News from manufacturers

How XPUs Meet a World-Class AI Factory

Nvidia introduced NVLink Fusion, an interconnect architecture that allows third-party custom accelerators to integrate into its scale-up infrastructure. The technology utilizes sixth-generation NVLink to support fabrics of up to 72 devices, with future roadmaps expanding to 1,152 chips. It also employs NVLink-C2C to enable high-efficiency chip-to-chip communication with processors like Nvidia Vera.

“The end-to-end latency for XPU-to-XPU transfers is 3x lower than alternative solutions based on off-the-shelf Ethernet, and the packet rate is 10x higher.” — NVIDIA

Related: NVIDIA B300 GPU accelerator , AC1200 Wi-Fi Bluetooth 5.0 PCIe Adapter , AXE5400 Wi-Fi 6E Bluetooth PCIe Adapter , AX1800 Wi-Fi 6 Bluetooth 5.2 PCIe Adapter , BE9300 Wi-Fi 7 Bluetooth 5.4 PCIe Adapter , AX3000 Wi-Fi 6 Bluetooth 5.2 PCIe Adapter , AXE5400 Wi-Fi 6E Bluetooth 5.3 PCIe Adapter , BE6500 Wi-Fi 7 Bluetooth 5.4 PCIe Adapter

With Groq 3 LPX in Full Production, NVIDIA Extends Vera Rubin Inference for Agents

NVIDIA announced that its Groq 3 LPX rack-scale inference system has entered full production, operating in conjunction with the Vera Rubin NVL72 platform. The codesigned infrastructure combines dedicated generation units, Vera CPUs, and Spectrum-X networking to accelerate long-context agentic workloads. Early enterprise deployments include infrastructure implementations by Nebius and CoreWeave.

“it delivered 3,400 output tokens per second for 100,000-token long-context use cases critical to agentic systems” — NVIDIA

Up to 30x More Work Per Watt: NVIDIA Vera Rubin NVL72 Sets a New Efficiency Standard for AI Agents

NVIDIA shared preliminary efficiency metrics comparing its upcoming Vera Rubin NVL72 rack architecture against GB300 NVL72 systems on complex AI agent benchmarks. Evaluated on the SemiAnalysis AgentX test suite, the platform aims to reduce energy consumption per token during long-context execution. The system also leverages DSX MaxLPS power management to support higher GPU density within fixed power limits.

“NVIDIA DSX MaxLPS technologies manages power across the GPU, rack and workload levels to provision up to 40% more GPUs within the same megawatt budget” — NVIDIA

Related: NVIDIA RTX PRO 4000 Blackwell , NVIDIA RTX PRO 1000 Blackwell Mobile , NVIDIA RTX PRO 4000 Blackwell Mobile , NVIDIA RTX PRO 2000 Blackwell Mobile , NVIDIA RTX PRO 2000 Blackwell , NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition , NVIDIA RTX PRO 5000 Blackwell , NVIDIA RTX PRO 3000 Blackwell Mobile

Intel Outlines Architectures for Agentic AI at Hot Chips 2026

At Hot Chips 2026, Intel detailed three hardware architectures designed for enterprise and artificial intelligence workloads. The lineup comprises the Diamond Rapids processor, the Crescent Island graphics processor for inference tasks, and the Wildcat Lake system-on-chip for edge and client applications.

Arm and UNICEF launch AI in Play to expand access to AI innovation

Arm and UNICEF have introduced AI in Play, a program designed to support artificial intelligence projects in emerging economies. The collaboration focuses on addressing challenges in healthcare, education, and climate across 12 countries. Targets include supporting 20 startups and providing foundational AI literacy instruction to 200,000 young individuals.

IBM and Arm: Expanding the Arm ecosystem for the next era of enterprise computing

IBM has introduced a dual-architecture processor intended for future enterprise mainframe systems. The hardware features individual cores that natively execute both IBM and Arm instruction sets, enabling Arm-native Linux software to operate directly alongside traditional z/OS and Linux environments on the same platform.

“each processor core is designed to natively execute both IBM and Arm instructions, allowing Arm and IBM software environments to coexist within the same enterprise system.” — Arm

Why agentic AI is a systems challenge

Agentic artificial intelligence represents an architectural shift from standalone model optimization to whole-system coordination. These workloads rely on complex pipelines linking specialized models, memory management, software harnesses, and heterogeneous processors to execute iterative tasks. Consequently, overall system efficiency, reliability, and orchestration matter more than raw token throughput.

Bring the Fire: Play Games on GeForce NOW With New Firefox Browser Support

NVIDIA has introduced Mozilla Firefox browser compatibility for its GeForce NOW cloud streaming platform on Windows. The web interface allows Ultimate tier subscribers to run titles at resolutions up to 1440p at 120 frames per second without installing a dedicated client. The service also integrated twelve additional games into its library.

“The experience delivers GeForce RTX-powered performance at up to 1440p and 120 frames per second for GeForce NOW Ultimate members.” — NVIDIA

Related: NVIDIA GeForce RTX 3090 Ti , NVIDIA GeForce RTX 5050 , NVIDIA GeForce RTX 4070 Ti , NVIDIA GeForce RTX 4090 Mobile/Laptop , NVIDIA GeForce RTX 2080 Ti , NVIDIA GeForce RTX 3080 Ti , NVIDIA GeForce RTX 2080 Max-Q , NVIDIA GeForce RTX 2070 Super Max-Q

Six in 10 Leaders Bet Big on Robots. Only Four in 10 Are Ready.

A survey of enterprise and public sector decision-makers reveals that sixty percent plan to deploy robotic systems within five years. While respondents project significant output gains, only about two-thirds anticipate having the infrastructure to manage a combined human and robotic workforce by 2030.

AMD Appoints Tim Ryan to Board of Directors

AMD has appointed Tim Ryan to its board of directors. The addition represents an update to the company's governance and oversight leadership.