The PC moves from tool to teammate
At a moment when artificial intelligence has largely lived in distant data centers, NVIDIA is proposing something more intimate: a personal computer that carries the full weight of AI reasoning within its own chassis, answering only to its owner. The RTX Spark superchip, arriving in laptops this fall, unites a Blackwell GPU with a Grace CPU and enough unified memory to run the largest language models without ever leaving the device. In partnering with Microsoft to build security and agency controls directly into Windows, NVIDIA is not merely selling faster hardware — it is advancing a vision of computing where intelligence is private, portable, and genuinely personal.
- The central tension is one of trust: powerful AI agents have been held back not by capability but by the fear of what they might do with access to your most sensitive files, accounts, and workflows.
- RTX Spark's arrival disrupts the assumption that serious AI performance requires cloud infrastructure, placing one petaflop of compute and 128GB of unified memory inside a laptop as thin as 14 millimeters.
- NVIDIA and Microsoft are attempting to resolve the trust problem through architecture itself — new Windows security primitives and the OpenShell runtime let users define hard boundaries around what agents can access, share, or act upon.
- Adobe is rebuilding Photoshop and Premiere from the ground up for this platform, signaling that the creative industry is treating RTX Spark not as an upgrade but as a generational shift in how professional tools are designed.
- With over 100 software and game developers committed and five major PC makers shipping hardware this fall, the platform is landing with enough momentum to suggest a genuine inflection point rather than a product announcement.
NVIDIA has introduced RTX Spark, a superchip that combines a Blackwell GPU with 6,144 CUDA cores and a custom 20-core Grace CPU, connected by NVLink to form a unified system capable of one petaflop of AI performance. With 128 gigabytes of shared memory, it can run 120-billion-parameter language models with up to one million tokens of context entirely on-device — no cloud required. The machines built around it will be slim, light, and battery-efficient, with OLED displays suited to creative and gaming work.
The deeper ambition is not speed but sovereignty. NVIDIA and Microsoft have co-developed new Windows security primitives — covering identity, containment, policy, and encryption — alongside NVIDIA's OpenShell runtime, which gives users precise control over what AI agents can and cannot do. Agents can be confined to local tasks, and any query that must reach the cloud can have personal information automatically stripped before it leaves the device. The user, not the agent, holds the keys.
Early agent developers including OpenClaw and Hermes Agent are building Windows applications on this foundation, enabling AI teammates that can reason across applications, generate media, write code, and search files — all within user-defined limits. Adobe is going further, rearchitecting both Photoshop and Premiere to exploit RTX Spark's memory and GPU architecture, promising twice the speed for AI features like Generative Fill and a new video pipeline in Premiere capable of real-time editing on footage that previously demanded rendering time.
The broader ecosystem has responded in kind. More than 100 software providers and game developers have pledged support, spanning professional creative tools, major game studios, and the XBOX platform. Creators will be able to edit 12K video and render large 3D scenes on a device they can carry; gamers will run AAA titles at 1440p with ray tracing above 100 frames per second.
RTX Spark laptops and desktops from ASUS, Dell, HP, Lenovo, and Microsoft Surface are set to arrive this fall, with Acer and GIGABYTE to follow. Dell's XPS 16 Creator Edition, HP's ultra-thin OmniBook entry, and Microsoft's Surface Laptop Ultra each represent a shared conviction: that the next era of computing will be local, intelligent, and shaped around the individual rather than the infrastructure.
NVIDIA has built a new kind of computer. Not a faster version of what came before, but something fundamentally different—a Windows PC designed from the ground up to run personal AI agents locally, securely, and with enough raw power to handle the work that used to require cloud servers or specialized hardware.
The chip is called RTX Spark. It combines an NVIDIA Blackwell GPU with 6,144 CUDA cores alongside a custom 20-core NVIDIA Grace CPU, connected through an NVLink chip-to-chip interconnect. The result: one petaflop of AI performance, 128 gigabytes of unified memory, and the ability to run 120-billion-parameter language models with up to one million tokens of context entirely on your laptop. The machines themselves will be thin—as slim as 14 millimeters—and light enough to carry, with all-day battery life and color-accurate OLED displays built for creative work and gaming.
