{"slug": "scale-in-nvidias-fifth-pillar-of-ai-networking-rests-on-shaky-ground", "title": "Scale-in? Nvidia’s ‘fifth’ pillar of AI networking rests on shaky ground", "summary": "Nvidia Corp. introduced a new networking concept called 'scale-in' at the Hot Chips event in Palo Alto, California, claiming it as a 'fifth' pillar of AI networking that unifies north-south traffic by accelerating infrastructure services such as security, storage, and management. The concept relies on Nvidia's BlueField-4 data processing unit (DPU), Spectrum-X Ethernet fabric, and DOCA software stack to offload tasks from host processors, but the company has provided sparse technical details, drawing skepticism about its viability. Nvidia also spotlighted a multiplane configuration for its Spectrum-X offering, which splits network connections into independent planes to scale to 512,000 GPUs without a third tier.", "body_md": "Nvidia claims to have invented a “fifth” pillar of AI networking focused on unifying north-south traffic. Dubbed “scale-in,” the concept is sparsely outlined in an Nvidia [blog post](https://blogs.nvidia.com/blog/vera-rubin-lpx-spectrum-x-nvlink-fusion/#scale-in) of underwhelming announcements from the Hot Chips event in Palo Alto, California.\n\nNvidia said scale-in “extends purpose-built acceleration to the infrastructure services that secure, manage, and operate the AI factory.” In non-marketing speak, it seems Nvidia networking hardware like its BlueField-4 data processing unit (DPU) is used to speed up data flow between an internal network or data center and external systems. Leveraging its Spectrum-X Ethernet fabric and software optimizations via the DOCA software stack, Nvidia contends scale-in can be employed to offload heavy security and storage tasks from the main server processor.\n\nA more detailed [technical blog](https://developer.nvidia.com/blog/nvidia-bluefield-4-powers-new-scale-in-network-infrastructure-for-agentic-ai-factories?nvid=nv-csfg-653504) describes scale-in: “BlueField-4 provides host-independent acceleration and offload, DOCA provides a unified model for programming and operating infrastructure services, and Spectrum-X Ethernet provides high-performance Ethernet connectivity across the scale-in access path, including external storage and data sources. Together, they give operators consistent control over access, data movement, security, provisioning, and observability across the infrastructure that supports accelerated compute and data storage.”\n\nScale-in was slapped with the agentic label in Nvidia’s Hot Chips announcements, with the chip giant’s blog post suggesting storage and operations beyond AI compute need to be accelerated to meet agentic workload demands. Such an idea is widely accepted in the industry, but more in the sense that accelerators have exponentially improved while memory hardware has lagged behind. Nvidia’s agentic billing of its apparent networking concept means to provide “secure, efficient and manageable shared infrastructure for deploying and operating agentic AI at massive scale.”\n\n“Nvidia scale-in delivers multi-tenant networking, high-performance storage access, in-silicon security, elastic provisioning and real-time observability, while keeping infrastructure processing independent of host compute resources,” the blog post reads. “By accelerating and codesigning these services as part of the AI factory, scale-in helps security, data access and operations scale alongside AI compute.”\n\nOn the point of pillars, the main three pillars of networking are largely accepted: scale-up for inside the tray or rack, scale-out for rack-to-rack or cabinet-to-cabinet connections, and scale-across, which is essentially distributed computing ranging from campus-to-campus links to disparate facilities spread across tens of miles.\n\nThe so-called fourth pillar is where things get murky. Nvidia doesn’t inherently define what the fourth pillar is, though it seems to be related to its [context memory storage platform, or CMX](https://www.sdxcentral.com/news/nvidia-pushes-ai-from-edge-to-storage-with-jetson-t4000-and-bluefield-4-updates/). Unveiled back in January, the offering pitches the BlueField-4 data processing unit as an effective backbone to extend GPU memory capacity while enabling high-speed sharing of context across clusters.\n\n## Multiplane ‘spotlight’\n\nThe so-called fifth pillar wasn’t the only networking idea on display at Hot Chips from Nvidia. The chip giant was “[spotlighting](https://blogs.nvidia.com/blog/vera-rubin-lpx-spectrum-x-nvlink-fusion/#spectrum-x-multiplane)” its Spectrum-X offering in a multiplane configuration.\n\nThe concept sees server network connections split into independent planes on each of which runs its own lightweight two-tier network. Nvidia claims this simpler flat network can scale to 512,000 GPUs minus the additional overhead complexity of a third tier. A dedicated engine inside Nvidia’s ConnectX SuperNIC manages traffic across all of the planes, instantly rerouting around failures.\n\nIn an eight-plane topology, if one plane fails, the firm claimed networks can still maintain about 90% of their total bandwidth, “with hardware recovery that’s 11x faster than software-based multiplane load balancing. This translates to 1.6x higher AI factory output.”", "url": "https://wpnews.pro/news/scale-in-nvidias-fifth-pillar-of-ai-networking-rests-on-shaky-ground", "canonical_source": "https://www.sdxcentral.com/news/scale-in-nvidias-fifth-pillar-of-ai-networking-rests-on-shaky-ground/", "published_at": "2026-08-25 10:09:57+00:00", "updated_at": "2026-08-25 10:13:55.480925+00:00", "lang": "en", "topics": ["ai-infrastructure"], "entities": ["Nvidia", "BlueField-4", "Spectrum-X", "DOCA", "Hot Chips", "Palo Alto"], "alternates": {"html": "https://wpnews.pro/news/scale-in-nvidias-fifth-pillar-of-ai-networking-rests-on-shaky-ground", "markdown": "https://wpnews.pro/news/scale-in-nvidias-fifth-pillar-of-ai-networking-rests-on-shaky-ground.md", "text": "https://wpnews.pro/news/scale-in-nvidias-fifth-pillar-of-ai-networking-rests-on-shaky-ground.txt", "jsonld": "https://wpnews.pro/news/scale-in-nvidias-fifth-pillar-of-ai-networking-rests-on-shaky-ground.jsonld"}}