NVIDIA Developer Blog

[NVIDIA] NVIDIA said on November 10, 2026 that it is announcing two open source projects for building and operating managed Kubernetes services on GPU infrastructure, one of which is the NVIDIA Kubernetes Engine (NVKE). According to NVIDIA, the projects target Kubernetes cluster management on GPU infrastructure.

[NVIDIA] NVIDIA said in a blog post dated Nov. 10, 2026 that NVIDIA Snapshot, a new open-source component of the NVIDIA DSX AI Factory Platform, makes initialized GPU workloads reusable across inference and other scenarios, enabling GPU workloads to start and restore in seconds, according to NVIDIA.

[NVIDIA] NVIDIA said in a blog post dated Nov 10, 2026 that it is advancing Kubernetes support for AI agents at scale with an approach it calls Agent Substrate. According to NVIDIA, AI agents spend most of their time idle, waiting on a human prompt, a tool result or a model response, and Kubernetes was not built for that pattern; running one Pod per agent is problematic. The post did not disclose performance figures or availability timing.

[NVIDIA] NVIDIA said in a blog post dated Nov. 4, 2026, that its DSX OS can be used to configure, certify and operate Kubernetes GPU clusters. According to NVIDIA, a single underperforming GPU can slow an entire distributed training job, and the node may pass basic health checks and show normal utilization, leaving operators unable to spot the problem.

[NVIDIA] NVIDIA said in a blog post dated Nov. 4, 2026 that high bandwidth memory (HBM) is the fastest, most expensive and most contested memory tier in an AI factory, and outlined five co-design approaches aimed at maximizing the value of every byte in the AI factory. According to NVIDIA, every byte spent on a redundant weight copy carries a cost. The abstract did not disclose specific performance or cost figures.

[NVIDIA] NVIDIA said in a blog post dated Nov. 4, 2026 that as GPU clusters scale, operations span compute, network, hardware and scheduling, and site reliability engineers (SREs) often have to examine multiple dashboards. According to NVIDIA, it uses AI agents to streamline AI factory operations.

[NVIDIA] NVIDIA released CUDA 13.5 on November 4, 2026. According to NVIDIA, the release adds support for NVIDIA Vera Rubin GA. NVIDIA said the CUDA update arrives as artificial intelligence and accelerated computing continue to push the boundaries of what is possible.

[NVIDIA] NVIDIA on November 2, 2026 released its Factory Operations Blueprint, designed to coordinate multiple AI agents in investigating production issues. According to a Siemens report cited by NVIDIA, unplanned downtime costs an average large plant in the surveyed sectors $253 million per year. NVIDIA said the blueprint organizes AI agents to work together on tasks such as investigating a failed production issue.

[NVIDIA] NVIDIA said in a blog post dated November 3, 2026 that developers can use NVIDIA NeMo to turn production evidence from agent failures into prioritized issues and custom evaluations, helping improve prompts, tools, models, memory, skills and harness code. According to NVIDIA, the approach is meant to convert production failures into reusable tests and fixes.

[NVIDIA] NVIDIA said in a blog post dated Oct. 28, 2026, that its DSX AI Factory Platform brings together technologies to help the ecosystem design, build and operate more efficient AI factories, including open source components. According to NVIDIA, the NVIDIA Workload Map is intended to enable faster support for new AI frameworks.

[NVIDIA] NVIDIA Cosmos is aimed at building world models for interactive simulation, according to an NVIDIA blog post. NVIDIA said physical AI developers working with world models face two distinct challenges: adapting models to their sensors, tasks and data, and getting those models into use. The post did not provide further details.

[NVIDIA] NVIDIA said in a technical blog post that configuring and optimizing Kubernetes for new GPU hardware such as NVIDIA Vera Rubin is complex and error-prone, spanning drivers, kernels, runtimes and networking. According to NVIDIA, the goal is to make Kubernetes AI-ready for Vera Rubin from day one.

[NVIDIA] NVIDIA said in a technical blog post dated Nov 2, 2026 that as distributed AI training scales, failures across hardware and software dependencies increasingly disrupt training progress, making fast and reliable recovery more important. The post describes automated workload recovery to keep distributed training running, without disclosing specific performance figures or customer cases.

