Chips

US policymakers are debating whether to create regulatory risk around Chinese open-weight models. The release of Moonshot AI's Kimi K3 reignited the argument. Enterprises face not just performance questions but whether these models will remain easily accessible in a year.

NVIDIA has released Cosmos 3 Edge, a 4-billion-parameter open world model built to run on-device. It helps robots and vision AI agents understand surroundings, reason in real time, and generate robot actions locally. The Cosmos 3 family included Cosmos 3 Nano (16B) and Cosmos 3 Super (64B) shipped on May 31, 2026 at GTC Taipei. Edge is the third and smallest tier, at roughly one-sixteenth the size of Super. The problem is specific. Machines operate at the edge in factories, warehouses, and hospitals. They need data center–level performance on memory-constrained systems. Cosmos 3 Edge targets that gap. What does world model do here? A world model learns how an environment changes over time. It represents objects, motion, spatial relationships, and the effects of actions. Consider a robot reaching for an object. Recognizing the object is only the first step. The robot must also track where the object is, how its gripper moves, and what happens on contact. A world model reasons about these relationships. It can predict the visual result of an action, infer the action that caused a change, or generate an action to reach a goal. Cosmos 3 Edge brings these capabilities into one on-device model. Its shared representation lets a system understand the current world state, simulate possible futures, and connect those futures to actions. Two transformer towers, one shared representation. Cosmos 3 uses a Mixture-of-Transformers architecture with two towers, described in NVIDIA's technical report. The autoregressive tower processes vision and text tokens for understanding and reasoning. The diffusion tower processes vision, audio, and action tokens for prediction, generation, and neural simulation. The two towers keep separate normalization layers and multilayer perceptrons. They share multimodal attention layers, which align information across language, video, audio, and action. This lets the model reason about a scene before it generates an output. The attention pattern adapts to each modality. Language uses causal attention, where each token attends to earlier tokens. Diffusion tokens attend more broadly to the available context, supporting coherent prediction and generation. Depending on the task, the model emits reasoning tokens from the autoregressive tower, or denoised video and action tokens from the diffusion tower. Cosmos 3 Edge uses a 2B dense transformer for its reasoner, and follows Qwen3-VL-compatible message conventions for image and video inputs, per the Cosmos GitHub repository. One action representation across embodiments. Physical systems describe actions differently. A vehicle uses ego pose and movement. A camera uses camera motion. A robot arm uses the pose of its end effector, and a gripper adds grasp state. Cosmos 3 maps these embodiments into a common action representation. Actions are encoded as compact geometric vectors that capture translation, rotation, and manipulation state. This connects control to the visual structure of the world. The model associates pixel changes with physical motion and control inputs. Generated video then becomes more than a prediction. It represents how the world should change in response to an action. Supported action dimensions depend on the embodiment. The Cosmos GitHub repository lists camera motion (9D), autonomous vehicle (9D), egocentric motion (57D), single-arm robot (10D), dual-arm robot (20D), and humanoid robot (29D). Policy mode runs in both directions. As a policy, Cosmos 3 Edge predicts an action together with its expected visual consequence. Current state goes in; an action and its likely visual outcome come out. Action flows in both directions. The model can predict the effect of an action, or infer the action from its effect. This connects world modeling directly to robot policy training and evaluation. NVIDIA also released Cosmos 3 Edge Policy (DROID). It is a robot manipulation policy post-trained on the DROID dataset for pick-and-place tasks, with post-training scripts included. Developers can fine-tune on a small H100 cluster or an NVIDIA DGX Station before deployment. Is it Deployable? Cosmos 3 Edge delivers memory-efficient inference across NVIDIA edge computers. Targets include NVIDIA RTX PRO GPUs, NVIDIA DGX, GeForce RTX GPUs, and NVIDIA Jetson, including the newly announced Jetson T2000 and T3000 modules. As a post-trained world action model (WAM), the model operates at robot-control resolution of 640×360 observations. On NVIDIA Jetson Thor it generates 32 actions per inference, while achieving real-time control at 15 Hz. For generation, the Edge tier supports 256p and 480p resolutions, 12–30 fps, and 50–150 frames. Using the open Cosmos framework, developers can post-train Cosmos 3 Edge for a specific embodiment and sensor set in about a day. NVIDIA positions a GeForce RTX 3070 or better as a local on-ramp for prototyping. Key Takeaways: Cosmos 3 Edge is a 4B open world model (2B dense reasoner) that runs on-device, released July 20 on Hugging Face. A Mixture-of-Transformers design pairs an autoregressive reasoner tower with a diffusion generator tower through shared multimodal attention. It hits 640×360 control resolution, 32 actions per inference, and 15 Hz real-time control on NVIDIA Jetson Thor. Actions map to a common translation/rotation/manipulation representation, spanning camera, vehicle, single-arm, dual-arm, and humanoid embodiments. Benchmarks (#1 on VANTAGE-Bench at 4B) are internally claimed; the model ships under Linux Foundation OpenMDW-1.1. Sources: Hugging Face launch post, Cosmos3-Edge model card, Cosmos 3 collection, NVIDIA technical report (PDF), Cosmos GitHub, NVIDIA developer blog, Jetson Thor blog and NVIDIA Newsroom: Japan coalition. The post NVIDIA Releases Cosmos 3 Edge: A 4B-Parameter Open World Model That Reasons and Generates Robot Actions On-Device appeared first on MarkTechPost.

