging, interconnection to system integration.
The Arm ecosystem is taking shape, aiming to derive more than half of data center computing power from the Arm architecture.
Arm's ambitions are well-founded. With its Neoverse platform becoming the core of every layer of the AI technology stack, it is estimated that by 2025, half of the computing power shipped to hyperscale data centers will be based on the Arm architecture. Its high performance and low power consumption will be key to alleviating the surge in AI power consumption and achieving sustainable operations.
From micro-sensors to giant data centers, Arm emphasizes the power of its ubiquitous computing platforms, charting a course for open, efficient, and sustainable AI infrastructure. Joining the OCP board is just the first step in this hardware standards revolution.
