Volume 343, Issue 2 · 7 articles from 2 issues
NVIDIA explains why telcos are turning to open models such as Nemotron for cost, customization, trust and flexible deployment.
NVIDIA says telecom operators increasingly build AI strategies on open models, citing its State of AI in Telecommunications report, in which 89% of respondents called open source models and software important.
Open models offer lower-cost access to frontier-level intelligence, can be fine-tuned with network and customer data, provide more visibility and control for governance, and can run across public cloud, private and edge environments so telcos can sell locally adapted AI services.
SoftBank describes using NVIDIA Nemotron and other open models to build its SoftBank Large Telecom Model.
NVIDIA shows developers using frontier AI agents such as GPT-6 Astra with Omniverse libraries to build physics-based simulation apps.
NVIDIA describes how developers combine frontier AI models with NVIDIA Omniverse libraries, which supply GPU-accelerated physics, rendering and sensor simulation, to build simulation applications by directing agents in natural language.
In one example, Omniverse product manager Frank DeLise used GPT-6 Astra to turn a SimReady warehouse and humanoid robot into an interactive simulator with first- and third-person views.
The agent connected libraries for physics (ovphysx), scene updates (ovstage), rendering (ovrtx) and UI (ovui).
At OCP Summit 2026, NVIDIA contributed Vera Rubin rack and tray designs and MGX specs for power, cooling and networking.
NVIDIA used the Open Compute Project Global Summit to contribute open designs for AI factory infrastructure.
These include Vera Rubin rack and compute tray designs through its MGX architecture covering power, cooling and mechanical systems, with a fully liquid-cooled, cable-, hose- and fan-free modular tray, plus an updated MGX specification supporting racks of up to about 250kW.
On networking, NVIDIA and Cisco announced an ultra-dense small form-factor pluggable standard for 400Gb/s fabrics.
NVIDIA says the contributions aim to speed AI factory buildouts by letting the ecosystem standardize and build supply chains around common designs.
At OCP Global Summit, NVIDIA contributed Vera Rubin rack and compute tray designs to OCP via its open MGX architecture.
At the Open Compute Project Global Summit in San Jose, NVIDIA announced partnerships and contributions aimed at a common foundation for next-generation AI infrastructure.
NVIDIA contributed key Vera Rubin rack and compute tray designs to OCP through its open MGX architecture, giving system builders a shared base for AI factory hardware.
The efforts span the AI factory itself, the utility grid that powers it and the rack-scale architecture, as AI factories scale from megawatts to gigawatts.
A lab review finds NVIDIA's single-slot, 165W RTX PRO 4500 delivers near-L40S inference throughput for brownfield servers.
NVIDIA aimed the RTX PRO 4500 Blackwell Server Edition at air-cooled, brownfield data centers with tight rack power limits.
It is a single-slot, passively cooled 165W card with 32GB of GDDR7, FP4 Tensor Cores and Multi-Instance GPU support, and it replaces the L4 as the entry RTX PRO server card.
StorageReview tested it in a Dell PowerEdge R770 against the 350W L40S, the L4 and RTX PRO 6000 data.
Target workloads include small language model inference, video analytics, vector search, transcoding and virtual workstations.
NVIDIA released Nemotron 3.5 Super VL, a 120B open model for always-on agents, plus two Bolt edge models for local agentic workloads.
NVIDIA released Nemotron 3.5 Super VL, an open 120-billion-parameter model it says delivers the fastest task completion and highest accuracy for specialized always-on agents, alongside two new edge models: Nemotron 3.5 Bolt 74B-A9B 24GB, tuned for intelligence and speed per GB for local agentic workloads, and Nemotron 3.5 Bolt E4B 4GB for fast conversational agents.
Super VL and Bolt 74B-A9B are available now, with Bolt E4B due October 23.
Together the open models are meant to cover agent deployments from enterprise servers to edge devices and local hardware.
NVIDIA and Cisco unveiled USFP, a 400Gb/s-per-lane pluggable optical module for scaling Ethernet AI fabrics up to 12.8Tb/s.
NVIDIA and Cisco announced plans for Ultra-Dense Small Form Factor Pluggable (USFP), a high-density optical module designed as the first to support 400Gb/s-per-lane electrical signals, enabling up to 12.8Tb/s.
The companies say USFP offers 4x the bandwidth density and 1.3x the power efficiency of 200Gb/s-per-lane optics, letting AI factories connect more systems without extra network tiers.
NVIDIA and Cisco are working with TE Connectivity and Amphenol to build a multi-source hardware ecosystem around the form factor.