Nvidia and Microsoft Reinvent the Personal Computer for the Age of AI Agents

2026-06-01 data

San Jose, Monday, 1 June 2026.
For the first time in 40 years, the personal computer is being fundamentally redesigned. Nvidia’s new RTX Spark superchip can run a 120-billion-parameter AI model locally — no cloud required. Slim laptops arrive Fall 2026.

A New Superchip at the Heart of the Reinvention

On May 31, 2026, at the GPU Technology Conference (GTC) held in Taipei, Taiwan, Nvidia CEO Jensen Huang took the stage for a two-hour keynote that the industry is already calling a watershed moment in personal computing [1][2]. The announcement at the center of it all: the NVIDIA RTX Spark™ superchip, a single processor that integrates an NVIDIA Blackwell RTX GPU — featuring 6,144 CUDA cores and fifth-generation Tensor Cores with FP4 precision — with a custom 20-core NVIDIA Grace CPU co-designed with MediaTek, all connected via NVLink-C2C [2][5]. The result is a chip capable of delivering 1 petaflop of AI performance with up to 128 GB of unified memory [2][5]. In Huang’s own words, delivered from the Taipei stage: “The PC is being reinvented. For forty years, you launched apps. Click. Type. With RTX Spark and Microsoft Windows, you ask — and the PC does the work.” [2]

What the RTX Spark Can Actually Do

The practical capabilities of the RTX Spark superchip move well beyond marketing language. The chip is engineered to run 120-billion-parameter large language models (LLMs) locally, with a context window of up to 1 million tokens — entirely on-device, without routing queries to a cloud server [2][5]. For creative professionals, the chip supports 12K 4:2:2 video editing, the rendering of 3D scenes exceeding 90 GB, and AAA gaming at 1440p resolution at over 100 frames per second [2][5]. The hardware natively supports the NVIDIA CUDA software stack, DLSS, NVIDIA Reflex, and G-SYNC, as well as FP4 Tensor Cores, RT Cores, AV1 encoders, and 4:2:2 hardware encode and decode [5]. Adobe has already committed to rearchitecting both Photoshop and Premiere Pro to leverage the new platform, targeting up to 2x faster AI performance by taking full advantage of RTX Spark’s unified memory, Blackwell GPUs, and TensorRT [2][8]. More than 100 software providers have adopted the platform ahead of its commercial release [2].

Microsoft and the Secure Agent Stack

The RTX Spark superchip does not arrive alone. Nvidia has partnered deeply with Microsoft to deliver what the two companies are calling a native Windows experience for personal AI agents [2][4]. The collaboration introduces new Windows security primitives — covering identity, containment, policy, and end-to-end security — alongside the NVIDIA OpenShell runtime, which manages local and cloud query routing and user privacy policies [2][8]. Together, these form a secured on-device stack that allows AI agents to run privately on a user’s primary Windows device, with plans to embed RTX Spark-powered agent experiences directly into the Windows taskbar [2]. Microsoft chairman and CEO Satya Nadella, based in Redmond, Washington, described the ambition clearly: “Our goal is to deliver unmetered intelligence to every home and every desk with Windows. RTX Spark marks a real breakthrough towards that vision.” [2] The collaboration will be further showcased at Microsoft’s Build developer conference, scheduled for June 2–3, 2026, in San Francisco, where Windows agent capabilities and new security primitives will be demonstrated to developers [2][4].

Open-Source Agents and a Growing Software Ecosystem

Prior to the May 31 announcement, open-source AI agent projects such as OpenClaw and Hermes had gained rapid adoption on GitHub, but their growth had been constrained by the inability to run securely and privately on local personal computers [8]. The introduction of NVIDIA OpenShell changes that equation directly. The runtime, built on Microsoft’s new security primitives, enables projects like OpenClaw and Hermes Agent to deploy securely on RTX Spark-powered devices [8]. Vincent Koc, chief architect at the OpenClaw Foundation, stated: “Running solutions like OpenShell and the Microsoft security primitives on RTX Spark will enable users to leverage a fully integrated stack for private, personal agents running on device.” [2] On the performance side, NVIDIA has also announced a 2x inference performance boost in llama.cpp and vLLM for RTX platforms, as well as up to 2x performance gains in ComfyUI, alongside a 2.6x inference speedup in vLLM using optimized NVFP4 checkpoints for Qwen 3.6 35B models [8]. H Company is also planning to launch the Holo Desktop app and a desktop agent harness for RTX and DGX PCs, using quantized Holo Computer Use models that deliver a 2x speedup and a 35% memory reduction on NVIDIA GPUs [8].

