You ask and the PC does the work.
For decades, the personal computer has been a tool that waits for instruction — obedient, but inert. Now, NVIDIA and Microsoft are proposing something older in spirit and stranger in form: a machine that anticipates, reasons, and acts. Unveiled at GTC Taipei, their RTX Spark platform reimagines the Windows PC not as an instrument you play, but as a collaborator you direct — one capable of running powerful AI agents locally, with the user retaining meaningful control over what those agents may touch and know.
- The personal computing paradigm that has held for forty years — open an app, issue a command, receive a result — is being directly challenged by a shift toward autonomous software agents that act on your behalf.
- The deepest friction is trust: granting an AI system broad access to personal files, applications, and data remains deeply uncomfortable for many users, and that discomfort could stall adoption before it begins.
- Microsoft's OpenShell runtime attempts to answer that concern directly, offering users defined privacy boundaries and the ability to keep sensitive requests processed locally rather than sent to the cloud.
- Over 100 developers — including Adobe, Xbox, and Blender — are already building for the platform, suggesting the industry is moving in this direction regardless of whether consumers are ready.
- The first RTX Spark devices from ASUS, Dell, HP, Lenovo, and others are expected before year's end, making this a near-term test of whether the agent-centric vision can find a real audience.
At NVIDIA's GTC Taipei conference, NVIDIA and Microsoft unveiled RTX Spark — a computing platform designed to transform Windows PCs into something more like intelligent collaborators than passive tools. The announcement signals a fundamental rethinking of personal computing: rather than launching applications and issuing commands, users would direct AI agents capable of reasoning through problems and executing multi-step tasks on their own. Jensen Huang captured the ambition plainly: "You ask and the PC does the work."
The hardware underneath this vision is substantial. RTX Spark pairs NVIDIA's Blackwell GPU architecture with a custom Arm processor developed with MediaTek, delivering up to one petaflop of AI performance and support for 128 gigabytes of unified memory. That is enough to run large language models with up to 120 billion parameters entirely on-device, process million-token context windows, and generate video — all without leaning on cloud infrastructure. Traditional PC capabilities, including high-frame-rate gaming and professional creative work, are preserved alongside these new capacities.
The more consequential announcement may have been OpenShell, a Windows-native runtime that gives users governance over their AI agents — defining what files and data agents can access, and routing sensitive requests to local models rather than remote servers. Satya Nadella framed the broader initiative as bringing "unmetered intelligence" to homes and workplaces, but the privacy architecture suggests Microsoft understands that capability alone will not drive adoption. Trust remains the central obstacle.
The ecosystem forming around RTX Spark is already broad. Adobe is redesigning core elements of Photoshop and Premiere Pro for the platform, while more than 100 developers — including Blender, Blackmagic Design, Xbox, and Remedy Entertainment — have committed support. Devices from ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI are expected before the end of the year, spanning thin laptops and compact desktops. Whether consumers will genuinely embrace a shift from apps to agents remains unresolved — but the industry, it seems, has already decided the question is worth asking.
NVIDIA and Microsoft are betting that the next era of personal computing won't be defined by the applications you launch, but by the agents you command. At NVIDIA's GTC Taipei conference, the two companies unveiled RTX Spark, a new computing platform designed to turn Windows PCs into something closer to intelligent assistants—machines that can reason through problems, plan multi-step tasks, and execute them without constant human direction. Manufacturers including ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI are already preparing devices built around this architecture, with the first machines expected to arrive before year's end.
The shift marks a departure from how personal computers have operated for decades. Historically, the PC has been a tool you control directly—you open an application, you give it commands, you watch it respond. What NVIDIA and Microsoft are proposing is different: software agents that sit between you and your machine, intermediaries capable of understanding what you want and handling the work themselves. NVIDIA CEO Jensen Huang framed it simply at the conference: "With RTX Spark and Microsoft Windows, you ask and the PC does the work."
At the hardware level, RTX Spark combines NVIDIA's Blackwell graphics architecture with a custom Arm-based processor developed alongside MediaTek. The platform delivers up to one petaflop of AI computing performance and supports as much as 128 gigabytes of unified memory—specifications designed to let machines run large language models with up to 120 billion parameters locally, process context windows reaching one million tokens, generate video, and handle complex AI workflows without constantly reaching out to cloud services. NVIDIA claims the system will also maintain traditional PC capabilities: AAA gaming at over 100 frames per second at 1440p resolution, professional creative work, the full range of tasks users expect from a modern computer.
But the hardware announcement was almost secondary to what NVIDIA and Microsoft are building in software. The two companies introduced OpenShell, a new Windows-native runtime that gives users control over what their AI agents can access and how they behave. You can define boundaries for agents, determine what information leaves your device, and route sensitive requests to local AI models rather than cloud services. This addresses what may be the central obstacle to widespread adoption of agent-based computing: trust. As AI systems become more capable, many people remain deeply uncomfortable granting them broad access to files, applications, and personal information. Running these systems locally, with stronger controls, could change that calculation. Microsoft CEO Satya Nadella described the initiative as part of the company's push to bring what he called "unmetered intelligence" to homes and workplaces.
The ecosystem is already forming around the platform. Adobe announced it is redesigning core components of Photoshop and Premiere Pro to take advantage of RTX Spark's unified memory architecture and AI acceleration, aiming to deliver faster AI-assisted editing and content creation. More than 100 software developers and gaming companies are supporting the platform, including Blackmagic Design, Blender, CapCut, ComfyUI, NetEase, Remedy Entertainment, and Xbox. For developers working with AI models, NVIDIA is positioning RTX Spark as a portable workstation capable of running advanced local AI workloads that previously required either desktop-class hardware or cloud infrastructure.
The first devices will span a range of form factors—ultra-thin laptops aimed at creators and developers, compact desktops designed for AI workloads and gaming. But the larger ambition extends far beyond a hardware refresh cycle. For the personal computer industry, this represents a fundamental rethinking of what a PC is for. Whether consumers actually embrace the shift from apps to agents remains an open question. But as AI increasingly moves from answering questions to performing actions, NVIDIA's latest announcement suggests the industry believes the next computing platform may not be defined by what applications you can run, but by what agents can do on your behalf.
Notable Quotes
With RTX Spark and Microsoft Windows, you ask and the PC does the work.— NVIDIA CEO Jensen Huang
Unmetered intelligence brought to homes and workplaces.— Microsoft CEO Satya Nadella, describing the company's broader ambition
The Hearth Conversation Another angle on the story
Why does it matter that these AI agents run on the device itself rather than in the cloud?
Control and trust. If your agent is making decisions about your files, your calendar, your sensitive information, you probably don't want that happening on someone else's server. Local processing means the data stays with you, and you can set boundaries.
But doesn't that require a lot of computing power? Isn't that why cloud AI exists in the first place?
It does, which is why the hardware matters so much. RTX Spark is designed to handle models with 120 billion parameters locally. That's genuinely powerful. A few years ago, you'd need a data center for that.
So this is about shifting power back to the user?
Partly. But it's also about speed and responsiveness. If your agent doesn't have to make a round trip to the cloud, it can act faster. And for creative work—video editing, image generation—that latency matters.
What's the risk here? What could go wrong?
Adoption is the big one. People have been using apps for forty years. Agents are a different mental model. You have to trust the system to do what you intended. And there's the question of whether local hardware can really handle everything people need.
Is this just a marketing story, or is something genuinely changing?
The fact that Adobe is redesigning Photoshop around this, that over 100 developers are on board—that suggests it's real. But whether it becomes the dominant way people use computers? That's still open.