TMTPost reported that China’s A-share indexes fell on November 13 after an early surge, with the ChiNext Index down 3.32% and the Shanghai Composite down 0.59%. The commentary linked weakness in computing-hardware stocks to Nvidia’s leadership change, while noting a rotation into consumer sectors and selected gains in domestic chip and computing-rental names. It also examined three first-day limit-up stocks and two consecutive limit-up stocks, highlighting differences in earnings, valuation and the extent of their AI exposure.
The Veyra repository released Veyra-NRVideo 2.1.6, an open-source Windows video player and capture-enhancement tool with real-time preview for local video, capture cards, images, and game streams. Version 2.1.6 adds instant replay, microphone mixing and separate audio tracks, HDR-control refinements, and presentation, logging, and variable-frame-rate fixes. The project describes AI super-resolution, denoising, frame generation, HDR, capture, and console streaming as core features, but its DLSS 5 and PS5 GTA6 120 FPS claims lack independent validation.
Elon Musk said on X that a petition seeking Nvidia CEO Jensen Huang’s reinstatement was underway and that he would sign it. WIRED, as cited by TechNews and QbitAI, reported that an internal Nvidia post drew more than 1,600 comments supporting Huang, but comment volume does not establish a formal employee petition.
At Hannover Messe, NVIDIA demonstrated factory AI Agents, digital twins and GPU simulation with more than 20 manufacturing partners. The report highlighted a claim that PepsiCo’s digital twin increased production-line throughput by 20%, but did not specify the measurement period, baseline, or whether the result covered one line or an entire plant.
CNBC reports that Altimeter Capital Management and Atreides Management, which disclosed about $2.13 billion in Nvidia stock as of June 30, urged the board to conduct a deliberate leadership search and keep Huang's possible return among the options. The letter did not demand his unconditional reinstatement and respected the board's final authority. Separately, TechRadar reported that a “Stand with Jensen” petition seeking immediate reinstatement had exceeded 175,000 signatures, though signers' identities were not independently verified.
TMTPost analyzes T-head and Kunlun as potential beneficiaries of NVIDIA’s leadership change, comparing their customer demand, product roadmaps, software ecosystems and valuation prospects. The report says Alibaba disclosed progress toward a Hong Kong listing for T-head, while Kunlun filed its application earlier in the year. It also cautions that reported valuation scenarios of $25 billion to $62 billion for T-head and about $50 billion for Kunlun were projections or listing targets, not completed public-offering prices.
Nvidia director Mark Stevens sold about 4.2 million shares in three batches between August 31 and September 18, receiving approximately $947 million, according to SEC filings. Unlike a prior sale by Jensen Huang that was announced six months in advance under a Rule 10b5-1 plan, Stevens had no prearranged trading plan. The sales occurred roughly two months before allegations involving Huang surfaced, but the reports do not establish that Stevens knew of the allegations or acted improperly.
Nvidia’s board announced that co-founder Jensen Huang resigned as president, CEO and director with immediate effect, according to TMTPost. Chief financial officer Colette Kress was named interim CEO while continuing to oversee finance, Harvey C. Jones became the company’s first chair, and the board opened an internal and external search for a permanent successor; Nvidia shares initially fell nearly 14% on November 12.
NVIDIA said on November 12 that Jensen Huang had resigned as president, chief executive officer and board member, effective immediately. TMTPost reported that NVIDIA shares fell more than 10% in premarket trading on November 12 and closed down 4.34% after a rebound. The commentary linked the volatility to uncertainty over NVIDIA’s strategy and leadership after Huang’s departure.
Reuters, citing internal emails and two knowledgeable people, reported that Jensen Huang described his relationship with Alice Huang and Megaspeed International as routine business dealings to senior Nvidia executives, including general counsel Tim Teter and chief marketing officer Alison Wagonfeld. The report said Huang also discussed suing The New York Times over its reporting, but no lawsuit or legal letter followed. Nvidia’s independent directors said the board had engaged an outside law firm for an independent investigation.
Elon Musk said a petition seeking to reinstate Nvidia CEO Jensen Huang is underway and that he will sign it, according to IT之家. The report cited a post claiming that roughly 1,600 Nvidia employees wanted Huang back, while also saying that the petition’s organization and actual participation had not been verified.
