Summary of Major Developments
• RTX Spark announced June 1 at Computex Taipei: Jensen Huang unveiled the NVIDIA RTX Spark superchip during his keynote at Computex 2026 in Taipei on June 1, 2026. The RTX Spark is NVIDIA’s first Windows on Arm platform for laptops and compact desktop PCs, combining the company’s Blackwell GPU architecture with a new custom Arm-based N1X CPU co-developed with Microsoft and MediaTek via NVLink C2C chip-to-chip interconnect.
• Flagship N1X specifications: The top-tier RTX Spark N1X superchip features 20 Arm CPU cores (10 Cortex-X925 performance cores at 4.1GHz + 10 Cortex-A725 efficiency cores), a Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor Cores, up to 128GB of unified LPDDR5X memory at 300 GB/s bandwidth, and 1 petaflop of FP4 AI compute. TDP ranges from 45W to 80W, positioning it against Apple M5 and Qualcomm Snapdragon X Elite for the premium thin-and-light market.
• OEM launch partners and timeline: More than 30 laptops and approximately 10 compact desktops using RTX Spark will launch in fall 2026 from Dell, HP, ASUS, Lenovo, MSI, and Microsoft. The Microsoft Surface Laptop Ultra, powered by the RTX Spark with 128GB of RAM, was shown on stage. NVIDIA’s CEO described the platform as the first complete reinvention of the PC in 40 years.
Technical Breakdown: RTX Spark Architecture and What It Delivers
The RTX Spark superchip is architecturally significant because it is the first time NVIDIA has entered the client PC processor market — a domain controlled for decades by Intel, AMD, and more recently Qualcomm’s Arm-based Snapdragon X series. The RTX Spark’s N1X chip fuses two previously distinct product lines: the Grace CPU architecture (developed for NVIDIA’s data centre Arm CPU product) and the Blackwell RTX GPU architecture (the current flagship data centre and discrete GPU generation). The connection between these two dies uses NVLink C2C — NVIDIA’s proprietary chip-to-chip interconnect — which enables cache-coherent memory access across CPU and GPU, eliminating the discrete memory transfers that constrain performance in traditional CPU-plus-discrete-GPU configurations.
The unified 128GB LPDDR5X memory pool is the most commercially significant capability in the RTX Spark platform for enterprise AI workloads. A 128-billion-parameter AI model requires approximately 128GB of memory to run at full precision — meaning the top-configuration RTX Spark can run a 128B parameter frontier model entirely in local memory without cloud offload. At 1 petaflop of FP4 AI compute, the chip delivers sufficient throughput for real-time inference on these models. For enterprise use cases — local document processing, code generation, agentic task execution — this creates a genuinely new category: a laptop that can run frontier-scale AI inference without a network connection or cloud compute dependency.
NVIDIA has disclosed a multi-generation RTX Spark roadmap alongside the N1X announcement, signalling a sustained commitment to the client PC market rather than a single-product entry. The base N1 variant — intended for thinner, more affordable devices — comes in 12-core (8+4) and 10-core (7+3) configurations with 2,560 and 2,048 CUDA cores respectively, at a 18-45W TDP for the efficiency-focused premium laptop market. NVIDIA’s overall strategy positions RTX Spark directly against Apple Silicon’s dominance in the creator and developer laptop segment, Qualcomm’s Snapdragon X series in the AI PC segment, and Intel and AMD in the mainstream productivity segment.
Gaming performance on RTX Spark is benchmarked at over 100fps in 1440p across triple-A game titles, which NVIDIA claims is the result of the Blackwell GPU’s full DLSS upscaling, ray tracing, and G-Sync support carried over from the discrete RTX 50 series. The chip’s RTX 5070-tier graphics performance in a thin-and-light form factor is the claim that positions RTX Spark as the first platform where creators, developers, and gamers can use a single device without performance compromise — a positioning that directly challenges MacBook Pro’s dominance in the premium portable performance market.
| Specification | NVIDIA RTX Spark N1X | Apple M5 (est.) | Qualcomm Snapdragon X Elite |
| CPU cores | 20 Arm (10 Cortex-X925 + 10 Cortex-A725) | ~12 cores (high/efficiency) | 12 Oryon cores |
| GPU | Blackwell, 6,144 CUDA cores | Apple GPU (38-core est.) | Adreno X1-85 |
| AI compute | 1 petaflop FP4 | Not publicly disclosed | ~45 TOPS |
| Unified memory (max) | 128GB LPDDR5X | ~48GB (est. M5 Pro) | 64GB LPDDR5X |
| Memory bandwidth | 300 GB/s | ~300 GB/s (est.) | 135 GB/s |
| CPU-GPU interconnect | NVLink C2C (cache-coherent) | Apple Fabric (proprietary) | Oryon SoC (integrated) |
| TDP range | 45W-80W | ~25-60W (varies by model) | ~23-45W |
| OS platform | Windows on Arm | macOS | Windows on Arm |
| Launch timeline | Fall 2026 | Late 2025 / 2026 | Currently shipping |
| Competing against | Apple M5, Qualcomm X Elite, Intel/AMD | RTX Spark, Qualcomm | RTX Spark, Apple M5 |
Commercial and Enterprise Market Impact
For enterprise IT procurement, the RTX Spark announcement has two distinct implications. First, it introduces a credible third Windows on Arm platform — joining Qualcomm Snapdragon and Intel’s future Arm-based offerings — with a differentiated value proposition centred on AI compute density rather than power efficiency. Enterprise buyers evaluating AI PC refresh cycles, which accelerated significantly following Microsoft’s Copilot+ PC certification programme, now have an NVIDIA-branded option with a familiar CUDA/RTX software ecosystem that eliminates driver and application compatibility concerns that constrained some early Qualcomm Snapdragon X deployments.
