Microsoft’s proprietary Cobalt chips will be integrated into Azure starting next week

Microsoft will introduce its customized Cobalt 100 chips to customers as a public preview during its upcoming Build conference.

During an analyst briefing prior to Build, Scott Guthrie, the executive VP of Microsoft’s Cloud and AI group, made a direct comparison between Cobalt and AWS’s Graviton chips, which have been accessible to developers for a considerable period of time. Guthrie stated that Microsoft’s chips will provide a 40% improvement in performance compared to other ARM chips available on the market. Adobe, Snowflake, and other companies have already begun utilizing the new chips.

Microsoft initially unveiled its cobalt chips in November of last year. These chips are built on the Arm architecture and have a 64-bit design with 128 cores.

Furthermore, Microsoft will offer Azure clients access to AMD’s MI300X accelerators alongside the Cobalt chips starting next week. Although AMD is a prominent producer of GPUs, it has historically lagged behind Nvidia in the field of artificial intelligence (AI). However, due to the high cost of Nvidia’s chips, major cloud providers are now seeking alternatives, and AMD has made progress in this area by providing improved software support. As a result, these new chips from AMD have become highly sought-after.

Guthrie characterized it as the most economically efficient GPU currently available for Azure OpenAI.

One of the pieces of news we discovered is that Microsoft will reduce the cost of accessing and operating large language models at the upcoming Build event. The precise appearance of that remains ambiguous, however.

Microsoft will also introduce a new “real-time intelligence system” that enables the streaming of real-time data into Fabric, Microsoft’s data analytics system, during the preview phase. This system will provide seamless integration with Kafka, as well as support for AWS Kinesis and Google Cloud’s Pub/Sub data-streaming systems.

Additionally, Microsoft will unveil a collaboration with Snowflake. Fabric will now incorporate Snowflake’s Iceberg format alongside Databricks’ Parquet, facilitating seamless compatibility with Snowflake. This integration allows for bidirectional data flow between Snowflake and Fabric.

Microsoft intends to introduce a new functionality for Copilot enthusiasts, enabling developers to directly oversee their Azure resources through Copilot by utilizing natural language. “This will allow for a more efficient developer process with the use of natural language throughout your development stack and Azure,” Guthrie stated. The system is constructed using a widely used extensibility mechanism, allowing other providers to easily integrate with it and provide similar functionalities.