NVIDIA is shipping AI hardware and tooling across data center, workstation, robotics, and media workflows while raising server prices to its largest customers and absorbing a memory shortage that has pushed its consumer GPUs above 1 million yen in Japan. It is also funding startups and customers at a scale that has drawn the label of central bank of AI.
NVIDIA's most recent moves cover a wide range. The company added AI tools for broadcast work, including 3D Body Pose, Video Frame Generation, Video Super Resolution, TrueHDR, LipSync, Active Speaker Detection, and Studio Voice, to its AI for Media collection. It announced the RTX PRO 5500 Blackwell Workstation Edition with 21,760 CUDA cores, 84GB of GDDR7 memory with ECC support, 1398GB/s bandwidth, and 600W maximum power. Driver 616.92 added DLSS Super Resolution to PUBG: BATTLEGROUNDS and Dynamic Multi Frame Generation to WARDOGS, Active Matter, and Aniimo, with Active Matter also getting Ray Reconstruction and 007 First Light receiving path tracing and Ray Reconstruction on September 15, 2026.
On the business side, the stories track rising costs. NVIDIA told major customers that servers using its AI chips, including Vera Rubin and Grace Blackwell, will cost more than 15% more, with the increase applied to shipments from early 2027 after memory prices climbed. In Japan, GeForce RTX 50 series cards rose, with MSI's VENTUS RTX 5070 at 121,010 yen and RTX 5080 at 242,222 yen, and on September 11 a Sofmap store listed a single ASUS ROG Astral RTX 5090 at 1,022,800 yen. NVIDIA invested more than $70 billion in startups over three years and provided $300 billion in customer financial support, per The Economist, while Amazon, Google, Meta, and Microsoft, which account for about half of sales, design their own semiconductors.
The DGX line continues to expand. The DGX Station, announced at GTC Taipei 2026, runs a GB300 Grace Blackwell Ultra Desktop Superchip with 20 petaflops of FP4 performance, up to 748GB of memory, and support for models up to 1 trillion parameters. It is scheduled for the fourth quarter of 2026 through ASUS, Dell, GIGABYTE, HP, MSI, and Supermicro. The DGX Spark ranked second in Kakaku.com's desktop popularity ranking as of August 21. On the roadmap, NVIDIA canceled original Rubin Ultra development after CoWoS-L interposer warping made prototypes non-functional, introduced a stopgap NVL576 rack, and repurposed the Rubin Ultra name.
Synthesized by Yomimono from the cited Yomimono stories below, each itself sourced, then editorially reviewed. Every fact links the story it came from.
Structured graph also available as JSON at /public/entities/nvidia.
CC BY 4.0.
When a claim about NVIDIA was confirmed, debunked, or disputed against open-web sources. The record stays even after a claim drops off the facts list.
Pre-orders for the duck-shaped robot Microduck surpassed 20,000 units after NVIDIA CEO Jensen Huang operated one by voice command on stage at GTC 2026 in Berlin. Hugging Face CEO Clem Delangue posted that the keynote drove the surge and said Pollen Robotics, the developer, needs to grow its team. Microduck is 25 cm tall, weighs 800 g, carries 15 motors plus a camera and sensors, and costs $399 before tax and shipping.
NVIDIA released Kumo Relational, a pretrained foundation model that makes predictions directly against corporate relational databases. Given a prediction query, it gathers up to 10,000 past cases as context and forecasts churn, supply and demand, and fraud in a single instruction. NVIDIA says it needs no task-specific training, no feature engineering, and no flattening of tables, and that it surpassed supervised machine learning models on benchmarks of common relational databases.
At an October 16 media interview after the RTX Spark Launch Party: Taipei, NVIDIA Senior Vice President Jeff Fisher denied rumors that production of the GeForce RTX 5090 has ended, saying shipments will continue. German outlet PC Games Hardware had reported on October 15, citing distribution and retail officials, that new supply of the GB202 chip used in the card had halted. Fisher said production capacity has constraints but NVIDIA is doing what it can to get products to market.
Nvidia detailed its RTX Spark system on a chip for Windows PCs at a technical briefing in the United States on October 8, after Microsoft, ASUS, MSI and others announced machines built around it the previous day. Nvidia calls the part RTX Spark N1X and denied the existence of a rumored lower-tier N1. Memory comes in five LPDDR5X-9400 capacities from 24GB to 128GB.
NVIDIA released Nemotron 3.5 Super VL on October 15, 2026, an open-weight model that reads text, images, and video in one system and operates external tools. It carries 120 billion parameters and, through mixture-of-experts routing, uses 12 billion per token. NVIDIA claims up to five times the throughput of open models at the same scale. Scale AI tuned a version for Skydance that cut processing costs to about one-fortieth.
