Titan Media and 36Kr report that S&P moved Meta's AA- outlook to negative, citing heavy AI infrastructure spending and uncertainty over monetization. They also say Microsoft, Alphabet, Amazon, Meta and Oracle generated $542.9 billion in operating cash flow but spent $556.9 billion on capital expenditures, turning combined free cash flow negative and cutting share repurchases by 83%. For Anthropic, 36Kr cites KBRA's BB+ issuer rating and a revised Blackstone expectation that the company may remain speculative grade even after an IPO, despite a reported $63 billion compute-financing arrangement.
Nvidia publicly presented Optics-on-Interposer (OOI), a next-generation co-packaged optics approach, at SEMICON West 2026, according to 36Kr’s report citing Cailian Press. OOI integrates the optical engine directly on the silicon interposer carrying the GPU and uses a larger interposer in place of the current organic package substrate. Nvidia researcher Nikola Nedovic said prototype samples were available, but the technology remains in research with no commercialization timetable.
Meta and Fourier reportedly built the Max humanoid robot prototype in about three months, according to The Information and outlets citing it. The approximately 1.5-meter, 48-kilogram robot uses Nvidia’s Jetson Thor chip and includes an expression screen, speakers and five-finger hands; Meta says it is intended for internal research with no plans to sell the units. Development of Max 2 began in September, with a redesigned torso, improved hands, a projector and additional sensors for stairs planned.
The open-source nv-uefi-unlocker project runs from a bootable USB drive before the NVIDIA driver loads, temporarily removing the Tensor Core limiter on selected RTX 20-, 30- and 40-series desktop and laptop GPUs without flashing VBIOS or patching the driver. Reported Windows tests include a 65% FP16 increase on an RTX 4060 Laptop and a 67–75% increase on an RTX 2070, but the tests differ by card and should not be compared directly. Some RTX 20-series models have known boot or handoff issues, and RTX 50-series GPUs are unsupported. Power, thermal limits and real-world AI gains vary.
Darkbloom told members of its decentralized AI inference network that falling token prices could reduce revenue from spare Mac compute capacity. The network estimated that an M5 Pro Mac Mini with 64GB of unified memory could earn up to $192 monthly under the best-case scenario, but about $36.77 after accounting for a reported 77% decline in Anthropic’s price per million tokens. Darkbloom is analyzing model demand, exploring workloads beyond inference, improving network performance, and seeking clients outside OpenRouter.
NVIDIA NeMo Speech 3.0 is available as release v3.0.0 and through the 26.07.00 NeMo Speech NGC container, according to the NVIDIA-NeMo/Speech repository. The framework supports speech AI, including automatic speech recognition and text-to-speech, along with multimodal and language-model development. The repository also points to a Hugging Face collection of open-weight checkpoints and demos.
IT之家, citing Reuters, reported that the Nasdaq 100 lost more than 5.8% in October as investors questioned when heavy AI infrastructure spending would pay off, with projected market-cap losses of over $2 trillion. Meta, Alphabet and Amazon raised spending plans, while Meta reportedly sought a record $97.5 billion in financing and Microsoft planned about $25 billion in investment-grade bonds. Strong revenue and cloud growth at Amazon, Meta and Microsoft showed that returns were uneven rather than universally deteriorating.
According to TMTPost, Horowitz Andreessen Academy (HAA), backed by Andreessen Horowitz (a16z), plans to enroll about 50 high-school graduates and first- or second-year university students in a free one-year program. The program does not award a diploma or university credit. Students would receive project budgets, compute credits and short courses from leaders including Jensen Huang, Sam Altman, Satya Nadella and Fei-Fei Li, followed by at least three months of full-time work at partner companies. HAA is not currently accredited, and TMTPost says the program’s ability to improve student outcomes remains uncertain.
Apple reclaimed the world’s highest market capitalization from Nvidia on Friday, according to the supplied reports. Apple’s closing market value was $5.03 trillion, compared with Nvidia’s $4.93 trillion. The reports said investors sought the perceived safety of the iPhone maker amid concerns about artificial intelligence spending.
NVIDIA opened the model weights for Nemotron 3.5 Transcribe, an open-weight MoE speech model with about 34 billion total parameters and 4.3 billion active per forward pass, according to TMTPost. The model combines transcription, speaker attribution, translation, word-level timestamps and audio question answering, supporting 40 transcription languages and as many as eight speakers in one hour of audio under the reported OpenMDW-1.1 license.
