Nvidia Jetson AGX Thors 7.5X FASTER AI Agent Brain For Robots Does This ($3499)

2’s Helix AI to process its vision-language action model in real-time even faster. And on top of handling massive datasets and multi-billion parameter models, as tasks or environments become more challenging, Fig. 2’s Helix can adapt in real-time. This allows the robot to improve fine motor skills and cognitive decision-making, such as when identifying, manipulating, and neatly folding various types of clothing completely autonomously, all with increasing speed. But this is only the beginning as NVIDIA also just released its new Jetson AGX Thor developer kit for AI robots as part of their three-computer solution to accelerate the adoption of physical AI.

This ecosystem consists of NVIDIA DGX for training foundational AI models, NVIDIA Omniverse for generating synthetic data and simulating real-world environments, and the all-new NVIDIA Jetson AGX Thor developer kit for runtime robotics. Under the hood, it’s powered by NVIDIA’s Blackwell GPU and 128 GB of memory, capable of delivering up to 2070 FP4 Terra floating-point operations per second of AI compute. That means the developer kit can effortlessly handle even the most advanced generative AI models and real-time robotics workloads. But it also runs the full NVIDIA AI software stack, which encompasses NVIDIA Isaac for robotics, NVIDIA Metropolis for video analytics, and NVIDIA Holoscan for sensor processing.

This unified approach ensures a seamless cloud-to-edge workflow, empowering developers to accelerate physical AI projects from simulation to deployment. In fact, AGX Thor delivers up to seven and a half times the performance compared to the previous AGX Orin, supporting modern VLA models like NVIDIA’s own Isaac Groot N1, as well as large language and vision language models widely used in AI research. Developers can even fine-tune these models with synthetic data generated in Omniverse, and then deploy them directly onto Jetson AGX Thor for real-world testing and operation. To illustrate this, NVIDIA demonstrated a full hardware-in-the-loop workflow with Isaac Groot N1 using Omniverse and then deployed to Jetson Thor.

This setup is also designed to support advanced video analytics at the edge. For example, NVIDIA’s video search and summarization tool, part of the Metropolis platform, can be run locally on Jetson Thor. This enables real-time contextual alerts, video summarization, and question-and-answer capabilities for live camera feeds. Plus, the NVIDIA Holoscan platform streamlines real-time AI at the edge even further. With features like the Holoscan SensorBridge, developers can connect high-throughput sensors, like Ethernet cameras streaming Ultra HD video, directly to the GPU, minimizing CPU usage and latency. This allows for immediate inference by edge-deployed AI models, powering applications ranging from industrial inspection to autonomous robotics.

As for parts, Jetson AGX Thor Developer Kit includes a Jetson T5000 module, a reference carrier board, an active heatsink and fan for cooling, as well as an external power supply. For connectivity, it features two USB-A ports, an Ethernet port, a display port, an HDMI port, two USB-C ports, a QSFP slot, and a microfit port for extending power if necessary. As for storage, it comes with a 1TB NVMe drive installed in an M.2-key M slot, plus a wireless networking card pre-installed in the M.2-key E slot. It also features physical controls, including dedicated buttons for power, force recovery, and reset.

And in generative AI video editing, Google just announced that its essential video editing tools within Google Vids are now free for all users, removing the need for a Gemini subscription. This means that anyone can now create videos using templates, add text overlays, and utilize animation effects directly within the platform. Plus, a wave of new features is arriving. Users can now convert photos into 8-second video clips with audio narration powered by the VO3 model. What’s more is that Google Vids introduces AI avatars designed for scripted presentations, alongside automatic audio cleanup and real-time transcription for improved clarity and accessibility.

Furthermore, Google has stated that additional options, such as portrait video formats, creative filters, and expanded background selections, will be available soon. These upgrades are aimed at content creators producing social media posts, YouTube intros, and training materials. And to assist newcomers, the platform includes the Vids on Vids learning series, offering step-by-step guidance through the video creation process. Simultaneously, Google DeepMind just released its latest nano-banana image editing model, which is now fully integrated into the Gemini app. In fact, early feedback has highlighted this tool as one of the most advanced in the world, especially when it comes to maintaining a subject’s true appearance through multiple edits.

The focus here is on subtle consistency, so photos of friends, family, or even pets retain their unique look, whether you’re adding a new hairstyle, or imagining a pet in a costume. Plus, Gemini now allows users to blend photos, swap backgrounds, and experiment with multi-stage edits, like redecorating a room or updating a wardrobe, all while preserving the main subject’s likeness. Another new feature is design mixing, which lets you apply the texture or color from one object to another for even more creativity. As for availability, both paid and free users worldwide can access these tools starting now. Furthermore, every image created or edited in Gemini includes a clear watermark, plus an invisible digital marker to ensure transparency regarding AI generation.

And in Agenic AI News, Anthropic is piloting Claude for Chrome, its AI-powered browser assistant, starting out with 1,000 Max Plan users. This early version helps manage calendars, draft emails, and test website features. However, prompt injection attacks, such as hidden commands on web pages, remain a concern, but new safeguards have cut attack success rates in half. Anthropic says the insights collected from this controlled rollout will guide their future safety improvements before a broader release. And finally, Google just rolled out Gemini 2.5 Flash Image, bringing its advanced generative image technology directly to Adobe Firefly and Adobe Express users.

With this integration, creators can now quickly produce stylized graphics, explore multiple visual directions, and easily transition their work into professional tools like Photoshop or Illustrator for further editing. On top of this, Adobe assures users that any content generated will not be used for training future AI models to maintain user privacy and transparency. For now, paid Adobe Firefly and Creative Cloud Pro members can enjoy unlimited generations with Gemini 2.5 Flash Image until September 1st, while new Firefly users on the free plan can access up to 20 generations. Anyways, check out this video here for more of the latest NVIDIA AI news.