NVIDIA AI Updates: October 17, 2026

1. RTX Spark PCs Go On Sale Starting at $2,599 With a Local Agent Stack

NVIDIA. RTX Spark Windows PCs are available globally as of October 16, starting at $2,599, from ASUS, Dell, HP, Lenovo, Microsoft (Surface Laptop Ultra), and MSI. The superchip pairs a Blackwell RTX GPU with a Grace CPU co-designed with MediaTek, with up to 128GB of unified memory and up to 1 petaflop of FP4 AI compute. The software bundle includes Nemotron 3.5 Bolt 74B-A9B in a 24GB 2-bit build, guided installers for agents such as GitHub Copilot, Hermes, OpenClaw, Perplexity, and Pi, sandboxing through NVIDIA OpenShell or Microsoft Execution Containers, and NVIDIA PAIR for serving inference to other machines on a home network. Source

2. NVIDIA Details the RTX Spark Developer Stack, Including PyTorch-CUDA for Windows on Arm

NVIDIA. NVIDIA’s developer guide for RTX Spark covers the first PyTorch-CUDA build for Windows on Arm, Python 3.15 as the native default, over 2,000 Conda packages, and Arm-compatible builds for more than 97% of the top 15,000 Python packages, alongside WSL and a Windows Developer Config built with Microsoft. Supported frameworks include llama.cpp, Ollama, vLLM, Unsloth, and LM Studio, and the open NVIDIA Skills repository teaches coding agents to use CUDA-X libraries. Against an Intel Core Ultra X9 388H baseline, NVIDIA reports 4.6x on a local coding agent benchmark (Qwen3.8-27B), 2.7x on Unsloth LoRA fine-tuning of Llama-3.1-8B, and 1.8x on a Linux kernel build. GitHub Copilot can use local models through Windows ML with a llama.cpp backend. RTX Spark small form factor desktops for always-on agents are slated for November. Source

3. Mercedes-Benz Expands Its NVIDIA Partnership to Hyperion, Alpamayo, and Level 4 Work

NVIDIA. Mercedes-Benz will build future automated driving on NVIDIA’s Hyperion compute and sensor architecture and the Alpamayo Stack, a full-stack driving software with AI reasoning, running alongside MB.OS for over-the-air updates. NVIDIA Halos adds a parallel safety stack for redundancy and guardrails. Target features include point-to-point urban assistance, active safety, and automated parking, and the companies are also developing SAE Level 4 applications such as robolimousines. No vehicle models or launch dates were given. Source

4. Volvo Cars Adopts Hyperion, Halos OS, and the Alpamayo Stack for Eyes-Off Driving

NVIDIA. Volvo Cars expanded its partnership with NVIDIA, which dates to 2017, to use the Hyperion sensor and compute architecture, Halos OS, and the Alpamayo Stack in an unnamed next-generation Volvo. The planned system targets door-to-door operation across highways, urban streets, and parking, with eyes-off and mind-off driving introduced gradually. NVIDIA did not give dates or compute figures. Source

5. Adobe Optimizes Photoshop and Premiere for RTX Spark and Builds Brand Concierge on Nemotron

NVIDIA. Adobe and NVIDIA expanded their collaboration: Photoshop, Premiere, and Substance 3D are optimized for RTX Spark with up to 2x faster AI processing, and Photoshop was rearchitected around GPU-accelerated compositing. Adobe’s Brand Concierge conversational AI is optimized on Nemotron open models and will reach enterprise customers later this year. Frame.io’s updated ingest pipeline uses NVDEC and TensorRT with CUDA, with Adobe reporting about 5x faster embedding generation, a 4-5x improvement in P95 processing latency, and an 80% smaller search proxy on early workloads. Source