Microsoft will use AMD’s AI-optimized Helios racks in Azure

siliconangle.com Jul 21, 2026

Microsoft will use AMD’s AI-optimized Helios racks in Azure

Curated by Shubham · TechCapsules · Jul 21, 2026

AI-summarised brief · reviewed before publication

Microsoft announced it will integrate Advanced Micro Devices’ Helios rack design into its Azure cloud services. This partnership introduces three new virtual machine instance families powered by AMD silicon. The Helios reference design features 72 Instinct MI455 GPUs per system, utilizing CDNA 5 architecture and 432 gigabytes of HBM4 memory. These graphics processing units are supported by Pensando data processing units and Epyc CPUs from the Venice series. The Venice processors leverage TSMC’s two-nanometer process and SoIC chiplet stacking technology. Helios racks employ liquid cooling and UALoE networking protocols to manage heat and data exchange efficiently. AMD CEO Lisa Su highlighted the extended collaboration across AMD’s full AI solution stack. Shipping begins later this year. Microsoft will deploy these systems for the ND MI455X v7 series, optimized for AI inference tasks like agents and search. Additionally, the HDv2 series supports CPU-intensive data preparation, while the HXv2 series targets electronic design automation workloads.

💡 Why It Matters

· The shift to liquid-cooled, two-nanometer racks marks a critical infrastructure evolution for handling dense AI inference loads.
· By offloading network tasks to specialized data processing units, Microsoft preserves valuable CPU capacity for customer applications.

Read full article on siliconangle.com

Microsoft adopts AMD Helios AI platform for Azure expansion

Microsoft is significantly expanding its Azure cloud infrastructure by adopting AMD’s next-generation Helios AI platform. This strategic move involves deploying AMD Helios Rackscale Solutions, which integrate Instinct MI455X GPUs, 6th Gen EPYC “Venice” CPUs, and Pensando networking technology. The deployment aims to support large-scale AI inference for Microsoft’s services and customer workloads. Additionally, Microsoft will introduce new Azure virtual machine families, specifically the HDv2 and HXv2 series, based on the latest EPYC processors. These additions target agentic AI, data pipelines, and semiconductor design applications. AMD plans to begin shipping Helios systems to customers, including Microsoft, in the second half of 2026. This collaboration strengthens the long-term partnership between the two tech giants, addressing growing demand for flexible, high-performance AI infrastructure. The expansion provides Azure customers with enhanced performance and choice for developing foundation models and running production AI applications through Azure Foundry Managed Compute.

💡 This shift toward integrated rack-scale solutions marks a departure from standalone accelerators, offering European developers a robust, unified environment for building complex AI applications. By embedding AMD hardware directly into Azure’s core infrastructure, Microsoft validates the viability of non-NVIDIA ecosystems for enterprise-grade AI workloads.

quantumcomputingreport.com · Jul 21

Joint Study by AWS, NVIDIA, LBNL, and NASA Establishes Framework for Quantum-HPC…

A joint research team from Amazon Web Services, NVIDIA, Lawrence Berkeley National Laboratory and NASA has released a quantitative model that determines when quantum processing units (QPUs) must be physically co‑located with high‑performance computing (HPC) supercomputers and when remote cloud access suffices. The paper, posted on arXiv and highlighted on the AWS Quantum Technologies Blog, separates quantum‑classical interaction into a sub‑microsecond real‑time control tier—requiring ultra‑low latency for tasks such as error‑correction decoding—and an application tier where latency only affects overall runtime. By decomposing execution cycles into compute and communication components, the authors introduce a communication‑to‑computation ratio (Rcc) that classifies workloads as compute‑bound (remote access viable) or communication‑bound (tight integration needed). Benchmarks on IBM’s 77‑qubit Heron processor and a 10‑qubit device illustrate the model’s guidance, while future fault‑tolerant scaling will increasingly demand low‑latency coupling.