[NVIDIA] NVIDIA published a blog post on November 2, 2026 introducing its Attribution tool for diagnosing failures in distributed AI workloads. According to NVIDIA, a single error in such workloads can produce a cascade of symptoms: a GPU or fabric event may surface as NCCL errors on other ranks and affect scheduler state, among other effects. The tool is positioned to help trace such cascading failures back to their source.

[NVIDIA] NVIDIA released NemoClaw on October 30, 2026, for building specialized agents on a stack developers control. According to NVIDIA, building an autonomous agent requires a model, a harness and domain-specific tools, and deploying an agent more securely requires a runtime that enforces constraints. The announcement was published on the NVIDIA developer blog.

[NVIDIA] NVIDIA said in a blog post dated Oct. 30, 2026 that developers can run AI models for edge application development with just a few commands. According to NVIDIA, AI is expanding beyond the data center as developers build a new generation of agents that act at the edge, such as monitoring driver attention and cabin conditions. The post did not disclose the specific commands, the list of supported models or performance figures.

[NVIDIA] NVIDIA published a blog post on building a validated Kubernetes runtime for NVIDIA Dynamo. According to NVIDIA, Dynamo is a distributed inference system, and teams building AI applications need a repeatable path from a model configuration to a working inference endpoint. The post is aimed at developers deploying Dynamo on Kubernetes and describes how to set up and validate the runtime environment.

[NVIDIA] NVIDIA on October 30, 2026 published a blog post on how to check and configure NVIDIA GPUDirect Storage (GDS) for different platforms. According to NVIDIA, GDS accelerates GPU storage access in on-premises and cloud environments and provides a robust, portable and efficient data path. The post is aimed at users deploying GDS and covers the checking and configuration process.

[NVIDIA] NVIDIA published a technical blog on October 29, 2026, titled "Unlocking Tensor Core Performance for Machine Learning Interatomic Potentials." According to NVIDIA, atomistic simulations help researchers understand how molecules and materials behave, from chemical reactions and phase transitions to mechanical failure. The post addresses how Tensor Core performance can be unlocked for machine learning interatomic potentials.

[NVIDIA] NVIDIA published a blog post outlining three methods to boost mixture-of-experts (MoE) inference performance on DGX Spark. According to NVIDIA, generating tokens faster starts with identifying what limits inference performance, and the post presents analysis from a batch-size-1 evaluation of Ling-3.0-flash from Inclusion AI on NVIDIA hardware. The abstract did not disclose specific performance figures.

[NVIDIA] NVIDIA published a blog post on October 26, 2026, describing how to fine-tune its Nemotron model so that it knows local rules, not just local languages. According to NVIDIA, as AI adoption expands globally, people across languages and cultures are using it in more aspects of their daily lives, and general-purpose language models fall short in serving those needs. The post covers fine-tuning approaches for localized rule knowledge; specific technical details and data were not disclosed in the summary.

[NVIDIA] NVIDIA on Oct. 28, 2026 introduced RTX Spark, which the company described as a new Windows PC platform for AI development, creation and gaming in slim laptops and compact desktops. According to NVIDIA, the platform is powered by an NVIDIA Blackwell RTX and powers neural rendering and AI characters in games. NVIDIA said RTX Spark targets AI development, creation and gaming workloads.

[NVIDIA] NVIDIA said in a blog post dated October 27, 2026 that many AI workloads need only a fraction of a GPU, and giving each one an entire device can leave capacity idle, while sharing GPUs lets platform teams serve more workloads. The post describes memory-enforced GPU sharing in Kubernetes with NVIDIA CUDA 13.4.

[NVIDIA] NVIDIA said in a blog post dated Oct. 23, 2026 that AI factories are among the most technologically complex operational environments ever created, housing hundreds of thousands of GPUs alongside NVIDIA Spectrum-X Ethernet. According to NVIDIA, its Minerva system is designed to turn telemetry chaos into peak GPU yield.

[NVIDIA] NVIDIA released Mission Control 2.5, bringing autonomous recovery to Kubernetes, according to NVIDIA. The company said training large-scale AI models on thousands of GPUs is a high-stakes operation in which monitoring and automation must be built in to optimize cluster resiliency.