NVIDIA just launched the Nemotron Coalition, teaming up with Mistral AI, Cursor, Perplexity, LangChain, and other top AI labs to build open frontier foundation models. Their first milestone is a joint base model co-developed with Mistral AI on DGX Cloud, which will be open-sourced to underpin the upcoming Nemotron 4 model family.
Perplexity just integrated NVIDIA’s Nemotron 3 Super across its entire product suite, rolling it out to the main app, Agent API, and Perplexity Computer. The addition gives both search users and developers building automated agent workflows immediate access to NVIDIA's high-performance model alongside existing options.
Nvidia just released Nemotron 3 Super, an open 120B-parameter model that activates only 12B parameters at inference to deliver 5x higher throughput for agentic workflows. It combines Mamba layers, multi-token prediction, and a 1-million-token context window to prevent goal drift across long tasks. It essentially solves the massive context explosion and thinking tax that hold multi-agent systems back.
NVIDIA just rolled out major updates for ComfyUI, adding a simplified App View that hides complex node graphs behind a clean prompt box. Creators also get an RTX Video Super Resolution node delivering 30x faster 4K upscaling, alongside NVFP4 precision on RTX 50 Series GPUs that slashes VRAM use by 60%. Why it matters: Running production-grade AI video locally on consumer hardware is becoming dramatically more practical.
NVIDIA just made a major investment in Thinking Machines Lab, the new frontier AI startup founded by Mira Murati. The multiyear deal plans to deploy at least one gigawatt of next-gen Vera Rubin AI systems targeted for early next year to power frontier model training and serving. The AI compute arms race is officially entering the gigawatt era.
Nvidia is planning to launch NemoClaw, an open-source AI agent platform designed for enterprise software companies, according to Wired. The platform packs security and privacy controls to let companies deploy task-handling agents—and it won't require Nvidia chips to run. Why it matters: Jensen Huang called open-source agent "claws" the most important software release ever, and now Nvidia wants to make them safe enough for corporate boardrooms.
Nvidia CEO Jensen Huang just crowned open-source AI agent framework OpenClaw as "the single most important release of software ever." Speaking at an investor conference, Huang noted that while Linux took 30 years to reach its scale, OpenClaw surpassed it in just three weeks. Why it matters: Agentic workflows are poised to drive inference token consumption up 1,000x, triggering a massive compute surge.
Jensen Huang just said Nvidia’s recent checks into OpenAI and Anthropic are likely its last, pointing to upcoming IPOs closing the investment window. The pullback follows Nvidia trimming its OpenAI pledge to $30B and Anthropic landing on a Pentagon blacklist. Why it matters: As its key customers clash and face mounting political fallout, Nvidia appears to be backing out of an increasingly messy web of alliances.
OpenAI just announced a massive $110B funding round at a $730B pre-money valuation, backed by Amazon, NVIDIA, and SoftBank. The company also revealed ChatGPT now commands over 900M weekly active users and 50M paying subscribers, securing gigawatts of compute to fuel its next phase.
NVIDIA just released its second annual State of AI in Healthcare survey, revealing that 70% of organizations now actively deploy AI. A striking 85% of executives report direct revenue gains, led by payoffs in medical imaging and drug discovery. Why it matters: Healthcare AI has officially crossed from pilot experiments into revenue-generating clinical infrastructure, with 82% relying on open source to power it.
A senior Trump administration official claims Chinese startup DeepSeek trained its upcoming model on banned Nvidia Blackwell chips, reportedly clustered in an Inner Mongolia data center. Officials also suspect the firm will scrub technical markers of chip usage. The leak further inflames Washington's bitter debate over export control enforcement.
Taalas just introduced the HC1, an AI accelerator that bakes the Llama 3.1 8B model directly into hardware to generate a blistering 17,000 tokens per second. By merging compute and storage on a single chip, it claims 40 times the speed of Nvidia's B200 at a fraction of the cost, though it is permanently tied to that specific model.
Meta just announced a multi-year partnership with NVIDIA to overhaul its AI data center roadmap, securing NVIDIA's next-gen Vera Rubin platform for cutting-edge clusters. The deal also rolls out Spectrum-X Ethernet across Meta's infrastructure and brings confidential computing to WhatsApp private chats. Zuckerberg says the endgame is delivering personal superintelligence to everyone.
NVIDIA just introduced LoRAWeave, an image editing approach that learns transformations from visual examples instead of text prompts. Users provide a before-and-after pair alongside a target image, allowing the model to dynamically blend LoRA weights and replicate complex styles like claymation or sketches. It makes tricky, hard-to-prompt visual edits far more precise.
NVIDIA just dropped a guide on running the popular OpenClaw AI agent entirely for free on GeForce RTX GPUs and DGX Spark hardware. By pairing local runners like Ollama or LM Studio with open models, users can let an always-on agent manage local files and schedules without bleeding cash on cloud API tokens or handing over private data. It brings powerful autonomous desktop workflows right to consumer hardware.
Monarch RT just brought real-time AI video generation to consumer hardware, clocking 16 FPS on a single Nvidia RTX 5090. Its structured Monarch matrix attention runs up to 4x faster than standard FlashAttention while retaining about 95% of full-dense visual quality. Why it matters: developers no longer need multi-GPU server clusters to run live video diffusion models.
Nvidia researchers just introduced Dynamic Memory Sparsification (DMS), a technique that compresses LLM KV cache by up to 8x without degrading accuracy. Instead of rigid deletion heuristics, DMS trains models like Llama 3 or Qwen 3 to intelligently decide which tokens to drop using a delayed-eviction window. The feature is available in Nvidia's Model Optimizer framework.
Nvidia and university researchers just released DreamDojo, a robot world model trained on 44,000 hours of egocentric human video. Instead of collecting endless, costly physical demonstrations, robots learn basic physical intuition by watching humans before fine-tuning for specific hardware. It also runs in real time at 10 FPS, making pre-deployment simulations far easier.
OpenAI just launched GPT-5.3-Codex, an agentic model that is 25% faster and sets high marks on SWE-Bench Pro. Remarkably, the Codex team used early versions of the model to debug its own training run and manage deployment. It turns coding into an interactive, real-time collaboration that spans the entire software lifecycle.
Intel just announced plans to produce graphics processing units (GPUs), venturing straight into Nvidia’s dominant territory. CEO Lip-Bu Tan revealed the move at the Cisco AI Summit, emphasizing the effort will be tailored around customer demand. Intel has already recruited veteran engineers like former Qualcomm executive Eric Demers to help drive the turnaround effort.