Intel shares closed down nearly 29% at $65.76 after the company forecast quarterly revenue of $15 billion to $16 billion, below analysts' roughly $17 billion expectation, TechSpot reported. Intel also warned that it would not generate positive cash flow this year, citing weak PC demand, rising memory costs linked to AI data-center demand, and the shutdown of its Kiryat Gat fab. TechNews and cnBeta reported declines of about 30% and 28.58%, respectively, alongside a broader semiconductor sell-off.
NVIDIA exhibited its latest NVLink switch board at the October OCP summit and is considering PTFE for high-speed signal transmission, PCB industry analysts told TechNews. The reported plans include introducing low-dielectric-loss PTFE into the Rubin Ultra switch board or a next-generation orthogonal midplane to reduce signal attenuation. PTFE’s low dielectric loss could support high-speed PCB design, but its poor compatibility with other materials and difficulty forming multilayer boards require a mixed-material approach; the report says PTFE alone cannot be used to make dozens-of-layers AI server boards.
NVIDIA Technical Blog details how NVIDIA AI Cluster Runtime (AICR) packages NVIDIA Dynamo and its supporting Kubernetes infrastructure into versioned, validated recipes. The Dynamo recipe pins the dynamo-platform chart to version 1.4.2 and defines component versions, compatibility constraints, deployment checks, and conformance checks covering GPUs, scheduling, observability, and ingress. The walkthrough also compares Dynamo Frontend and Gateway API/EPP routing and explains the planned native Rust EPP path in Dynamo 1.5.
NVIDIA says the 64GB DGX Spark is available for embedded developers prototyping, fine-tuning, and testing models and agents. Its technical tutorial shows how the trtedge CLI and TensorRT Edge-LLM run NVIDIA Nemotron-3.5 Lightning on DGX Spark, expose it through an OpenAI-compatible API, and optionally enable DSpark speculative decoding. NVIDIA reports that DSpark raised mean generation speed from roughly 73 tokens per second to 105–127 tokens per second across three tested workloads, with peak total memory use of 27.9GB.
Creator MonsterHouse built an interactive Halloween projection that listens to trick-or-treaters and responds as a grumpy pirate ship, becoming friendlier when visitors are polite. The system combines NVIDIA Nemotron Voice Agent, Nemotron Speech ASR, Nemotron 3.5 Bolt E4B 4GB, Magpie TTS Zeroshot and Audio2Face-3D, with the conversational and animation workload running on a single workstation. Its pipeline also uses LoRA for personality, Nemotron 3 Nano Omni for costume descriptions, and YOLO with ByteTrack for visitor counting and tracking.
NVIDIA introduced NemoClaw, a set of open blueprints that combine tuned open models, agent harnesses, optimization and telemetry tools, and OpenShell runtime governance. The blueprints support OpenClaw, Hermes Agent, and LangChain Deep Agents, allowing developers to assemble specialized agents on infrastructure they can own and customize. NVIDIA says its tuned Nemotron 3 Ultra configuration for LangChain Deep Agents reached near-frontier performance at roughly one-tenth the per-run cost of leading closed models, but that claim is vendor-reported.
A Chinese review of Joanna Stern’s debut book, “I Am Not a Robot: My Year Using AI to Do (Almost) Everything,” examines her yearlong experiment delegating chores, planning, learning, writing and other activities to AI and robots. The review highlights both useful capabilities—information filtering, factual analysis and planning—and failures involving physical manipulation, intent recognition, hallucinations and weakened human creativity. It also discusses the book’s reported appearance on The New York Times bestseller list.
U.S. technology shares fell sharply as investors questioned the returns on enormous AI infrastructure budgets, weaker free cash flow and rising financing needs. According to BBC Technology, TechCrunch AI and IT之家, Alphabet lost more than $500 billion after a 12% plunge, Intel dropped nearly 30%, and the Nasdaq fell about 3.5%, while Nvidia, Broadcom and Qualcomm also declined. Alphabet, Amazon, Meta and Microsoft outlined more than $740 billion in 2026 capital expenditures, although some investors described the selloff as an ordinary correction after unusually large gains.
NVIDIA introduced an optional agent skill for gds-diag, an open-source command-line utility that checks GPUDirect Storage environments, identifies problems, and suggests improvements. The skill adds natural-language routing, multi-step diagnostic orchestration, and model-guided interpretation of results, while the underlying tool provides deterministic checks and JSON output.
PsiQuantum announced an optical circuit switch prototype for AI data-center networks, built from its low-loss silicon photonics and designed to pass light between fibers without electrical conversion. The company says its 8×8 prototype typically loses 1.2 dB in lab tests without an amplifier and reconfigures in under 3 milliseconds. Tom’s Hardware reported that the performance figures have not been independently verified and that the device remains a small prototype.