Second, and more strategically, NVIDIA’s entry into client silicon closes the last gap in its full-stack AI infrastructure play. Before RTX Spark, NVIDIA had data centre AI (H100, H200, B200, GB200), edge AI inference (Jetson), and workstation AI (RTX 5000 series discrete GPUs), but no integrated client platform. RTX Spark creates a unified NVIDIA compute continuum from laptop to hyperscale data centre — all running CUDA, all interoperable through NVIDIA’s NIM (NVIDIA Inference Microservices) and NeMo frameworks. For enterprises standardising on NVIDIA’s AI software stack, RTX Spark allows endpoint devices to run the same model inference code as data centre deployments, which eliminates the software complexity of managing different AI runtimes at the edge versus the cloud.
“NVIDIA is doing to the PC what Apple did to the phone — but the mechanism is different. Apple’s advantage was vertical integration of hardware and software. NVIDIA’s advantage is the CUDA ecosystem and enterprise AI software stack. RTX Spark gives enterprise developers a continuous compute environment from laptop to data centre, running the same frameworks. That is a stronger enterprise lock-in argument than Apple ever had.” — Principal Enterprise Architect, AI Infrastructure, Fortune 500 technology company, June 1, 2026
“The 128GB unified memory configuration is the specification that matters for enterprise agentic AI deployments. Any laptop that can run a 128B parameter model locally — without cloud inference, without network latency — changes the architecture of agentic workflows for security-sensitive enterprise use cases where cloud processing is restricted.” — Enterprise AI Platform Analyst, independent research, June 1, 2026
Frequently Asked Questions
What is NVIDIA RTX Spark and when will it be available?
NVIDIA RTX Spark is a Windows on Arm superchip announced at Computex 2026 on June 1, combining an Arm N1X CPU (20 cores, co-developed with Microsoft and MediaTek) and a Blackwell GPU (6,144 CUDA cores) via NVLink C2C interconnect, with up to 128GB unified LPDDR5X memory and 1 petaflop of FP4 AI compute. RTX Spark will launch in fall 2026 in more than 30 laptops and approximately 10 compact desktop PCs from Dell, HP, ASUS, Lenovo, MSI, and Microsoft.
How does NVIDIA RTX Spark compare to Apple M5 and Qualcomm Snapdragon X Elite?
RTX Spark N1X matches or exceeds Apple M5 and Qualcomm Snapdragon X Elite on AI compute (1 petaflop FP4 versus undisclosed M5 figures and ~45 TOPS for Snapdragon X Elite) and unified memory capacity (128GB versus approximately 48GB for M5 Pro and 64GB for Snapdragon X Elite). RTX Spark targets the Windows on Arm ecosystem, competing directly with Snapdragon X Elite for AI PC buyers. For gaming, NVIDIA claims RTX 5070-tier performance — a benchmark neither Apple M5 nor Qualcomm match for triple-A titles.
Why does the 128GB unified memory on RTX Spark matter for enterprise AI?
128GB of unified CPU-GPU memory allows the RTX Spark to run a 128-billion-parameter AI model entirely in local device memory without cloud offload. This is significant for enterprise use cases in regulated industries (legal, financial, healthcare, government) where data cannot be sent to external cloud inference providers. A laptop capable of local frontier-scale AI inference eliminates cloud dependency for security-sensitive agentic workloads, fundamentally changing the architecture of enterprise AI edge deployments.
Sources
CNBC. (2026, June 1). Nvidia jumps into PCs with new Arm-based chip debuting in laptops from Microsoft, Dell, HP. https://www.cnbc.com/2026/05/31/nvidias-new-chip-to-power-fresh-line-of-windows-laptops-by-dell-hp.html
Tom’s Hardware. (2026, June 1). Nvidia unveils RTX Spark Superchip for laptops and desktop PCs at Computex 2026. https://www.tomshardware.com/laptops/nvidia-unveils-rtx-spark-superchip-at-computex-2026
Phoronix. (2026, June 1). NVIDIA announces RTX Spark Superchip for laptops and desktops. https://www.phoronix.com/news/NVIDIA-RTX-Spark
Smartprix. (2026, June 1). Nvidia enters the consumer laptop market at Computex 2026 with its RTX Spark Superchip. https://www.smartprix.com/bytes/nvidia-enters-the-consumer-laptop-market-at-computex-2026-with-its-rtx-spark-superchip/
91mobiles. (2026, June 1). Computex 2026: NVIDIA RTX Spark announced, a new Arm based chipset in collaboration with Microsoft. https://www.91mobiles.com/hub/computex-2026-nvidia-rtx-spark-announced/
TechRadar. (2026, June 1). Watch out Apple — Nvidia just unveiled its RTX Spark Arm superchip. https://www.techradar.com/computing/computing-components/watch-out-apple-nvidia-just-unveiled-its-rtx-spark-arm-superchip-to-take-on-the-m5-at-computex-2026
Benzinga. (2026, June 1). Nvidia has become an infrastructure company: Jensen Huang shows off RTX Spark. https://www.benzinga.com/markets/tech/26/06/52895858/nvidia-infrastructure-company-jensen-huang-computex-2026-ai-factory
