Jensen Huang stood on stage at Computex 2026 and declared that “40 years of traditional PCs is now at an end.” Then he unveiled RTX Spark — Nvidia’s first PC chip — and positioned it as the replacement for everything Intel and Qualcomm have built, and a direct challenge to Apple’s M5.

This is not Nvidia licensing its GPU to a partner. This is Nvidia designing the entire computer — CPU, GPU, memory architecture, and AI stack — as a single superchip. It is the most significant structural shift in the PC industry since Apple Silicon.

The Specs: What RTX Spark Actually Is

RTX Spark combines a 20-core Grace CPU (Arm-based, co-developed with MediaTek) with a Blackwell RTX GPU packing 6,144 CUDA cores and 5th-generation Tensor Cores with FP4 support. Up to 128GB of LPDDR5X unified memory — double what Apple’s M4 Ultra offers. The GPU and CPU communicate over NVLink-C2C at 600GB/s, eliminating the bandwidth bottleneck that cripples discrete GPU laptops.

AI performance: 1 petaFLOP at FP4 precision. For context, that’s data center-class AI compute in a form factor that scales from single-digit wattage (tablet mode) to 80W (desktop). Memory bandwidth hits 300GB/s.

Dell, HP, Lenovo, Microsoft, Asus, and MSI have all committed to shipping RTX Spark devices. First systems arrive before holiday 2026. Broader availability in early 2027.

Why This Matters: Nvidia Owns the Full Stack

The strategic significance isn’t the specs — impressive as they are. It’s that Nvidia is now vertically integrated from data center to laptop.

Consider what Nvidia controls after this announcement: the GPU training the AI models (H200, Blackwell, Vera Rubin), the networking connecting them (InfiniBand, Spectrum-X), the software running them (CUDA, TensorRT, NeMo), and now the client device that deploys them locally (RTX Spark). A developer can train a model on Nvidia’s cloud hardware, optimize it with Nvidia’s software, and deploy it to an Nvidia-powered laptop — never leaving the ecosystem.

Apple achieved something similar with its own silicon stack. But Apple’s advantage is limited to its own hardware. Nvidia’s RTX Spark runs Windows — which means it plugs into the enterprise ecosystem that Apple has never cracked at scale.

The Apple M5 Problem

Apple’s M-series chips redefined what a laptop could do. Unified memory, power efficiency, and integrated GPU performance set a standard that Intel and AMD spent three years failing to match. Qualcomm’s Snapdragon X Elite made progress but couldn’t close the gap.

RTX Spark is the first chip that doesn’t just compete with Apple Silicon — it attacks from a position Apple can’t match. Apple has no answer to 6,144 CUDA cores, 5th-gen Tensor Cores, or NVLink-C2C bandwidth. Apple Intelligence runs a 3-billion-parameter on-device model. RTX Spark can run models 10-50x larger locally, with real-time ray tracing and DLSS on top.

The 128GB unified memory ceiling is particularly aggressive. Apple’s M4 Ultra maxes out at 192GB, but that’s a $5,000+ desktop chip. RTX Spark brings 128GB to laptops — meaning local LLM inference with 70B+ parameter models becomes a laptop workload, not a workstation one.

The MediaTek Partnership: Why Not AMD or Qualcomm?

Nvidia chose MediaTek — not AMD, not Qualcomm, not Samsung — to co-develop the Grace CPU. This is strategic for three reasons.

First, MediaTek has no competing GPU business. AMD (Radeon) and Qualcomm (Adreno) would be co-developing a CPU for a chip that competes with their own GPUs. MediaTek has no such conflict.

Second, MediaTek dominates the smartphone SoC market in volume — over 40% of Android devices globally. They understand Arm architecture at a scale that few companies match.

Third, MediaTek gives Nvidia a path into mobile. If RTX Spark succeeds in laptops, a MediaTek partnership could extend to tablets, phones, and embedded devices — markets Nvidia has never competed in.

“Agentic AI OS” — What Jensen Really Means

The phrase Jensen used — “turn Windows into an agentic AI OS” — is not marketing fluff. It describes a specific technical vision: a PC where AI agents run continuously in the background, managing workflows, processing context, and taking actions on behalf of the user.

This requires two things current PCs lack: enough local compute to run large models without cloud latency, and enough memory to maintain persistent context across sessions. RTX Spark’s 1 petaFLOP and 128GB address both.

The competitive implication is clear. Intel and Qualcomm are selling “AI PCs” with NPUs that handle basic tasks — image generation, voice transcription, simple summarization. Nvidia is selling a PC that can run the same models deployed in data centers. It’s not an incremental improvement. It’s a category difference.

How AI Is Changing This

AI is fundamentally reshaping the competition between Nvidia’s RTX architecture and Apple’s upcoming M5 chip, with each taking distinct approaches to neural processing workloads. Nvidia’s RTX 4090, for example, leverages dedicated Tensor cores alongside 16,384 CUDA cores to deliver exceptional performance in AI training tasks like fine-tuning large language models, where its 24GB VRAM allows handling of massive datasets. Meanwhile, Apple’s anticipated M5 chip is expected to integrate enhanced Neural Engine units directly into its unified memory architecture, optimizing for on-device AI inference rather than training. This divergence reflects broader industry trends: Nvidia dominates datacenter AI training with raw computational power, while Apple focuses on efficient, privacy-preserving AI execution for consumer applications. The M5’s unified memory design will likely excel at real-time AI features like computational photography and voice processing, whereas RTX maintains superiority in demanding creative workflows requiring substantial parallel processing power.

What Happens Next

RTX Spark ships holiday 2026. If it delivers on the specs, the PC market splits into three tiers: Apple Silicon for the creative/consumer ecosystem, RTX Spark for AI-first Windows professionals, and legacy x86 for everyone still on the upgrade treadmill.

Intel is the biggest loser. The company that defined the PC for four decades now faces an Nvidia chip that makes x86 irrelevant for the highest-value workloads. AMD retains relevance through gaming GPUs and EPYC servers, but in the laptop market, RTX Spark leapfrogs everything AMD offers.

Jensen wasn’t being hyperbolic. The traditional PC era — defined by x86 CPUs with discrete or integrated GPUs — is ending. What replaces it is a superchip architecture where CPU, GPU, and AI accelerator are one unified system. Apple proved the concept. Nvidia just scaled it to a level Apple can’t match.

For the full structural map of the AI economy, read The Map of AI Redrawn on Business Engineer.