The release of Chinese open-source AI model Kimi has triggered a public feud within Trump's AI advisor circle. Kimi rivals the intelligence of paid models from OpenAI and Anthropic but is free, creating economic and political problems for the president. Factions disagree on whether to embrace open source or impose tighter controls.

France's abundant nuclear power provides low-cost electricity, giving it a competitive advantage in AI, but American tech giants may snap up that capacity first.

The AI industry faces two opposing strategies: spending billions on nuclear power to support larger models, and releasing free open-source models. The article argues these are two sides of the same bet on whether intelligence has a ceiling. Biology offers a precedent: the human brain under energy constraints achieved efficiency through architecture innovations like sparsity and in-memory computing. Current open models apply these tricks but only cover mid-level tasks, while frontier models maintain an edge. The outcome depends on where the ceiling sits.

This paper presents SpecLA, a speculative decoding runtime for stateful linear-attention models. It verifies chains and trees with topology-aware kernels, stores compact factors to recover accepted states, and uses confidence pruning plus a target-aligned EAGLE-style drafter. On an NVIDIA H100 with a GDN-1.3B target, SpecLA achieves up to 1.70x end-to-end speedup over autoregressive decoding.

OpenLanguageModel (OLM) is an open-source PyTorch library for building and pretraining small language models with visible internals. Model code mirrors architecture using components like Block, Residual, Repeat, and Parallel. It integrates tokenizers, datasets, optimization, mixed precision, callbacks, checkpoints, and hardware-aware execution, enabling seamless transition from teaching notebooks to full pretraining. The library includes 27 presets across nine model families and documentation from fundamentals to research. Validation shows close agreement with reference implementations, 90.6% weak-scaling efficiency on four GPUs for a 348M-parameter model, and positive usability feedback. OLM is MIT-licensed and available on PyPI, GitHub, and its documentation site.

This paper proposes an operator-aware mixed-precision tolerance calibration method that mines accumulated cloud GPU run data to automatically determine optimal absolute tolerances for tensor kernel correctness tests, achieving much tighter tolerances than hand-picked ones and significantly improving bug-detection recall.

This paper presents ARGO, a smart eyewear platform leveraging the STM32N6 microcontroller and its integrated NPU for on-device machine learning, minimizing latency and preserving privacy. Through holistic co-design of hardware, firmware, and AI, an optimized YOLOv11 model is deployed for real-time urban obstacle recognition, introducing Head-wise Parallel Attention (HPA) for efficient NPU execution. The model achieves mAP50-95 of 24 with only 2.483 MB memory footprint. The platform integrates multimodal sensors, runs at 10 FPS, and provides ~113 minutes of autonomy on a 200 mAh battery, demonstrating the feasibility of high-performance, privacy-preserving assistive devices.

