On November 28, news emerged that according to the latest information disclosed by Arm, the recently released Cobalt 200 processor by Microsoft is the world’s first processor built on Arm Neoverse CSS V3, designed for cloud and AI infrastructure.
In the era of artificial intelligence (AI), the industry has undergone a significant transformation from generic off-the-shelf systems to customized infrastructure. From traditional network services to scalable data analytics, and large-scale model inference, various workloads have now been integrated into AI-driven intelligent processing chains. The architectural design of modern data centers is no longer just a pile of independent computing resources, but needs to be constructed into a unified system capable of efficiently handling diverse, high-load tasks. To meet this demand, the industry is comprehensively optimizing computing technology from the ground up, prioritizing performance, scalability, and energy efficiency.
Azure Cobalt 200: Accelerating the Development of Integrated AI Data Centers
On November 19 of this year, Microsoft announced the launch of the new generation server processor Azure Cobalt 200, which is the next-generation Arm architecture-based CPU designed specifically for cloud-native workloads, manufactured using TSMC’s 3nm process.
According to reports, Cobalt 200 is an important milestone in Microsoft’s strategy to continuously optimize every layer of the cloud stack (from chips to software). Our design goals are: full compatibility with workloads using existing Azure Cobalt CPUs; a performance improvement of up to 50% compared to Cobalt 100; and integration with the latest Microsoft security, networking, and storage technologies.
Like its predecessor, Cobalt 200 is optimized for common customer workloads and provides unique features for Microsoft cloud products. The first batch of Cobalt 200 servers is now online in its data centers, with broader deployment and customer availability expected in 2026.
According to the latest information released by Arm, Microsoft Cobalt 200 is the first publicly released chip to adopt the Arm Neoverse Computing Subsystem V3 (CSS V3), which is Arm’s latest performance-optimized core and architecture. Cobalt 200 fully embodies Microsoft’s vision for integrated AI data centers: general-purpose CPUs closely collaborating with dedicated accelerators, equipped with customized networking, storage, and security offload technologies to achieve large-scale training, fine-tuning, and deployment of cloud and AI workloads.
Each Cobalt 200 SoC contains 132 cores, with each core equipped with 3MB of L2 cache and 192MB of L3 system cache, providing exceptional performance for customer workloads.
A unique innovation of Cobalt 200 is the dynamic voltage and frequency scaling (DVFS) for each core. In Cobalt 200, this technology allows each of the 132 cores to operate at different performance levels, thereby achieving optimal power consumption regardless of the workload.
Building the Future of AI Infrastructure Together
Arm and Microsoft share a common vision to create a more efficient, scalable, and sustainable cloud environment—from the chip to every layer of the software stack, optimizing not only for token generation but also maximizing end-to-end intelligence per watt of energy. The release of Cobalt 200 marks another important milestone in the collaboration between the two companies, showcasing the innovative results that can be achieved by combining world-class computing architecture with top-notch cloud engineering technology.
Dermot O’Driscoll, Vice President of Product Solutions for Arm’s Cloud AI Division, stated: “The trend of building customized systems for AI and cloud-native workloads is reshaping the design paradigm of infrastructure. Our collaboration with Microsoft on Cobalt 200 fully demonstrates the core value of Arm Neoverse CSS V3, which can accelerate this innovation process and create highly optimized chip solutions for various data-driven applications.”
Selim Bilgin, Vice President of Chip Engineering at Microsoft, stated: “Cobalt 200 is the result of long-term joint innovation between Microsoft and Arm, fully demonstrating the excellent results that can be achieved through the collaborative optimization of chips, servers, and software. We are pleased to be the first company to implement Neoverse CSS V3 and, leveraging Arm’s scalable architecture, have achieved unique features such as per-core dynamic voltage frequency adjustment (DVFS), customized compression and encryption accelerators, and Azure Boost. The computing infrastructure we built together will significantly enhance the performance and energy efficiency of actual cloud workloads.”
Cobalt 200 is another achievement following the successful collaboration on the Cobalt 100 processor. The Neoverse-based Cobalt 100 instances have performed excellently in various real-world application scenarios, such as databases, networking, and quantitative finance, surpassing traditional competitors in both performance and energy efficiency.
Neoverse CSS Accelerates Custom Chip Innovation
Neoverse CSS V3 is a pre-integrated and silicon-validated platform that helps chip designers—from hyperscale cloud service providers to Arm Total Design partners, including ASIC design companies, chip startups, and established commercial chip suppliers—complete the design and mass production deployment of custom chips at unprecedented speed. This platform is built on Arm’s most powerful Neoverse CPU core to date, supporting Armv9 architecture features such as Confidential Compute Architecture (CCA) and second-generation Scalable Vector Extensions (SVE2).
This model allows partners to focus on the most critical areas of innovation—precisely optimizing at the chip and system level for their specific workloads. Meanwhile, Arm continues to provide a world-leading computing platform from cloud to edge, supported by a highly efficient AI-enabled software ecosystem from cloud to edge, providing strong support for ecosystem partners. Microsoft has modeled over 350,000 configuration parameters based on extensive workload analysis to optimize performance for customer needs and internal applications such as Microsoft Teams.
Building the Future Cloud AI Data Center
As AI continues to penetrate the core issues of data centers, computing architecture must be comprehensively optimized for various workloads—not only covering AI training and inference but also supporting critical tasks that drive modern business operations, such as web services, storage, and data services and processing. In the AI era, these workloads are the core support for transforming tokens into intelligence. The integration of general computing and AI-optimized computing at a macro level is becoming the cornerstone of the next generation of cloud infrastructure. With the Cobalt 200 processor based on Neoverse CSS V3, Arm and Microsoft are becoming the core force in this computing revolution, providing a more superior performance, outstanding energy efficiency, and greater flexibility for future data centers optimized for intelligence.
Editor: Chip Intelligence – Lang Ke Jian
Previous Exciting ArticlesGuokai Microelectronics Terminates Acquisition of SMIC Ningbo!Not Afraid of Google TPU Competition! Huang Renxun: GPUs Have General Advantages, NVIDIA’s Position is Solid!DRAM Price Surge Just Beginning, Expected to Rise Another 58% Next Year?Intel CEO Chen Liwu: Fully Supports Luo Weiren, TSMC’s Accusations are Groundless!China’s Ministry of Commerce and the EU Reach Consensus: Urging Nexperia to Conduct Internal Communication!HP Announces Global Layoffs of 4,000 to 6,000 People!Once a Huawei “Genius Boy”, 90s “Zhi Hui Jun” Becomes Chairman of Shangwei New Materials!TSMC Sues Luo Weiren: Stealing 2nm Secrets to Jump to Intel!ADATA Chairman Chen Libai: Storage Shortage, Can’t Buy Even with Money!Memory Prices Surge, NVIDIA and AMD Can’t Stand It Either!Yangtze Memory Temporarily Shuts Down Production Line? Fake News!Investing 500 Million! Yidao Information Partners with Huafeng Technology to Enter Advanced Packaging!Changxin Memory Showcases DDR5 and LPDDR5X New Products for the First TimeWintek Technology Has Filed an Appeal with the Dutch Government!Huang Renxun: AI Chip Revenue in the Chinese Market is Zero!Demand is Three Times Supply, Samsung Plans to Shift Some NAND Production Lines to DRAMResults of the Academician Election Announced! BYD’s Lian Yubo, CATL’s Wu Kai, and others ElectedBan on Sales Crisis Resolved, BOE and Samsung Display Reach Settlement!
For industry communication and cooperation, please add WeChat: icsmart01Chip Intelligence Official Communication Group: 221807116