TechNews reports that Siliconware’s 2.5D packaging development became the foundation for its CoWoS foundry business, which the company identifies as a current growth driver. R&D executive Jiang Dongsheng said standardized development processes and more than a decade of technical exploration supported its move toward co-packaged optics (CPO). Nvidia CEO Jensen Huang announced a plan for Siliconware’s CPO capacity to grow more than 15-fold by mid-2027, but Jiang cautioned that CPO is a new business with a small base.
TechCrunch reports that Nvidia invested tens of billions in startups annually and participated in 105 funding rounds this year, often as the largest investor. Investors and founders say Jensen Huang’s departure may reduce capital for AI-native startups, while interim CEO Colette Kress is expected to make investment strategy more conservative; the resulting effects on valuations, dealmaking and competition remain expectations rather than confirmed outcomes.
A guide shows how to deploy LiteLLM with the NVIDIA NeMo Switchyard routing plugin on AWS, allowing coding-agent requests to be routed among models on Amazon Bedrock. Stage routing uses recent tool activity and request-history results to select one of two candidates, while Random routing supports weighted traffic splits and A/B comparisons. Sample benchmark runners compare the cost per solved task for a routed model group with single-model baselines. The Switchyard LiteLLM integration is experimental, and the sample code is intended for non-production use in a non-production account.
NVIDIA Technical Blog says CUDA Toolkit 13.4.2 introduces the Ozaki-II and 2M algorithms in cuBLAS to accelerate double-precision real and complex matrix multiplication on integer Tensor Cores. Ozaki-II requires fewer integer multiplications than Ozaki-I, while 2M reduces a complex multiplication from four real multiplications to two. NVIDIA reports that cuBLAS can apply the acceleration automatically without application-code changes, with benchmark results for scientific-computing applications.
NVIDIA applied deep learning research VP Bryan Catanzaro denied signing an online letter asking the board to reinstate former CEO Jensen Huang, calling it “obviously fake” and saying he would not sign it even if genuine. He said he supports NVIDIA’s board and interim CEO Colette Kress. IT之家 reported that the letter copied the template of the 2023 OpenAI staff letter supporting Sam Altman and included names of former or misidentified employees.
Nvidia said speculation that it is already engaging potential CEO candidates is premature. The company remains in the process of selecting a firm to find a successor to ousted leader Jensen Huang.
Nvidia’s board told CNBC it is still selecting a search firm to find a permanent successor to Jensen Huang and has not formally engaged any potential candidates. The statement conflicts with a Financial Times report that an external agency had already been hired, while Reuters-linked reporting said Arm CEO Rene Haas was among the potential candidates under consideration. Nvidia said it would consider both internal and external candidates, but it has not announced a successor or selection timetable.
Bloomberg Technology reports that Jensen Huang’s departure from Nvidia has left Silicon Valley uncertain about the future of the AI boom. The change raises questions about a central part of the financial infrastructure supporting artificial intelligence.
The Wall Street Journal, as reported by The Next Web, said interim CEO Colette Kress is expected to tighten Nvidia’s financing of AI cloud providers and neoclouds after Jensen Huang resigned. The company reportedly reduced its commitment to OpenAI’s Ohio data center lease, paused some deals and began requiring more scrutiny of guarantees. Huang had also urged neocloud partners to collect upfront customer payments before renting out Nvidia capacity.
RadixArk partnered with NVIDIA to run its open-source Miles post-training framework end to end, with every isolated Firecracker sandbox hosted on NVIDIA Vera CPUs. The reported run trained GLM-4.7-Flash for 68 steps across more than 2,400 Terminal-Bench and SWE-bench episodes in synchronous and fully asynchronous modes; one gRPC task’s pass rate rose from 40% to 82% between training halves. RadixArk also compared Vera with AMD EPYC Turin and Intel Xeon 61 systems, reporting 12% faster episodes under tested busy-server conditions.
Bill Ackman said it was sad to see Jensen Huang leave Nvidia and predicted he would do something great next. He also said Anthropic CEO Dario Amodei’s presentation at the company’s IPO roadshow event in New York impressed him.
AWS Machine Learning Blog details AssemblyAI’s production speech-to-text inference architecture, combining Amazon ECS, EC2 Spot and On-Demand capacity across multiple NVIDIA GPU families, and Amazon SQS retries. The design supports regional data-residency requirements and absorbs Spot interruptions; AssemblyAI reports that it reduced inference-compute costs by 70% compared with On-Demand pricing.
