Microsoft and Nvidia Put Local AI Agents at the Center of the Surface Laptop Ultra

Microsoft and Nvidia are putting locally running AI agents at the center of a new class of Windows laptops, with the Surface Laptop Ultra powered by Nvidia’s RTX Spark platform. The companies are showcasing the hardware in San Francisco as they push AI workloads that can run directly on a Windows PC rather than depending entirely on cloud computing. The RTX Spark platform combines Nvidia’s Blackwell GPU architecture, a Grace CPU and up to 128GB of unified memory, with Microsoft’s Windows security infrastructure designed for on-device agents.

The Laptop Is Becoming an AI Computer

For years, powerful AI workloads have largely been associated with remote data centers. A user sends a request to the cloud, servers perform the computation and the result comes back to the device.

Microsoft and Nvidia are now pushing a different model: put enough computing power inside the PC that an AI agent can perform substantial tasks locally.

That is the idea behind the Surface Laptop Ultra, Microsoft’s high-performance Surface laptop built around Nvidia’s RTX Spark technology. Microsoft has positioned the machine for developers, creators and AI builders working with demanding local workloads.

The shift is larger than simply adding another AI feature to Windows. It is an attempt to make the computer itself the place where an AI agent operates.

Nvidia’s RTX Spark Brings Data-Center-Class AI Hardware Closer to the PC

At the center of the system is Nvidia’s RTX Spark superchip.

The platform combines a Blackwell-based RTX GPU with a 20-core Nvidia Grace CPU and high-speed chip-to-chip connectivity. Nvidia says the platform can deliver up to 1 petaflop of AI computing performance and support configurations with as much as 128GB of unified memory, allowing CPU and GPU workloads to share a large memory pool.

That memory architecture matters for local AI because large models can require substantial amounts of memory simply to load and operate.

Nvidia says RTX Spark systems can run AI models with up to 120 billion parameters locally, depending on the configuration and software. Microsoft similarly describes the Surface Laptop Ultra as capable of demanding local AI workflows, rendering, compilation and multimodel workloads.

The significance is not that every AI task will suddenly move away from the cloud. Rather, the hardware expands the range of workloads that can be performed directly on the machine.

What Makes an AI Agent Different From a Chatbot?

The companies are targeting a broader role for AI than simply answering questions.

An AI agent can be designed to take a sequence of actions—such as writing and testing code, working with files, using software tools or carrying out a multi-step workflow.

Running such an agent locally creates different technical requirements from running a conventional chatbot. The computer needs enough processing power and memory to keep models active while also providing the operating-system controls necessary to prevent an autonomous process from accessing resources it should not use.

That is where Microsoft’s Windows work becomes important.

Microsoft Is Building Security Around Local Agents

Microsoft and Nvidia have been developing a software layer intended to make autonomous agents safer to run directly on Windows.

Microsoft’s Microsoft eXecution Containers (MXC) provide an operating-system-level mechanism for isolating agents from the broader system. Nvidia’s OpenShell runtime builds on that foundation with capabilities including policy management, inference routing and protection mechanisms for sensitive information.

The goal is to give an AI agent enough access to perform useful work without giving it unrestricted control of the computer.

That becomes increasingly important as agents move from generating text to actually taking actions.

A local agent that can read files, execute code or interact with applications represents a fundamentally different security problem from a chatbot that simply returns an answer.

Less Dependence on the Cloud

One of the biggest changes could be the economics of AI computing.

Cloud AI requires data-center infrastructure, and many AI services charge according to usage. Local processing shifts at least some of that computation to hardware that the user already owns.

Nvidia has described RTX Spark as a platform for running personal AI agents locally, while Microsoft has promoted the idea of Windows becoming an environment where agents can operate directly on the user’s device.

For users working with sensitive files, local processing can also reduce the need to send certain information to remote servers.

That does not mean local AI automatically provides complete privacy. Applications, models and cloud-connected workflows can still transmit information depending on how they are designed. But keeping computation on the device can provide an additional option for sensitive workloads.

The Surface Laptop Ultra Is Built Around This Strategy

Microsoft’s Surface Laptop Ultra is the clearest expression of the partnership in laptop form.

Microsoft says the device combines Nvidia’s Blackwell RTX GPU, full CUDA support and up to 128GB of unified memory in a thin laptop design. The company has also positioned it for local AI, 3D rendering, software compilation and other demanding professional workloads.