NVIDIA announced on October 14 that the PC version of Hogwarts Legacy will add DLSS 5 support on November 5, 2026. Avalanche Software, the developer, used the DLSS 5 SDK to tune the integration itself, and the announcement states the studio preserved the game's artistic intent and style. DLSS 5 introduces 3D-Guided Neural Rendering, which uses AI to generate lighting and material detail in real time on top of already rendered frames. The update reaches GeForce RTX 50 series, GeForce NOW, and RTX Spark users. Hogwarts Legacy is the open-world action RPG set in the wizarding world of the 1800s, published by Warner Bros., released in 2023 with more than 40 million copies sold. The PC version already supports DLSS multi-frame generation and ray reconstruction, and the other DLSS features will move to their latest versions alongside the new one. NVIDIA also said DLSS 5 support arrives for Where Winds Meet in late November and for RTX Remix on October 23.
NVIDIA and Cisco announced USFP, a pluggable optical module form factor for AI factories, unveiled in Santa Clara, California. The module is designed to support 400Gb/s electrical signaling per lane and reach up to 12.8Tb/s of capacity, with bandwidth density said to improve 4x and power efficiency about 1.3x over conventional 200Gb/s-per-lane pluggable modules. TE Connectivity and Amphenol are expected to join the multi-source ecosystem.
NVIDIA announced the 64GB unified memory model of its desktop AI system DGX Spark on October 2. In the United States it goes on sale October 23 from six companies, Acer, ASUS, Dell, Gigabyte, HP, and MSI, at a starting price of $4,999, about 790,000 yen. The 64GB version runs AI models up to 100 billion parameters locally using the same GB10 Grace Blackwell superchip, with up to 1 PFLOP (FP4) and 273GB/s bandwidth.
NVIDIA announced on October 8, 2026 that it will invest about $1 billion over five years to support U.S. science and technology development. The plan covers support for higher education research institutions, investment in quantum computing, and backing for cloud providers running U.S. government missions. NVIDIA joins the national project Genesis Mission as a cooperating company on quantum computing, nuclear fusion, accelerator design, and microelectronics.
A Hot Chips session on NVIDIA's Vera server chip walks through the custom Olympus core, which fetches 16 instructions per cycle and decodes and dispatches 10 per cycle across 18 issue ports. The design keeps the decode throughput of the earlier Arm AGI CPU but pairs it with instruction fusion inside the decode queue and different micro-ops, which the writeup reads as an IPC advantage. NVIDIA designed Vera's core itself, where the earlier Grace chip used Arm's Neoverse V2.
Asmec Corporation has begun handling Vecow's EAC-7000, an Arm-based fanless edge AI computer built on NVIDIA Jetson Thor and aimed at robots and autonomous mobile bodies. The higher-end Jetson T5000 module carries a 14-core Neoverse V3AE CPU and a 2,650-core Blackwell GPU, with AI performance up to 2,070 FP4 TFLOPS. A T4000 configuration reaching 1,200 FP4 TFLOPS is also offered. Five models cover camera input, PoE+, and 10GigE variants.
Microsoft and NVIDIA opened pre-orders for the Surface Laptop Ultra on October 8, 2026, with the release set for October 16 at 9:00 PM. Sofmap is accepting reservations on a dedicated page. The notebook runs on the NVIDIA RTX Spark N1X, a chip pairing an Arm-based Grace CPU with an RTX GPU. The top configuration carries a 6144-core Blackwell RTX GPU, a 20-core Grace CPU, and up to 128GB of unified memory, enough to run AI models above 120B parameters locally and reach up to 1 petaflop of AI performance. The 15-inch PixelSense Ultra touchscreen peaks at 2000 nits. The chassis stays under 18mm thick and under 2kg, using a new thermal design rated at up to 2.5 times the capacity of existing Surface Laptops. Storage is user-removable, and charging moves to a magnetic USB Type-C system called Magnetic Connect. Microsoft's US price starts at $2599.
ASUS has opened global preorders for its ProArt P14 and P16 creator laptops built around NVIDIA's RTX Spark chip, announced at IFA 2026. US pricing runs from about $2,600 (roughly 410,000 yen) for a 14-inch 18-core model to about $7,000 (roughly 1.11 million yen) for a 16-inch 128GB configuration. Shipping starts October 16. Japanese availability is announced without pricing yet.
Greg Lui, 38, CEO of Earthmade Computer, was arrested on a three-count federal indictment. He is accused of conspiring with five others to ship export-controlled servers carrying NVIDIA A100, H100, GeForce RTX 4090 and GeForce RTX 5090 GPUs to customers in China without U.S. Commerce Department licenses, routing goods through Malaysia and Singapore. Prosecutors put the total value above $300 million.