Gary Marcus argues that China's Kimi K3 model has caught up with US top models, disrupting American AI business models. He recounts his warnings since 2025 that the US focus on LLMs would lead to a tie, not victory. Marcus proposes seven strategic options, ranging from inaction to making AI a global public good via an international 'CERN for AI' initiative.

daft-physical-ai is a new open-source Python library built on Daft that provides ready-made operations for turning raw robot recordings into training-ready data. It currently supports hand tracking and reward scoring, with lazy batch execution and distributed processing.

Dean Ball, OpenAI's chief of strategic futures, argues that open-weight models are inherently decelerationist because they deter capital expenditure. This article examines the economic logic behind the claim, considering China's capacity to build AI infrastructure, shifting training paradigms, and where value accrues in the AI stack. It concludes that open-weight models may actually be accelerationist by lowering costs and broadening participation, even if they disrupt current business models.

Italian engineer Vincenzo (JustVugg) created Colibrì, a proof-of-concept that runs the 744-billion-parameter GLM-5.2 model (1.5TB) on a modest CPU with only 25GB RAM and 1GB/s NVMe. Despite extremely slow speeds (0.05-0.1 tokens per second), it leverages the Mixture-of-Experts architecture to load experts per token, achieving frontier-level answer quality. The project is open-source and aims to explore running large models on consumer hardware.

This article explores the implications of SpaceX's fast-track inclusion into the Nasdaq-100 index for index fund investors. It discusses the mechanics of index funds, concerns about Elon Musk's governance, and whether investors should worry about AI concentration in the market.

Standard RAG excels at simple fact retrieval but fails on portfolio and counting questions. GraphRAG promised to solve this with knowledge graphs but is expensive and complex. The author presents a 'graph-like RAG' approach that uses ordinary databases, lazy summarization, and a fixed ontology, achieving similar benefits at a fraction of the cost and operational burden.

Founding members -- including Nvidia, Meta and Samsung -- aim to reduce their reliance on rare materials used in chips.

Stoke is a lightweight Rust gateway that enforces hard budget caps, loop detection, and rate limits on AI agent API calls before they reach the provider, preventing runaway spending. It also offers local-first routing across multiple machines, auto-routing with cost/speed optimization, and a fail-closed architecture.

This post shows how Amazon Quick can serve as the business-user front door for specialized agent workflows, using the NVIDIA NeMo Agent Toolkit to build a supply-chain risk example that helps planners move from dashboard and knowledge context to guided mitigation recommendations.

A newly opened San Francisco restaurant faced intense community backlash and vandalism after using AI-generated images on its menu display. The owners removed the images and are now planning community events to rebuild their reputation.

Moonshot AI released Kimi K3, a 2.8T parameter MoE model with open weights, ranking high on benchmarks. The article discusses the narrowing gap between Chinese and US AI models, China's commitment to open source, economic impacts of open models, and China's efficiency advantages.

The article argues that AI, deployed as a system rather than a collection of pilots, is the first technology to simultaneously address the three taxes of trust, time, and cost that prevent enterprises from acting on their data. It outlines three movements: turning dark data into signal, expanding the question space with probabilistic reasoning, and decoupling human capability from headcount. Governance is highlighted as the true competitive moat.

NVIDIA unveils major advancements at SIGGRAPH 2026, including MCP-connected AI agents for creative tools, a synthetic video detector NIM microservice, and the open-source Cosmos 3 Edge world model for edge physical AI. These innovations span content creation, media verification, and robotics.

The article analyzes how American AI companies' closed and proprietary model is losing ground to China's open strategy. AI models have little moat, and China's open release turns compute disadvantage into distribution advantage, closing the performance gap.