But the hardware is only half the story. NVIDIA and Microsoft have partnered to solve a problem that has limited AI agent adoption so far: how do you run powerful autonomous agents on your primary computer without surrendering privacy or control? The answer involves new Windows security primitives—identity, containment, policy, and end-to-end encryption built into the operating system itself—paired with NVIDIA OpenShell, a runtime that lets users define exactly what agents can and cannot do. If an agent needs to send a query to a cloud model, OpenShell can automatically redact personal information. If a task should stay local, the system routes it there instead. The user remains in control.
Leading agent developers are already adopting the stack. OpenClaw and Hermes Agent, two projects that have gained significant traction on GitHub and OpenRouter, will ship Windows apps built on these security foundations. The agents can execute tasks across multiple applications, reason through complex workflows, generate images and video, write code, and search local files semantically. They become teammates rather than tools—always available, always learning your preferences, always working within boundaries you set.
Adobe is rearchitecting two of its flagship applications from the ground up for RTX Spark. Photoshop and Premiere will be rebuilt to take full advantage of the unified memory and GPU acceleration, delivering up to twice the speed for AI-powered features like Generative Fill and Generative Extend. The new Photoshop engine will support GPU-accelerated compositing, live filters, and high dynamic range work. Premiere will feature a new video pipeline that taps directly into RTX Spark's memory architecture, enabling real-time editing and color correction on timelines that would have required rendering time before. Both applications will eventually allow users to work alongside Windows agents—a creative collaborator that can handle repetitive tasks, suggest edits, and accelerate workflows.
The creative and gaming ecosystem is rallying around the platform. Over 100 software providers and game developers have committed to RTX Spark support, from Blackmagic Design and OTOY to KRAFTON, Riot Games, and XBOX. Creators will be able to render 90-gigabyte 3D scenes, edit 12K video, and generate 4K AI video on a portable device. Gamers will play AAA titles at 1440p resolution with ray tracing and DLSS at over 100 frames per second.
The first RTX Spark machines arrive this fall. ASUS, Dell, HP, Lenovo, and Microsoft Surface are all shipping models, with Acer and GIGABYTE to follow. Dell is building the XPS 16 Creator Edition. HP is designing one of the thinnest RTX Spark laptops in its OmniBook line. Microsoft is releasing the Surface Laptop Ultra, engineered for creators and developers who need serious performance in a portable, thoughtfully designed package. The machines represent a bet that the future of computing is local, intelligent, and personal—not cloud-dependent, not generic, but genuinely yours.
Citas Notables
For forty years, you launched apps. Click. Type. With RTX Spark and Microsoft Windows, you ask—and the PC does the work.— Jensen Huang, NVIDIA CEO
Our goal is to deliver unmetered intelligence to every home and every desk with Windows.— Satya Nadella, Microsoft CEO
La Conversación del Hearth Otra perspectiva de la historia
Why does it matter that agents run locally instead of in the cloud? Doesn't cloud computing already handle this?
Because every query you send to the cloud is data about you—what you're working on, what you're asking, what you care about. Running agents locally means your private information never leaves your machine. You get speed, privacy, and control all at once.
But doesn't that limit what the agent can do? A cloud model has access to more knowledge.
Not necessarily. RTX Spark can run 120-billion-parameter models locally. That's enormous. And when you do need a frontier model in the cloud, OpenShell redacts your personal details before sending anything. You get the best of both worlds.
So the security layer is the real innovation here, not the chip itself?
The chip enables it. You can't run these models locally without the compute power. But yes—the partnership with Microsoft to build security and containment into Windows itself is what makes it practical. Developers can trust that agents will behave safely.
Adobe is rearchitecting Photoshop and Premiere. That's a massive undertaking. Why now?
Because unified memory changes everything. Before, you had to move data between the GPU and CPU constantly. Now it's all in one pool. Photoshop can apply filters in real time. Premiere can edit 12K video without rendering. The hardware finally makes it possible.
What happens to the cloud GPU market if everyone runs models locally?
It doesn't disappear. Frontier models—the cutting-edge ones—will still live in the cloud. But routine tasks, creative work, personal agents—those move to the device. It's a shift in what compute lives where, not the end of cloud computing.
These laptops launch in the fall. What's the realistic adoption curve?
Creators and developers will move first. They've been waiting for portable power like this. Gamers will follow. The mainstream consumer market takes longer. But once people experience an agent that actually understands their workflow and runs privately on their machine, the expectation changes.