ASRock announced on October 6 that its DeskSlim X600 and DeskSlim B760 compact desktop platforms now support the NVIDIA RTX PRO 4000 Blackwell SFF Edition and RTX PRO 2000 Blackwell graphics cards. The 5L-class chassis takes low-profile, dual-slot GPUs up to 200mm. Both platforms carry four DDR5 UDIMM slots, two M.2 SSD slots, and two SATA ports. A Japan rollout announcement is set for a later date.
A GitHub project called OpenDLSS-NR reimplements NVIDIA's DLSS 5 neural rendering network in Vulkan. It runs the same 71-block Swin/ViT network as DLSS-NR build 310.8.0 using FP8 on Tensor Cores, and the output is described as bit-for-bit identical to the original, with all 75 block boundaries matching byte-for-byte. It requires Windows, an NVIDIA Ada Lovelace or later GPU, and the user's own DLSS 5 model. The project states it is unrelated to NVIDIA.
China's Ministry of Industry and Information Technology has asked select domestic companies, including Alibaba and ByteDance, to report how many NVIDIA RTX Pro 5500 GPUs they want and what they will be used for, and plans to approve the purchases, The Information reports. The RTX Pro 5500 is a workstation GPU NVIDIA announced in September 2026 for large-scale AI models and simulations.
NVIDIA released Nemotron 3 Diarization, a speech recognition model that identifies speakers while transcribing in real time. It handles up to eight speakers and labels them in order of appearance as Speaker 1, Speaker 2, and so on, without registering voices in advance. Training used public datasets and data licensed from David AI. The model is open, under the OpenMDW-1.1 license, and available on Hugging Face.
NVIDIA released the GeForce 617.14 WHQL driver on September 22, 2026, adding support for CONTROL Resonant, The Witcher 3: Wild Hunt Remaster, AION2, and Gears of War: E-Day. All four support DLSS 4.5 Super Resolution and Multi Frame Generation, while CONTROL Resonant gains ray tracing through DLSS 4.5 Ray Reconstruction. The driver also fixes Sega engine launch issues and Samsung Odyssey G95NC HDMI display problems.
NVIDIA announced the Open Agent Safety Platform, a framework for monitoring and constraining AI agents that act autonomously. The platform has two components. NVIDIA OpenShell is open-source software that runs on an NVIDIA Vera CPU and checks each tool an agent tries to use, letting operators set rules on files and network connections the agent can reach and blocking anything outside those rules. NVIDIA Sentry is a separate monitoring system that runs on the BlueField-4 data processing unit and can stop an agent that moves outside its boundary in milliseconds. NVIDIA CEO Jensen Huang called the platform a browser for agents and said trust in how AI is built and deployed is a condition for the AI industry's success. A company spokesperson said the July incident in which an OpenAI AI agent intruded into Hugging Face could possibly have been prevented with the platform.
Apple is developing an AI server built around M8 Ultra chips connected through NVIDIA's NVLink Fusion, according to The Information. Two configurations are in development, one with two M8 Ultras and one with four. Apple plans to sell the servers to enterprise customers, with launch reported around 2029. Manufacturing would involve TSMC's N3P process, Broadcom, and Foxconn.
NVIDIA has invested more than $70 billion in startups over the past three years and provided $300 billion in financial support to its customers, according to The Economist, which reported that the company is being called the central bank of AI. Giant technology companies including Amazon, Google, Meta, and Microsoft account for about half of NVIDIA's sales and are designing their own semiconductors.
NVIDIA added several AI tools to its AI for Media collection for the broadcast industry in September 2026. The new additions include 3D Body Pose, which converts joint positions into 2D and 3D motion data, Video Frame Generation for raising frame rates two to four times, Video Super Resolution for upscaling, TrueHDR for real-time SDR to HDR conversion, LipSync for matching mouth movements to audio, Active Speaker Detection, and Studio Voice for reducing noise and reverberation.
NVIDIA announced the RTX PRO 5500 Blackwell Workstation Edition, a professional GPU with 21,760 CUDA cores, the same count as the GeForce RTX 5090. Its GDDR7 memory is 84GB with ECC support, 52GB more than the RTX 5090's 32GB. Memory bandwidth is 1398GB/s, and the maximum power consumption is 600W.
Retail pricing for the NVIDIA GeForce RTX 5090 has crossed 1 million yen in Japan. On September 11, the Sofmap Omiya store listed a single ASUS ROG Astral unit at 1,022,800 yen including tax, a card that sat in the lower 600,000-yen range when it launched in Japan in June 2025. The store said it received one unit. Japanese prices were in the upper 300,000-yen to 400,000-yen range around September last year, and the climb has continued through this summer. In Germany, the same card has passed 5,000 euros, roughly 890,000 yen, with the cheapest listings at 5,249 euros and 5,299 euros. A retail industry source had predicted the cheapest class would reach 900,000 yen and special editions would clear 1 million yen. On September 6, Apply warned customers that some graphics cards are scheduled to rise further.