UK's AISI finds the gap between open and closed models on cybersecurity is shrinking. Chinese company Moonshot AI releases Kimi K3, a frontier-level model with some brittleness. Demis Hassabis proposes a FINRA-like standards body for AGI. Research shows LLMs can smuggle side channel tasks, evading monitors.

This article introduces a new series on running LLMs locally from Go. It explains inference as next-token prediction using frozen weights, the autoregressive loop, tokenization, the GGUF file format, and memory mapping for efficient loading.

Bristol Myers Squibb (BMS) is deploying its second NVIDIA DGX SuperPOD, built on eight DGX Vera Rubin NVL72 systems, making it the most powerful and energy-efficient AI cluster in life sciences. The new system will be accessible to every scientist, removing wait times and limits, and supporting AI applications across the drug discovery pipeline. BMS is also integrating the new cluster with its existing one to create a unified environment with AI-native tooling.

Chinese AI leaders Moonshot and Alibaba released models that claim to match top US systems at lower cost. Their open-source approach challenges US dominance and raises questions about the effectiveness of export controls and massive spending.

Airbus plans to migrate around 900 applications from AWS to French cloud provider Scaleway to assert digital sovereignty. Concerns over US data protection drive the move, but supply chain and AI remain challenges.

Moonshot AI’s Kimi K3, a 2.8-trillion-parameter open-weight model, uses mixture-of-experts, quantization, and attention caching to trade compute for memory, circumventing US chip restrictions. While it tops benchmarks in coding, deployment requires data-center infrastructure, pricing is high, and software support is incomplete.

Company aims to develop AI software that cuts research times and use of rare metals in chipmakers’ supply chains.

The article argues that by 2026, the main constraint for AI infrastructure will shift from GPU supply to power grid capacity. It projects 40% of AI data centers will be power-constrained by 2027, with approval timelines for new grid connections lasting 24-36 months. It provides detailed analysis of power demands, inference driving the power curve, and strategies like efficiency improvements, scheduling, and geographic distribution. It promotes Spheron's distributed GPU network as a workaround.

Moonshot AI's Kimi K3 tops the Frontend Code Arena benchmark, sparking debate on US AI regulation. While strong on frontend coding, it lags behind Claude Fable 5 overall. The release intensifies the policy fight over AI regulation and open-weight models, with implications for stocks like NVIDIA and Anthropic.

Thinking Machines Lab has released Inkling, its first general-purpose open-weights foundation model. It is a multimodal MoE model with 975B parameters, 41B active parameters, and a 1M-token context window. Designed for customization, Inkling excels in reasoning, coding, agentic workflows, and multimodal tasks. This guide covers its architecture, training, benchmarks, deployment, and fine-tuning workflow.

Agentic AI systems can autonomously plan, decide, and act to achieve goals, marking a shift from reactive chatbots to proactive digital teammates. This article covers its definition, driving factors, use cases, challenges, and infrastructure needs.

In mission-critical scenarios like disaster inspection and search-and-rescue, communication-limited robots must make reliable onboard decisions. Episodic memory reuse, though low-cost, can be unsafe due to changed topology or insufficient resources, leading to 'memory traps'. This paper presents MemoGuard, a lightweight adaptive runtime that validates memories against topology, resource, and outcome contracts before reuse, invoking fallback only when validation fails. In a corridor-inspection simulator, MemoGuard reduces battery safety violations by 76.6% over similarity-only top-1 reuse and reduces fallback calls by 21.4% over always reasoning. On an NVIDIA Jetson AGX Xavier with local llama3.2:3b fallback, it avoids 3.67 s and 36.97 J overhead per trial.

A Partial Information Decomposition (PID) framework is used to select the most informative MRI contrast pair prior to training, reducing computational cost for multi-contrast 3D brain tumor segmentation. Applied to T1n, T1c, T2w, and T2-FLAIR, it selected T1c+T2-FLAIR, which achieved mean Dice 0.676 vs 0.687 for all four inputs on lightweight 3D U-Nets.

