NVIDIA and Microsoft unveil RTX Spark AI PC platform

NVIDIA and Microsoft revealed RTX Spark, a Windows-native PC platform designed around local AI agents, content-creation workloads and RTX graphics, announced at Computex 2026 on May 31, 2026. According to NVIDIA's newsroom release, the RTX Spark superchip offers 1 petaflop of AI performance and up to 128GB of unified memory, pairing a Blackwell GPU with up to 6,144 CUDA/RTX cores and a 20-core Arm-based CPU co-developed with MediaTek (NVIDIA Newsroom; Windows Experience Blog). Microsoft and OEM partners including Microsoft Surface, ASUS, Dell, HP, Lenovo, MSI, Acer and GIGABYTE are announced as early device partners; over 30 laptops and 10 desktops are expected this fall (Tom's Hardware; NVIDIA Newsroom).

Per NVIDIA and Microsoft announcements, RTX Spark integrates CUDA, RTX, DLSS and TensorRT technology into a single superchip targeted at thin-and-light laptops and compact desktop Windows PCs (NVIDIA Newsroom). Vendor materials claim the platform can run 120B-parameter LLMs with 1 million tokens of context using local agents, render 90GB+ 3D scenes, edit 12K 4:2:2 video, and deliver AAA gaming at 1440p and over 100 fps; these are performance claims from NVIDIA (NVIDIA Newsroom; Windows Experience Blog). NVIDIA also highlighted a secure agent execution component called NVIDIA OpenShell in its press materials.

Packaging a high-throughput Blackwell GPU, a multi-core Arm CPU and large unified memory into a single superchip shifts tradeoffs previously handled by discrete CPU/GPU and separate VRAM pools. Devices with large unified memory and tensor throughput reduce inter-component data movement, which benefits large-context LLM inference and multimodal agent runtimes. Developers building latency-sensitive agent chains and multimodal pipelines may see lower local inference latency than cloud-dependent setups, although real-world performance depends on software maturity, model weight formats and tooling support.

Public reporting frames RTX Spark as part of a broader push to move agentic and creative workloads onto primary devices rather than into the cloud. The NVIDIA-Microsoft partnership combines silicon claims with Windows-level integration and OEM distribution, which could accelerate on-device agent experimentation in creative and developer toolchains. Ecosystem readiness -- model compression, memory bandwidth optimizations, toolchain support, and ISV application ports -- will determine how many of NVIDIA's claimed workload targets translate into day-to-day productivity gains.

Observers should track three practical indicators:

• •third-party benchmark results for 120B-parameter inference and multi-hour content workflows versus cloud baselines
• •availability of software stacks -- OS-level primitives, secure agent runtimes and vendor toolchains -- from Microsoft and OEMs
• •application ports from major ISVs, notably Adobe, which NVIDIA and Microsoft cited as rearchitecting Photoshop and Premiere for RTX Spark (NVIDIA Newsroom). Power and thermal profiles on thin-and-light devices will also matter given petaflop-class performance expectations

Jensen Huang said at NVIDIA's Computex 2026 announcement: "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" (NVIDIA Newsroom).