MSI Computer Japan released the GeForce RTX 5060 8G VENTUS 2X OC V1, a graphics card built on NVIDIA's Blackwell architecture. It has no RGB lighting and uses an original dual-fan cooler with MSI's TORX FAN 5.0 design, which links the outer edges of three fan blades for steadier airflow to the heatsink. The dual-fan layout keeps the card compact for a range of cases, and the warranty on GeForce RTX 50 series cards is two years.
NVIDIA released Game Ready driver 616.92, which adds DLSS Super Resolution to PUBG: BATTLEGROUNDS and Dynamic Multi Frame Generation to WARDOGS, Active Matter, and Aniimo, with Active Matter also getting Ray Reconstruction. Path tracing and Ray Reconstruction for 007 First Light arrive September 15, 2026. The release fixes browser flickering on some sites, while a power management bug carried over from version 610.47 remains, where Prefer Maximum Performance may not apply correctly.
NVIDIA's Agentic Variation Operators (AVO) agent system scored 100% on the public set of ARC-AGI-3, a benchmark measuring AI reasoning in unknown game-like environments. The base model 5 used inside AVO scored about 30% in a separate evaluation by ARC Prize, the benchmark's developer. AVO cleared all 183 levels across 25 environment types.
NVIDIA has notified some major customers that prices on servers equipped with AI chips, including the Vera Rubin and Grace Blackwell platforms, will rise by more than 15%. The increase is expected to apply to shipments from early 2027. The move follows soaring memory prices, with demand for memory growing faster than production volume.
Nvidia's AI development PC DGX Spark ranked 2nd in Kakaku.com's Desktop PC Popularity Best Seller Ranking as of August 21. The roughly 1 million yen local AI PC appeared alongside gaming PCs priced at 100,000 to 300,000 yen, showing strong demand among local AI developers and users.
NVIDIA canceled original Rubin Ultra development after CoWoS-L interposer warping made prototypes non-functional. The 180-layer CoWoS-L warped in multiple directions, causing dies and wiring to peel. NVIDIA shifted to calling a different configuration Rubin Ultra and introduced a stopgap NVL576 rack solution using eight NVL72 (V2) units.
GeForce RTX 50 series GPU prices are rising in Japan, following increases in China. MSI's VENTUS GeForce RTX 5070 now sells for 121,010 yen, up nearly 20,000 yen, while the RTX 5080 is 242,222 yen, up nearly 30,000 yen. NVIDIA has not made an official announcement.
NVIDIA announced NVIDIA Halos for Robotics, a full-stack safety system for physical AI and robots. The system spans hardware, software, and sensor layers, building on the Halos architecture previously used for autonomous vehicles. Agility is the first adopter, integrating safety features into humanoid robots deployed at Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada.
Nvidia has announced ArtiFixer, an AI model based on the Wan 2.1 video generation AI that can generate 3D scenes from multiple photos. The model fills in missing parts using generative processing, improving on conventional 3D Gaussian Splatting methods that often fail to maintain scene consistency. ArtiFixer comes in three variants: ArtiFixer, ArtiFixer3D, and ArtiFixer3D+.
Nvidia announced ENPIRE, a harness framework developed with Carnegie Mellon University and UC Berkeley that lets AI agents like Claude Code and Codex autonomously improve real-world robot task execution. In tests, it achieved a 99% success rate on tasks like inserting a pin and cutting a cable tie by running a continuous improvement loop.
Nvidia's AI research team announced MotionBricks, a motion generation framework for game animation and robot control. It handles over 350,000 motion skills with a single neural backbone, achieving 15,000 FPS and 2 ms latency. The technology was presented as a SIGGRAPH 2026 paper accepted by ACM Transactions on Graphics.
Real World Inc. announced a joint initiative with NVIDIA to build common industry infrastructure for humanoid robot AI, centered on a new dexterous manipulation benchmark called DexBench, data standards for motion learning, and integration with NVIDIA's Isaac robotics platforms.
Nvidia announced the DGX Station, a Windows desktop PC built around the GB300 Grace Blackwell Ultra Desktop Superchip, during its GTC Taipei 2026 keynote. The system delivers 20 petaflops of FP4 performance, supports up to 748GB of memory, and can run AI models with up to 1 trillion parameters locally. It is scheduled to launch in the fourth quarter of 2026 through partners including ASUS, Dell, GIGABYTE, HP, MSI, and Supermicro.