The hardware is therefore not being marketed simply as a faster conventional laptop.

Its purpose is to provide enough local compute for workloads that have traditionally pushed users toward workstations, specialized AI hardware or cloud GPU infrastructure.

Microsoft’s broader Surface lineup also includes the Surface RTX Spark Dev Box, a compact developer-focused system using the same RTX Spark platform. Microsoft says it is designed for sustained AI development, local model fine-tuning and agentic workflows.

Windows Is Being Designed for an Agentic Computing Era

The hardware is only one part of Microsoft’s strategy.

At Microsoft Build earlier this year, the company described Windows as moving toward an agent-native runtime, with operating-system controls designed to manage how agents execute tasks.

The partnership with Nvidia extends across local devices and cloud infrastructure, rather than treating them as competing environments. Developers can build workflows that operate locally when possible while still using cloud resources when additional computing power or capabilities are required.

This hybrid approach may prove more practical than trying to eliminate cloud AI altogether.

A laptop can handle private documents, routine automation and smaller local models while sending exceptionally demanding workloads to remote infrastructure when necessary.

The Hardware Challenge Has Not Disappeared

The move toward local AI comes with a major trade-off: powerful AI hardware is expensive.

Running large models locally requires substantial memory and computing capability, which can increase both the cost and thermal demands of a laptop.

Reuters reported that analysts see hardware pricing as a potential obstacle to widespread adoption, particularly as memory costs rise. Nvidia’s own high-performance local AI systems illustrate how expensive large-memory configurations can become.

That creates an unusual situation for the AI industry.

The software is becoming capable enough to run sophisticated agents locally, but the hardware required to run those agents comfortably can remain outside the reach of many mainstream consumers.

Nvidia Is Entering a Different Part of the PC Market

The partnership also has implications for the semiconductor industry.

Windows PCs have historically been dominated by x86 processors from Intel and AMD. Nvidia’s RTX Spark platform uses an Arm-based Grace CPU alongside its Blackwell GPU architecture, creating a different architecture for high-performance Windows computing.

Nvidia is therefore not simply supplying another graphics processor.

It is attempting to establish a broader computing platform in which the CPU, GPU, memory architecture and AI software stack are designed around local AI workloads.

Microsoft, meanwhile, gains a hardware foundation for its vision of Windows as an operating environment where AI agents can operate directly on personal computers.

The Bigger Shift Is From AI Assistant to AI Computer

The most important change may ultimately have little to do with the Surface Laptop Ultra itself.

The larger trend is a shift in how the personal computer is expected to work.

Traditional software waits for the user to open an application, provide instructions and complete individual steps. Agentic computing aims to let users describe an objective while software handles a sequence of actions.

For that model to become practical at scale, AI needs access to local computing resources, operating-system controls and enough memory to run capable models without constantly depending on remote infrastructure.

Microsoft and Nvidia are now building those pieces together.

The Surface Laptop Ultra represents one early expression of that strategy: a Windows computer designed not merely to run AI applications, but to provide the computing environment in which AI agents can work.

Whether that model reaches mainstream PCs will depend on three things—cost, reliability and security.

The technical capability is advancing rapidly. The next question is whether the economics and trust required for everyday users will advance at the same pace.

Most Searched 5 FAQs

1. What is the Microsoft Surface Laptop Ultra?

The Surface Laptop Ultra is Microsoft’s high-performance Surface laptop built around Nvidia’s RTX Spark platform, targeting developers, creators and demanding local AI workloads.

2. Can the Surface Laptop Ultra run AI models without the cloud?

Yes. Nvidia and Microsoft designed RTX Spark Windows systems to support substantial AI workloads locally, including large language models and AI-agent workflows, although cloud computing can still be used when required.

3. What is Nvidia RTX Spark?

RTX Spark is Nvidia’s computing platform combining a Blackwell RTX GPU, Grace CPU and unified memory. Nvidia says the platform can provide up to 1 petaflop of AI performance and configurations with up to 128GB of unified memory.

4. What are AI agents on Windows?

AI agents are software systems designed to perform multi-step tasks rather than simply respond to individual prompts. Microsoft’s Windows agent infrastructure is intended to provide controlled environments in which such agents can execute tasks securely.

5. Does local AI mean cloud AI is going away?

No. Microsoft and Nvidia are developing a hybrid approach in which AI workloads can run locally or in the cloud depending on the task, available hardware, security requirements and computing needs.