NVIDIA Announces RTX Spark PCs and Native AI Solutions at IFA 2026

NVIDIA announced the NVIDIA PAIR router developed for local artificial intelligence agents and RTX Spark Windows PCs offering 1 Petaflop power.

The goal of freeing artificial intelligence from cloud dependency and bringing it directly to our personal computers is gaining momentum. Making important announcements at the IFA 2026 fair, NVIDIA shared its new ecosystem that will make local artificial intelligence agents much more accessible and powerful for everyone. Company; officially announced the open source NVIDIA PAIR router, which brings together the idle processing power of computers in the home or office network, the next generation RTX Spark offering 1 Petaflop power, and huge performance updates for Windows computers and popular open source inference engines.

What is NVIDIA PAIR? Idle Computers Are Turning into Artificial Intelligence Network

The biggest bottleneck faced by those working with native AI agents is queuing complex, multi-step tasks on a single graphics card. NVIDIA has made available the open source and free NVIDIA PAIR (Personal AI Router) tool, which it developed to solve this problem, as beta. PAIR automatically discovers compatible devices on your local network via mDNS and intelligently allocates the individual subtasks of the AI ​​agent you are working on to other suitable devices on the network.

On the security side, the system uses mutual TLS (mTLS) and local certificates to ensure that data traffic between devices remains completely on the local network and encrypted. The tool integrates with popular open source inference engines such as LM Studio and Ollama; It supports Windows, macOS and Linux platforms. On the hardware side, GeForce RTX 20 series and above GPUs, Turing and later RTX PRO workstations, DGX Spark and even Apple M4 or newer Apple Silicon processors can be included in the network pool.

1 Petaflop Power and 128 GB Memory: RTX Spark is Coming to Windows Computers

NVIDIA's most notable move on the hardware side was the RTX Spark platform, designed directly for personal artificial intelligence agents. This new computer category, which will start to take its place on the shelves next October, ranges from compact desktop designs to thin and light laptops.

RTX Spark Hardware Features and Featured Details

The RTX Spark platform aims to transform personal computers from a simple work tool into a proactive artificial intelligence assistant:

Processing Power: Blackwell architecture RTX GPU offering 1 Petaflop AI processing performance.

Central Processor: Highly energy efficient 20-core Grace CPU.

Unified Memory: Thanks to the high-bandwidth unified memory architecture that can be scaled up to 128 GB, local language models with huge parameters can be easily run.

Manufacturer Support: While Acer exhibited its compact desktop concept at IFA 2026, Lenovo officially announced the Yoga Pro 9n and Yoga 9n 2-in-1 models for October.

90 Percent Speed Increase in llama.cpp Extraction with RTX 5090

NVIDIA also shared the fruits of its optimization efforts with the open source community in figures. Thanks to optimizations at the kernel level, advanced speculative decoding, and new XQA attention cores on FlashInfer, llama.cpp inference efficiency increases up to 1.9 times (90 percent increase) on the GeForce RTX 5090 graphics card. Similarly, in the vLLM infrastructure, speed gains of up to 1.2 times are achieved in RTX PRO 6000 workstations with Blackwell architecture and up to 1.4 times in DGX Spark clusters.

While one-click installation support is available on Windows for Hermes Agent, which was developed in collaboration with Nous Research to eliminate local installation difficulties, tools such as Perplexity's Portable Computer agent and OpenClaw, which has become popular in the open source world, are also directly optimized for RTX hardware.

A New Era Begins in Local Artificial Intelligence

Privacy concerns, cloud subscription fees, and latency are increasingly driving users to AI solutions running on local hardware. NVIDIA's transformation of the processing power of all devices at home into a common pool with PAIR and its raising of the hardware bar with RTX Spark computers with 128 GB combined memory seems to bring the future of personal artificial intelligence assistants to completely local systems.