A Wall Street Journal profile, cited by Wccftech and TechSpot, says NVIDIA used its balance sheet to finance AI startups and data-center expansion so they could absorb shipments of TSMC silicon that the chipmaker had ordered in excess. The coverage says interim CEO Colette Kress reduced NVIDIA’s “circular financing” commitments to smaller cloud providers and AI firms including OpenAI and Anthropic. It also reports that NVIDIA sought to reduce upcoming chip orders and characterizes the financing as creating artificial demand to avoid a supply glut and lower GPU prices.
A New York Times report says Nvidia held stakes worth $48 billion in 229 private companies at the end of July, including OpenAI and Anthropic. It also says Nvidia offered $165 billion in cloud backstops. Some investors and founders said they had lost an ally in Jensen Huang.
NVIDIA says cuda-bindings 13.3.0 introduces cuda.bindings.cudla, a Python layer over the cuDLA C++ API for managing and orchestrating inference on NVIDIA Deep Learning Accelerator hardware. On Jetson AGX Orin, the bindings support hybrid workflows that coordinate GPU preprocessing and postprocessing with DLA inference through CUDA streams. The documented image-classification example covers loading a compiled network, allocating buffers, submitting work, synchronization, and retrieving output, while standalone mode remains unsupported in this release.
Gary Marcus argues that generative AI’s finances may be unraveling, citing lower prices, heavier competition, elusive profits and high training and inference costs. He says Nvidia, Oracle and CoreWeave have lost momentum, while Anthropic’s perceived valuation has declined, and interprets the figures as signs of a bubble popping, though he acknowledges his judgment could be wrong. Marcus also notes that technical improvement is possible but may not be fast enough to justify trillions in investment. He argues that many AI companies depend on Nvidia’s financing and could face trouble if that support changes.
Where’s Your Ed At alleges that Megaspeed and related entities form a multinational GPU-smuggling operation backed by NVIDIA. Citing Culper Research and The Wall Street Journal, the commentary says an alleged spy operated a purchasing front for the Chinese to buy GPUs. Culper estimated that the operation might account for more than 20% of NVIDIA’s fiscal 2026 revenue, or over $43 billion. The report says the situation was developing rapidly, and the allegations were not independently verified in the supplied material.
Cornelis has launched CN6000, a family of 800G network cards and switches designed for AI clusters and supercomputers, positioning the products as an alternative to Nvidia’s InfiniBand. The main card uses Intel’s Omni-Path technology and supports chips from AMD, Intel, Nvidia and Qualcomm, while Dell, Lenovo and Supermicro plan supporting systems. Cornelis says its card handles 1.6 billion messages per second and is 21% faster for small-message transfers, based on its own tests; partner systems are expected in early 2027.
A Bloomberg report says Nvidia CEO Jensen Huang stepped down after revelations of an extramarital relationship with a company vice president, which violated the company’s code of conduct. The report says Huang structured the company around himself, with no chairman, second-in-command or public succession plan, leaving his successor to manage an organization built around a leader who is no longer there. Bloomberg Opinion columnist Beth Kowitt discusses the case as an example of the “key-man” CEO issue.
What problem it solves
AI agents need to choose among models while balancing task success, cost, and latency for their workloads. Applications also need flexibility in where routing runs and who controls model clients, credentials, retries, and service operation.
Key features
Multiple routing algorithms — Routing can be configured with Auto, Task, Execution, or Composite strategies, including mode…
What problem it solves
Serving large language models requires efficient memory management, flexible model execution, and practical integration with existing models and APIs. vLLM addresses these needs with inference and serving capabilities spanning batching, caching, quantization, distributed execution, and multiple hardware platforms.
Key features
Efficient memory and serving — vLLM manages atte…
What problem it solves
AI agents need product-specific instructions to use NVIDIA software reliably across physical AI, robotics, simulation, CUDA-X libraries, RAG, and platform workflows. The catalog centralizes these skills while keeping them maintained in their respective product repositories and synchronized through an automated pipeline.
Key features
Agent-discoverable skill catalog — The NVI…
What problem it solves
AI agent skills need to be checked for quality, security, duplication, and their actual effect on agent behavior. Static checks alone cannot establish whether guidance improves task results, so evaluation must also support live comparison with and without a skill.