💡 It gives hardware planners a concrete tool to allocate capital efficiently, preventing costly over‑engineering of quantum‑HPC links for workloads that can thrive on standard cloud infrastructure.

hospitalitynews.in · Jul 20

Keytop Tech | IT Infrastructure Experts

Keytop Tech Solutions, a Gurgaon-based IT infrastructure provider founded in 2021, has established itself as a trusted partner for enterprises, startups, and government bodies across India. The company offers end-to-end solutions, including new and certified pre-owned hardware such as servers, storage systems, networking devices, and workstations. Keytop ensures rapid deployment through same-day or next-day delivery via major logistics partners. Beyond hardware, its service portfolio encompasses cloud solutions, data backup, structured cabling, power backup, and endpoint security from leading vendors. To support flexible scaling, Keytop provides leasing options for critical IT assets. The firm emphasizes robust customer support through annual maintenance contracts, remote troubleshooting, and multi-vendor assistance for major OEMs like HP, Dell, Cisco, and IBM. By combining extensive inventory with comprehensive lifecycle management, Keytop aims to transform technology into tangible business opportunities, helping organizations maintain performance and scalability in a competitive digital economy while minimizing heavy capital investment through strategic leasing and reliable service delivery.

💡 Keytop’s hybrid model of certified pre-owned hardware and leasing options lowers the barrier to entry for small businesses seeking enterprise-grade infrastructure. This approach allows organizations to scale operations rapidly without the financial burden of heavy upfront capital expenditure.

itbrief.asia · Jul 20

NetApp buys DataPelago to boost AI data processing

NetApp has acquired DataPelago, an AI data infrastructure company, to integrate its data-processing technology into NetApp’s storage business. Based in California, DataPelago develops software that processes data at the storage layer rather than moving it to separate computing systems. NetApp states this approach can reduce infrastructure costs by up to 80% and deliver performance up to 10 times faster for AI and analytics workloads compared to conventional methods. The deal incorporates DataPelago’s Nucleus engine, which utilizes CPUs and GPUs to address the time and cost challenges of preparing data for enterprise AI projects. CEO George Kurian noted that as AI models advance, enterprises require equally intelligent data infrastructure. This acquisition aims to help customers harness their data assets more effectively by simplifying data preparation, governance, and access across hybrid cloud and on-premises environments, thereby addressing a critical bottleneck in AI deployment.

💡 The acquisition shifts the industry focus from merely acquiring powerful AI chips to optimizing the underlying data infrastructure that feeds them. By processing data where it resides, NetApp directly tackles the hidden costs and latency that often stall enterprise AI initiatives.

forbes.com · Jul 19

AI Storage & Memory From Backblaze, CoreWeave, Panmnesia, Vast And Cloudera

Backblaze and CoreWeave finalized a $335 million agreement, positioning Backblaze to supply cost-efficient hard disk drive storage for CoreWeave’s artificial intelligence infrastructure. This partnership optimizes data placement for massive AI workloads. Simultaneously, Panmnesia advanced memory technology with new CXL controller and switch chips designed to enable low-latency memory expansion across multiple servers. Meta’s implementation of this CXL technology reportedly reduced distributed AI inference server requirements by 25 percent and decreased cache response times by 29 percent. In another development, Cloudera and Vast Data announced a strategic partnership to deliver an “AI data platform anywhere.” This collaboration merges Cloudera’s containerized data services with Vast’s AI Operating System. The resulting scalable solution facilitates data ingestion, refinement, and delivery to AI models. It supports GPU-accelerated workloads across hybrid and multi-cloud environments, addressing the critical need for speed and scale in training, inference, and retrieval-augmented generation processes.

💡 These partnerships directly address the bottleneck of data movement and storage costs that currently limits AI scalability. By integrating specialized memory controllers with optimized cloud storage, enterprises can significantly reduce hardware overhead while accelerating model training and inference speeds.

thestar.com.my · Jul 18

Musk’s SpaceX in talks to supply the Pentagon with computing power, WSJ…

Elon Musk’s SpaceX is reportedly negotiating with the U.S. Department of Defense to provide billions of dollars in data center capacity for running artificial intelligence models. The Wall Street Journal cited sources familiar with the matter, indicating the talks aim to expand the Pentagon’s existing partnership with the aerospace firm beyond rocket launches and satellite communications. SpaceX employees have discussed competing directly with neocloud providers like CoreWeave by offering lower-priced computing power. The Defense Department seeks additional cloud capacity to support intelligence agencies and military AI applications. This follows similar recent agreements, including a multi-year deal with Google for access to Nvidia chips and a partnership with Anthropic for the Colossus 1 facility. Amazon also recently announced a $50 billion investment in government AI capacity. Discussions remain ongoing and could collapse. Neither SpaceX nor the Pentagon responded to comment requests, and Reuters could not independently verify the report.

💡 SpaceX’s pivot into high-performance computing challenges the dominance of traditional cloud giants in the lucrative government AI sector. By leveraging its infrastructure to undercut competitors like CoreWeave, the company is redefining its role from a space logistics provider to a critical national security technology partner.