Microsoft unveils game changing AI data centre technology

Cobalt 200 exemplifies Microsoft’s vision of the converged AI data centre, where general-purpose CPUs and specialised accelerators work side by side with bespoke networking, storage, and security offload to train, fine-tune and deploy cloud and AI workloads at a massive scale.

Arm and Microsoft share a vision for a more efficient, scalable, and sustainable cloud – one where every layer of the stack, from silicon to software, is optimised not just for token generation but for complete end-to-end intelligence per watt.

The launch of Cobalt 200 marks another milestone in this collaboration, demonstrating what’s possible when world-class computing architecture meets world-class cloud engineering.

“The shift to purpose-built systems for AI and cloud-native workloads is transforming how infrastructure is designed,” said Dermot O’Driscoll, VP Products and Solutions, Cloud AI Business Unit, Arm. “Our collaboration with Microsoft on Cobalt 200 demonstrates how Arm Neoverse CSS V3 can be leveraged to accelerate this transformation and enable highly optimised silicon for a wide range of data-driven applications.”

Selim Bilgin, Corporate Vice President of Silicon Engineering, Microsoft, adds: “Cobalt 200 is the result of our longstanding co-innovation efforts between Microsoft and Arm, demonstrating what is possible when optimising silicon, server, and software together.

“We’re excited to be the first implementers of the Neoverse CSS V3 and taking advantage of Arm’s extensible architecture to implement unique capabilities like per-core DVFS, custom compression and cryptographic accelerators, as well as Azure Boost. Together, we are delivering compute infrastructure that improves performance and power efficiency for real cloud workloads.”

Cobalt 200 follows Arm's successful collaboration with Microsoft on its Cobalt 100 processor, which is delivering Neoverse-powered instances that are outpacing traditional competitors on performance and efficiency across various real-world workloads, including database, networking and quantitative finance.

Neoverse CSS V3 is a pre-integrated and silicon-proven platform that enables silicon designers – from hyperscalers to Arm Total Design partners including ASIC design houses, silicon startups or established merchant silicon providers – to design and deploy custom silicon faster than ever before.

It is built around Arm’s most performant Neoverse CPU core to date and includes support for Armv9 architectural features such as Confidential Computing Architecture (CCA) and Scalable Vector Extension 2 (SVE2).

This approach allows partners to focus on innovation where it matters most – optimisations in silicon and at the system level specific to their workload profile – while Arm continues to deliver the world’s leading platform for cloud-to-edge compute, backed by a cohesive cloud-to-edge AI-enabled software ecosystem.

Microsoft did exactly this by modelling more than 350,000 configuration parameters based on extensive workload profiling to optimise performance for both customer needs and internal applications like Microsoft Teams.

Arm says that as AI continues to permeate the data centre conversation, compute must optimise for every workload – certainly AI training and inference, but also the critical web, storage and data serving and processing workloads that form the backbone of modern commerce and are critical in the AI era to translate tokens into intelligence.