Nvidia didn’t just unveil new chips in Taipei. It quietly rewired the economics of the AI server, and the move hits Intel and AMD far harder than the market priced in.

Intel fell 6% and AMD fell 5% on the RTX Spark Superchip, an Arm-based laptop processor Nvidia built with MediaTek and Microsoft that fuses a Blackwell GPU and a CPU onto a single package and reaches premium Windows machines this fall, CNBC reported. Nvidia rose about 4%. The story wrote itself. Nvidia is finally coming for the PC, and the firms that have sold its silicon for 30 years took the hit.

The other chip barely moved the tape, and it is the one that matters. Nvidia put its own CPU, called Vera, at the center of the AI server, the part of the business that actually pays Intel's and AMD's bills.

For 30 years the server processor was a product you bought. On Monday, Nvidia started turning it into a part it hands you inside a system it designed.

The chip is called Vera, and Nvidia is blunt about what it is for. Huang described it as the CPU for the age of agents, and the company put it into full production inside the Vera Rubin platform, the multi-rack system it spent the keynote industrializing.

A single Vera Rubin NVL72 rack pairs 72 Rubin GPUs with 36 Vera CPUs, wired together over NVLink, SiliconANGLE reported. Vera is an 88-core, Arm-based design that Nvidia says completes its target tasks roughly 1.8 times faster than an x86 CPU, according to TradingKey . It is also the first standalone data center processor the company has sold, aimed at what one account sized as a roughly $200 billion market, Seoul Economic Daily reported.

To see why that matters, look at the job the CPU does in an AI server. The GPUs do the heavy math. The CPU feeds them, schedules the work and keeps the expensive accelerators from sitting idle. For years that host socket belonged to a Xeon from Intel or an Epyc from AMD, and it was a good business. Data center processors carry high prices and high margins, and that cash funds the next generation of design. Last quarter AMD pulled a record $5.8 billion from its data center segment and Intel $5.1 billion from its data center and AI unit, according to DCD and Yahoo Finance .

The Socket Nvidia Wants To Take

Vera is built to take that socket. Because Nvidia owns the rack, it can fuse its own CPU to its own GPUs over NVLink, with a coherent memory link that a standard x86 host attached over the slower PCIe bus cannot match. The host stops being a part the customer shops for on the open market and becomes a piece of Nvidia's system. The pitch is not that Vera is a marginally better CPU. The pitch is that the CPU should no longer be a separate purchase at all.

The most demanding buyers already take the integrated path. OpenAI, Anthropic and SpaceX were the first to take Vera chips, with the earliest units hand-delivered in May, The Next Web reported. These are the customers building the largest training and inference clusters on the planet, and they are choosing a unified architecture over the multivendor server they could assemble from cheaper parts. That is supplier-and-customer behavior, not a forecast, which is the kind of evidence worth weighting.

Why Nvidia Owning Part Of Intel Matters

Here is where the obvious story breaks. It is tempting to cast this as Jensen Huang going to war with the companies whose lunch he is eating. The facts do not cooperate. In December, Nvidia closed a $5 billion investment in Intel, taking a stake of roughly 4%, and the two now co-design silicon. Intel is building custom x86 CPUs to Nvidia’s specification for the data center and x86 chips fused with Nvidia GPU chiplets for PCs, according to Fortune and The Tech Portal .

So the x86 that survives inside Nvidia's world increasingly survives as something Intel builds to Nvidia's order. That is the real shape of the move, and it is more durable than any price war. Nvidia is not trying to win the CPU. It is changing what owning the CPU is worth.

Whether the processor in the rack is an Arm-based Vera or a custom x86 that Intel fabricates on contract, the merchant vendor loses the thing that made server silicon lucrative in the first place: pricing power over a socket the customer had no choice but to buy from them.

Markman readers have seen this pattern. When Google flipped the switch on its own features in the 2000s, the small companies that had built businesses on top of Search found they were renting space in someone else's house. Value migrates from the component to the system that integrates it, and capital goes where it is treated best. In the AI rack, it is being treated best at the layer Nvidia controls.

There is a second beneficiary, and the market named it without quite saying so. While Intel and AMD fell on Monday, Arm Holdings jumped 14%, Tech Times reported. Both chips Huang unveiled — the Vera server CPU and the RTX Spark laptop part — run on Arm. The architecture underneath the shift collects a royalty on every socket that moves, whoever ends up making the chip.

What The Market Is Missing

None of this means Intel and AMD are finished, and the honest version of the story says so. Both data center lines still grow — AMD’s by 57% and Intel's by 22% year over year. Intel is also a strategic foundry asset with Washington's attention and now Nvidia's capital behind it. The erosion of the standalone x86 socket is a multiyear structural drift, not a collapse that shows up in this quarter's revenue.

That gap is the mispricing. The market spent Monday reacting to the PC chip, the front it could see, and under-weighted the rack-level shift that decides where the profits in computing pool over the next decade. The premium laptop market Nvidia just entered is real and worth fighting over. The question of who owns the processor at the heart of the AI server is worth far more.

The signal under the noise is simple. In this buildout, the money concentrates with whoever integrates the system and whoever owns the architecture beneath it. Nvidia is now both. For long-term owners, a 5% dip driven by a laptop headline is the kind of volatility to use, not the kind to fear, and Nvidia remains the cleanest way to own the company doing the absorbing. The companies losing the socket will keep printing growth for a while yet. The company turning the socket into a component is the one building the moat.