For years, the death of x86 was a prediction made mostly by people who wanted it to be true. On Tuesday in Taipei, it got a product launch. Nvidia took the Computex 2026 stage and unveiled the RTX Spark — a 20-core Arm-based Grace N1X processor co-developed with MediaTek and fabbed on TSMC's 3-nanometer process — and the market response was immediate and unambiguous. Arm Holdings surged more than 11% in pre-market trading, adding billions to a valuation that has already been one of the more remarkable stories in the semiconductor space over the past twelve months.

The RTX Spark is not a niche workstation curiosity. Nvidia has lined up more than 30 laptops and at least 10 desktop systems from Microsoft, Dell, HP, Asus, Lenovo, and MSI — the full roster of the Windows PC industrial complex — to ship hardware built around it. The chip integrates Blackwell GPU architecture with up to 128GB of LPDDR5X unified memory in a unified memory pool design, meaning the CPU and GPU share the same memory space rather than shuttling data across a bottleneck. That is the design choice Apple made in 2020 when it launched the M1. The rest of the industry spent five years calling it a niche. Now Nvidia is doing it for Windows, at scale, with the biggest OEMs on earth signed up.

The unified memory architecture matters more than the headline core count. Artificial intelligence inference workloads — the kind that will define the next generation of on-device computing — are notoriously memory-bandwidth-hungry. Running a capable large language model locally on a laptop requires moving enormous amounts of data between compute and memory constantly. Discrete GPU architectures, where the CPU and GPU maintain separate memory pools, introduce latency and bandwidth ceilings that become a fundamental performance constraint. Blackwell plus 128GB of shared LPDDR5X is a direct answer to that problem, and it is an answer built on Arm instruction sets, not Intel or AMD's.

Arm's business model means it collects a royalty on every chip that ships using its architecture and an upfront licensing fee from every company that designs one. Nvidia's volume commitments alone would be significant. But the RTX Spark announcement arrives alongside separate confirmation that ByteDance and Oracle have both adopted Arm's in-house AGI CPU — Arm's own silicon, not a licensed design — for data center workloads. That is a qualitatively different signal. ByteDance is one of the most computationally intensive companies on earth; Oracle's cloud infrastructure serves enterprise customers who have historically been the most conservative adopters of new architecture. Both choosing Arm-native silicon for AI compute is not a pilot program. It is a purchasing decision made at scale by organizations that cannot afford to bet on a losing architecture.

The combined picture — consumer PCs, laptops, desktops, and hyperscale data centers — represents the full stack of computing, top to bottom, pivoting toward Arm simultaneously. The establishment technology press has covered each of these announcements as separate business stories. Taken together, they describe something more structural: the x86 duopoly that Intel and AMD have maintained for decades is under coordinated pressure from multiple directions at once, and Arm is the common denominator in every alternative.

Nvidia's decision to co-design with MediaTek is worth examining on its own terms. MediaTek is the world's largest mobile chip designer by volume, with deep expertise in Arm-based SoC integration and aggressive manufacturing relationships with TSMC. Bringing MediaTek into the RTX Spark design partnership gives Nvidia access to integration knowledge built across billions of shipped mobile chips while MediaTek gains association with Nvidia's AI platform and access to the premium PC market it has historically not competed in. Both companies get something they could not easily build alone. The collaboration also signals that the competitive boundaries between mobile, PC, and AI accelerator silicon are dissolving — the chip categories that defined the last two decades of the industry are converging into a single design space.

For Arm Holdings specifically, the market surge reflects a recognition that the company's royalty stream is no longer dependent on any single end market. A year ago, the bear case on Arm was that smartphone growth was plateauing and PC penetration was uncertain. The Computex announcements, combined with the hyperscaler AI CPU adoption, effectively close both legs of that argument simultaneously. Arm's own internal projections — which the company has discussed in investor communications — point toward a trajectory of sustained royalty growth as average selling prices of Arm-based chips rise with AI integration.

None of this is without risk. Microsoft's Windows on Arm push has produced software compatibility friction before; the ecosystem of legacy x86 applications does not vanish because a new chip ships. Intel and AMD are not standing still, and both have significant roadmaps targeting AI PC workloads. TSMC's 3nm capacity is constrained and contested by every major chip designer simultaneously. But the direction of force is now unmistakable. When Nvidia, the company that has extracted more value from the AI hardware wave than any other, places its premium PC architecture bet on Arm and not x86, it is not making a speculative call. It is ratifying what the data center buildout has been telling anyone willing to read it for two years. The regime change is not coming. It shipped at Computex.