Key features
Three-tier evaluation pipeline — Independent tiers validate skill safety and structure, detect over…
What problem it solves
Running large language and vision-language models can require substantial setup and may depend on the available hardware. The project aims to make inference accessible across a wide range of local and cloud hardware with minimal setup.
Key features
Minimal-setup inference — The project provides installation options for common platforms, including scripts, Docker, pre-built b…
What problem it solves
Serving language models requires reading checkpoints, tokenizing requests, and running accelerated kernels while managing a local model server. TensorFold addresses these requirements with a native binary that can run on Apple Silicon and NVIDIA GPUs without requiring Python or MLX for serving.
Key features
Native Metal and CUDA inference — The Zig-based engine executes mode…
What problem it solves
Interactive multimodal and speech systems often require generation to span heterogeneous stages with different compute patterns, dependency structures, and resource needs. Serving these pipelines requires coordinated stage lifecycles, inter-stage transport, model integrations, and a unified API surface.
Key features
Coordinated multi-stage runtime — Models generation as coor…
What problem it solves
Data analysis workflows often require expressive query composition, flexible execution models, and interoperability with columnar data. Polars addresses these needs by allowing the DataFrame API and SQL to share the same optimizer and engine, with support for large workloads and custom extensions.
Key features
Unified DataFrame and SQL Queries — Queries can be written with e…
What problem it solves
Running frontier-scale open-weight models locally requires an inference platform that can make efficient use of the heterogeneous hardware available on a personal device. The project addresses this by treating GPUs, CPUs, host memory, and interconnects as a unified, elastic inference platform.
Key features
Edge-native MoE runtime — The runtime provides MoE serving with bandw…
What problem it solves
Autonomous AI agents often need to read files, install packages, call APIs, and use credentials, but unrestricted access can expose host data, secrets, and networks. OpenShell provides these capabilities within an isolated runtime where declared policies constrain agent access.
Key features
Kernel-level runtime enforcement — Agents run in isolated sandboxes where kernel cont…
What problem it solves
AI agent skills execute with implicit trust and often receive minimal vetting before installation. SkillSpector helps users assess whether a skill contains vulnerabilities, malicious patterns, or other security risks before installing it.
Key features
Multi-format scanning — Scan agent skills supplied as Git repositories, URLs, zip files, directories, or individual files.
So…
What problem it solves
Mosaics obscure portions of videos and images, making it difficult to view the affected content without changing the rest of the picture. Users need a restoration workflow that can identify selected mosaic regions and process them with GPU-accelerated video handling.
Key features
Targeted mosaic restoration — Detects mosaics in videos and restores them while leaving the rest…
What problem it solves
Deep learning models often need to be adapted for efficient deployment across different inference frameworks. Model Optimizer addresses this by providing a unified set of optimization techniques and export workflows for quantized checkpoints.
Key features
Post-training quantization — Compress models with quantization techniques available for Hugging Face, Megatron-Bridge, Di…
What problem it solves
Platform teams need to create and operate isolated Kubernetes clusters for customers on infrastructure they supply and control. They need a management plane that can provision GPU machines, manage cluster lifecycle operations, and establish node identity before issuing join credentials.
Key features
Independent workload clusters — Create workload clusters through the NVKE AP…
What problem it solves
Running AI models locally can require integrating multiple inference engines, model formats, hardware accelerators, and API surfaces. LocalAI addresses this with a small core, on-demand backends, and compatible APIs that work across supported hardware without requiring a GPU.
Key features
Composable inference backends — Backends are packaged separately and pulled only when a…
What problem it solves
LLM and Visual Gen inference requires specialized kernels, efficient runtime orchestration, and flexible deployment across NVIDIA GPU configurations. Developers also need a modular framework that supports customization while integrating with established inference tooling.
Key features
Optimized inference kernels — Custom kernels accelerate common inference operations, includ…
What problem it solves
Rust developers need host-side access to the CUDA runtime and tools for expressing GPU kernels while preserving Rust's ownership and borrowing rules. The platform provides separate programming models for workloads that require direct thread-level control and workloads naturally expressed through tensor operations.
Key features
Shared CUDA runtime — Host-side crates provide d…
What problem it solves
Agents can retrieve documents and query databases, but unreliable results undermine agentic analytics. Teams need a context layer that records where data lives, what its definitions mean, whether it is usable, how it should be queried, its provenance, and related concepts.
Key features
Enterprise ontology generation — Generate semantic ontologies from database metadata, infe…