From Slim Laptops to Deskside Supercomputers

The RTX Spark platform spans a wide range of form factors. On the consumer end, slim laptops measuring as thin as 14 mm and weighing as little as 1.36 kg will be released in Fall 2026, available in screen sizes ranging from 35.6 cm to 40.6 cm [2][6]. Manufacturing partners confirmed for the initial wave include ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI, with Acer and GIGABYTE models to follow at a later date [2][6]. Microsoft’s own Surface Laptop Ultra — targeting creators, developers, and engineers — forms part of this lineup [2]. At the enterprise end of the spectrum, Nvidia simultaneously announced the NVIDIA DGX Station™ for Windows, a deskside AI supercomputer powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, which connects a Blackwell Ultra GPU to a 72-core NVIDIA Grace CPU, offering up to 748 GB of coherent memory and up to 20 petaflops of FP4 performance [3]. The DGX Station is capable of running frontier AI models of up to 1 trillion parameters entirely locally and will be available in Q4 2026 from ecosystem partners including ASUS, Dell Technologies, GIGABYTE, HP, MSI, and Supermicro [3]. Pavan Davuluri, executive vice president of Windows + Devices at Microsoft, framed the enterprise ambition in direct terms: “This unlocks a new class of AI performance on Windows, the platform enterprises trust for security, manageability and compatibility.” [3]

Competing in a Market Apple Currently Dominates

Nvidia’s push into the PC chip market is not without competitive context. Technology analyst Max Weinbach of Creative Strategies noted bluntly: “Apple more or less owns this market today,” and that “Nvidia wants to build a laptop ecosystem for Windows that’s an alternative.” [7] With the RTX Spark, Nvidia is also entering territory historically held by Intel and AMD in the Windows PC space, adopting a chip architecture that, according to reporting from May 31, 2026, is similar in approach to Qualcomm’s — which was itself only enabled on Windows in 2021 after Microsoft opened the platform to non-Intel architectures due to Intel’s performance struggles [4][7]. Microsoft’s first AI PC effort, the Copilot+ PC launched on May 20, 2024, encountered significant setbacks tied to the “Recall” feature, including delays and security concerns, which in part motivated the pivot toward more capable local compute and on-device agents to reduce dependency on cloud infrastructure [4]. Industry analyst Carolina Milanesi of Current Strategies offered a measured assessment of what Nvidia’s entry means for the broader market: “From an industry perspective, it’s a good thing.” [4]

The Bigger Picture: Compute as a Strategic Asset

Jensen Huang’s Taipei keynote carried a message that extended well beyond hardware specifications. Addressing employment concerns directly on May 31, 2026, Huang stated that AI is not displacing software engineers, pointing to active hiring across the industry [1]. His broader vision was more expansive still: “I imagine someday an AI supercomputer in your house running agents. And you have to have it in your house and these in time become more like R2-D2 to you than a PC to you.” [1] For enterprises, Huang’s framing was equally direct: “Compute is revenue. The more you buy, the more you make.” [1] Chris Marriott, vice president of enterprise platforms at NVIDIA, elaborated on the enterprise rationale, noting that “as enterprises scale AI agents across their organizations, they need AI infrastructure that can connect directly to the applications and workflows that power their business.” [3] Historically, heavy-duty enterprise AI workloads required Linux-based data center systems, which created a significant gap for the majority of Fortune 500 companies that use Windows as their primary operating environment for daily productivity and creative workflows [3]. The DGX Station for Windows, combined with the consumer RTX Spark platform, represents Nvidia’s attempt to close that gap entirely — from the thinnest laptop to the most powerful deskside workstation — all running within the Windows ecosystem that enterprises already depend upon [2][3].

Bronnen

AI PC Nvidia GTC