The article details how Silicon Valley's embrace of AI is leading to widespread layoffs and a sense of precarity among tech workers, who once enjoyed immense job security and high salaries. Through personal stories and expert analysis, it illustrates the shift toward automation and its impact on the workforce.

A single 24GB GPU is the practical floor for serious local inference. This guide compares six open-weight models that fit one card at Q4_K_M, including Qwen3.6, Gemma 4, Mistral Small, gpt-oss-20b, and DeepSeek-R1-Distill. It covers VRAM fit, licensing, and the job each does best.

Headroom is a 30-second browser-based test that measures your GPU's memory bandwidth ceiling for local AI inference, requiring no model download, using synthetic weights, and needing WebGPU support. It measures bandwidth via three independent methods and detects a subgroup bug that can cause silent garbage output in browser LLMs. Validations show in-browser engines are launch-overhead-bound, with 2-3x hardware headroom.

Together AI and Y Combinator have partnered to provide YC startups with a dedicated GPU cluster, addressing the compute bottleneck. Startups can commit for weeks instead of two-year contracts, and manage GPUs directly through a self-service portal without YC involvement. The cluster is fully utilized and supports needs from single-node to large-scale expansion.

A curated collection of free resources for AI researchers, including over $1000 in free compute credits, research guides, community forums, and more to help students start AI research without spending a dollar.

In an exclusive interview with CNBC, TSMC CFO Wendell Huang said the company is accelerating its Arizona factory expansion to meet multi-year structural AI demand. An additional $100 billion investment raises total U.S. commitment to $265 billion, with full-year capex revised up to $60-64 billion. TSMC is converting 5nm capacity to 3nm, and 2nm technology will drive next quarter's revenue.

Feyn Labs has released SQRL, a family of text-to-SQL models that inspect a database with read-only probes before committing to a query. The flagship SQRL-35B-A3B reports 70.6% execution accuracy on BIRD Dev, edging Claude Opus 4.6, and distills into self-hostable 4B and 9B checkpoints.

Alibaba's Qwen team previewed Qwen3.8-Max-Preview, a 2.4 trillion-parameter multimodal MoE model it calls "second only to Fable 5." The preview is live on Token Plan, Qoder, and QoderWork at 10% of standard pricing. What is not live: any benchmark table, model card, license, per-token price, or active-parameter count. This breakdown separates what Alibaba confirmed from what it only claimed.

A Forbes article profiles luxury escorts in Silicon Valley who are paid for their knowledge of AI and crypto. However, this is not a new phenomenon—similar roles have existed for centuries in the form of geishas, courtesans, and salon hostesses. The article argues that the human desire for intellectual companionship long predates AI.

Moonshot AI temporarily halts new subscriptions after unexpected demand for Kimi K3 pushes GPU capacity to the limit.

This week's developments in AI shift focus from raw scale to distribution and openness. Highlights include Thinking Machines' open-weight Inkling, Moonshot AI's 2.8T parameter Kimi K3, PrismML's phone-runnable Bonsai 27B, OpenAI's self-play red-teaming system GPT-Red, and Xi Jinping's call for open-source AI as a global public good at the World AI Conference in Shanghai.

Artificial intelligence is evolving from rule-based programming (Software 1.0) to learned models (Software 2.0) and now to natural language interfaces with emergent reasoning (Software 3.0). LLMs are a general-purpose technology with broad impact, but they lack physical grounding. The next frontier is spatial intelligence and physical AI, driving autonomous vehicles and humanoids.

Zlvox is a platform offering 25 free developer tools covering AI, JSON, PDF, image processing, and more, with a strong emphasis on privacy (client-side processing, zero data retention) and performance. No sign-up required.

Three Chinese labs' flagship open-weight MoE models—Kimi K3, DeepSeek V4 Pro, and GLM-5.2—each excel in benchmarks, licensing, and cost. Kimi K3 leads in capability but is API-only; DeepSeek V4 Pro is cheapest and fully open; GLM-5.2 balances speed and deployability.