Microsoft is expanding its cloud platform Azure in the second half of 2026 with AMD’s new Helios rack system. Each rack consolidates 72 Instinct-MI455X accelerators for 1.4 ExaFLOPS of computing power in FP8 format. For Microsoft, this is the largest step away from its previous dependence on Nvidia chips.
Helios Racks Consolidate Computing Power for Large AI Models
As Microsoft announced in an official blog post, the company is expanding its computing infrastructure offering with AMD’s Helios system. One rack accommodates 72 Instinct-MI455X graphics accelerators with a total of 31 terabytes of HBM4 memory and achieves, according to a technical analysis by Tom’s Hardware, 260 terabytes per second of bandwidth within a rack. Between the racks, it is said to be 43 terabytes per second via the UALink connection technology based on Ethernet.
The system is complemented by the sixth generation of AMD EPYC processors codenamed Venice, as well as networking technology from the brand Pensando. Azure will receive two new virtual machine series: HDv2 is aimed at agent-based AI applications and data pipelines, while HXv2 is for chip design calculations. For comparison: Nvidia’s competing system Vera Rubin NVL72 achieves a similar bandwidth within a rack, but only reaches about 21.5 terabytes per second between the racks – about half as much as Helios. For the operation of the GPUs, AMD also relies on the open software platform ROCm instead of Nvidia’s proprietary CUDA ecosystem. The companies did not disclose specific delivery dates for individual data centers.
AMD Positions Helios as a Response to Nvidia’s Market Leadership
AMD CEO Lisa Su described the Microsoft partnership as an important milestone for the next generation of AI infrastructure. Microsoft CEO Satya Nadella stated that the expansion of the Azure infrastructure with Helios provides customers with more power, scalability, and choice for building and operating future AI applications. So far, Nvidia has almost completely dominated the AI data center market; analysts view Helios as AMD’s most comprehensive response to this, as the system provides computing chips, networking technology, and software from a single source.
The announcement coincided with AMD’s in-house event Advancing AI 2026, where the company presented further details about the Helios architecture. Financial markets reacted immediately: According to Blockonomi, AMD’s stock rose by about five percent after the announcement, while Microsoft’s stock slightly declined. In addition to Microsoft, AMD claims to also count Meta, OpenAI, and Oracle among the customers of the Helios platform – independently unverified. Both companies did not disclose financial details of the agreement between Microsoft and AMD.
More Choices for Azure Customers in Compute-Intensive AI Applications
For companies operating AI models via Azure, the expansion initially means more choice in the underlying hardware rather than an immediate switch. So far, most large language models on Azure run on Nvidia chips; with Helios, a second architecture is being added that Microsoft plans to operate in parallel. This could strengthen the negotiating position of cloud customers against chip suppliers and cushion supply bottlenecks, as have repeatedly occurred in the AI market recently.
Whether customers in Germany or the EU will have access to Helios capacities at market launch remains open from Microsoft and AMD. Prices for the new VM series HDv2 and HXv2 were also not disclosed by both companies; usage-based rates are typical for Azure, which are usually published shortly before general availability.
Crucial will be how much computing capacity Microsoft actually purchases from AMD and whether Helios can keep up in practice with the Nvidia-dominated systems. Both companies did not provide any figures on this, so the success of the partnership will only become apparent with the first deliveries in the second half of 2026.
