NVIDIA and Microsoft have announced RTX Spark, a new Windows PC platform aimed at running AI agents directly on the device rather than relying entirely on cloud services. The companies say the system is designed to make personal computers more capable of handling everyday AI tasks locally, with a focus on speed, privacy and control.
The pitch behind RTX Spark is straightforward. Instead of sending every complex request to the cloud, users could run more capable agents on their own machines and keep more data and activity close to home. That would allow PCs to take on functions such as working with files, opening applications and completing multi-step tasks without constant internet dependence or recurring cloud usage.
According to NVIDIA, RTX Spark systems will deliver up to 1 petaflop of AI performance and include 128GB of unified memory. The company says that configuration should allow the PCs to run models with as many as 120 billion parameters locally, giving the device enough headroom for more advanced on-device AI workloads.
Microsoft is contributing security features for Windows, while NVIDIA is adding OpenShell, a runtime meant to restrict what agents can access and how far they can operate. The security layer is intended to help keep local agents within defined limits as they interact with real applications and files on the user’s computer.
The companies said RTX Spark laptops and desktops are expected to arrive this fall from several major PC makers, including Microsoft Surface, ASUS, Dell, HP, Lenovo and MSI.
The launch reflects a broader industry push to move more AI functions onto personal devices. Cloud-based tools have made advanced models widely available, but they also add latency, cost and dependency on remote servers. NVIDIA and Microsoft are positioning RTX Spark as a way to reduce those tradeoffs by making the machine itself the main environment for everyday AI work.
That could matter most for users who want agents that can manage repeated computer tasks, handle personal data and operate with less friction. It also fits a growing interest in AI systems that do not just answer questions, but actively carry out work across the desktop.
For developers, Microsoft Build is expected to provide an early look at Windows agent platform demonstrations tied to the new effort. Wider availability will likely depend on the fall hardware launches and whether the software stack proves stable enough for real-world use.
RTX Spark also points to a strategic question for the AI industry. If agents become the main way people interact with software, then the device running those agents may become a crucial distribution channel. That could give Windows PCs a stronger role in the next phase of consumer AI, especially if local execution becomes practical for more users.
The move does not mean cloud AI is going away. NVIDIA and Microsoft are making a case that local devices should take on a larger share of routine tasks, while remote models remain available for heavier workloads. The result, if the companies succeed, could be a more split model of AI computing, with personal PCs handling many day-to-day jobs and the cloud reserved for more demanding work.
For now, RTX Spark remains a platform announcement rather than a shipping product. But it signals that two of the biggest names in computing are betting that the next wave of AI will not live only in data centers. It may also live in the laptops and desktops people already use.