IT之家 reported that China’s securities regulator accepted the materials for SiliconFlow’s overseas listing filing, a procedural step that does not constitute final approval or completion of its proposed Hong Kong listing. SiliconFlow says its platform uses its own inference engine to schedule chips from Nvidia, Huawei Ascend, MetaX and Moore Threads for models including DeepSeek, Qwen and Kimi. Separately, 36氪 reported that the company’s public-cloud business had a 2025 gross margin of -119%, while its serverless Token service generated about RMB 14.3 million in revenue and had not achieved sustained profitability.
Horizon Robotics founder and CEO Yu Kai criticized industry claims that label autonomous-driving chip compute at six to seven times dense compute. He proposed that Horizon products disclose two figures: “NV compute,” defined as dense compute multiplied by two, and “equivalent compute.” Under figures he provided, the Journey 6P is rated at 560 TOPS for NV compute and 1,960 TOPS for equivalent compute. Commentary linked the dispute to Newchip’s X7 Ultra launch, which was announced with 700 TOPS of large-model compute and scaled delivery based on company disclosures.
Tom's Hardware published a primer explaining core LLM concepts, including models, weights, tokens, quantization, context windows, KV cache, inference, and speculative decoding. It focuses on text LLMs used for chat, programming, and agentic bots, and frames the material as a reference for understanding local-model hardware such as NVIDIA RTX Spark.
SiliconANGLE reports that enterprises deploying AI agents must enforce each user’s data permissions, making role-based access control central to whether agents can be trusted. Cloudera positions Cloudera Anywhere Cloud as a hybrid data and AI platform spanning public clouds and on-premises environments. Its packaged blueprints and expanded partnerships with Mistral AI, Nvidia and Vast Data target private AI use cases such as fraud detection. Chief product officer Leo Brunnick said privacy concerns, sovereignty requirements and rising inference costs are pulling more AI workloads into corporate data centers.
IT之家 reports, citing Blue Whale Automotive and people familiar with the matter, that Horizon Robotics' high-compute chip and HSD urban assisted-driving system have entered BYD's higher-tier Tian Shen Zhi Yan B program, a step beyond its earlier chip-only role in the Tian Shen Zhi Yan C program. Under the reported arrangement, Horizon supplies a large-compute chip plus HSD, while BYD-developed chips and Nvidia supply remain, with Nvidia's share expected to shrink. The chip used in the B program is described as a distinctive part number with confirmed high compute, benchmarked against the single Nvidia Orin-X (254 TOPS) typically fitted to B-program vehicles, and Huaxing Securities research said the first Horizon-equipped B-program model launches in November.
36Kr reported, citing an announcement from Jin An Guoji Group Co., Ltd., that the company had not contacted NVIDIA or Huawei or directly conducted business cooperation with them as of the announcement. The company issued the disclosure in a stock-trading abnormal-volatility announcement after its closing-price increase deviation reached a cumulative 20% over three consecutive trading days. It said its high-frequency, high-speed copper-clad laminate products remained in development and customer sampling, had generated no revenue and carried uncertainty in their development progress and results.
Tom’s Hardware found Nvidia’s standalone Vera data-center CPU delivered chart-topping Python performance and strong socket- and system-level memory bandwidth, aided by its C2C interconnect. Vera also matched or beat comparable AMD Turin and Intel Granite Rapids-AP chips in several general-purpose server workloads, though x86 processors retained an advantage in core density and some frontend tasks. The results indicate competitiveness for AI-adjacent workloads, but no standardized end-to-end agentic CPU benchmarks yet exist.
The Rundown AI, citing The New York Times, reports that U.S. intelligence alleged Alice Huang, an executive at Singapore-based data center firm Megaspeed, passed intelligence to Beijing while maintaining close ties to Nvidia CEO Jensen Huang. Trump was reportedly briefed about the allegations about a year ago, but the White House did not act, and officials said they did not believe Jensen Huang knew of Huang's alleged role. Nvidia said the meetings were routine, while Megaspeed said it is cooperating with the U.S. and China called the allegations groundless.
The Financial Times reports that Nscale’s Nvidia-backed neocloud project, the 2,250-acre, 8GW Monarch Compute Campus in West Virginia, has been hit by construction issues, local opposition and financing risks. The supplied report does not provide further details about the scope or status of these problems, and no responses from Nscale or Nvidia are included.
OpenAI CEO Sam Altman met Foxconn chairman Young Liu in Taiwan, Liu confirmed at the company’s technology event. Liu said the companies’ AI infrastructure work has made progress and their next collaboration will become more concrete, though he did not disclose meeting details, a new agreement or any commercial order. OpenAI has not commented.
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…