Nvidia's RTX Spark Pushes PC Battle into New Territory as Computex Highlights Shift

Nvidia unveils RTX Spark ARM-based superchip with 1 petaflop AI performance, causing Intel shares to drop 4.1% and signaling a major PC market shift.

The semiconductor universe has rarely seen a single product announcement reorder competitive lines as swiftly as Nvidia's RTX Spark did at Computex 2026. What started as a routine keynote in Taipei morphed into a declaration of war on the desktop CPU duopoly, with Jensen Huang unveiling an ARM-based superchip that packs a petaflop of local AI performance. The reaction across the sector was immediate: Intel shares slid 4.1 percent as the implications of Nvidia's PC-first strategy sank in, while Nvidia's own stock edged up 0.38 percent to 193.64 euros, a modest move that belies the scale of the strategic pivot.

The RTX Spark — codenamed N1X — represents Nvidia's most direct assault yet on the traditional PC processor market. Jointly developed with MediaTek and Microsoft, the chip combines a 20-core Grace CPU with a Blackwell GPU featuring 6,144 CUDA cores, all sharing up to 128 GB of unified LPDDR5X memory on TSMC's 3-nm process. The headline number: one petaflop of FP4 compute, enough to run large language models with 120 billion parameters entirely on-device, with a context window stretching to one million tokens. For professionals, that means 12K video editing and 3D scenes exceeding 90 GB can be handled locally, eliminating the need to shuttle data to distant data centers.

Microsoft’s role in the project goes beyond chip design. The two companies are aligning Windows to support personal AI agents that plan, create content, and orchestrate applications across the operating system. Jensen Huang described the ambition as transforming the PC from a tool into a “team member.” Adobe is already adapting Photoshop and Premiere to tap the new hardware, underscoring Nvidia’s understanding that silicon alone is insufficient — a software ecosystem must make the power tangible to users.

First devices carrying RTX Spark are slated for autumn 2026, with an initial wave of OEMs that reads like a who’s who of the PC industry: ASUS, HP, Dell, Lenovo, MSI, and Microsoft’s own Surface Laptop Ultra. Acer and GIGABYTE are expected to follow. The breadth of partner support suggests Nvidia is not targeting a niche developer audience but the mass market — precisely the turf Intel and AMD have defended for decades.

Nvidia’s share price, at roughly three percent below its 52-week high of 195.06 euros, has gained 5.6 percent on the week and over 20 percent year-to-date. An ex-dividend date of June 4 is approaching, with a payout of 0.25 dollars per share. Cathie Wood’s ARK Investment added 300,000 shares on Monday, a bullish institutional vote of confidence.

The Computex stage also saw Nvidia extend its influence into robotics and autonomous systems. Cosmos 3, an open foundation model for physical AI, was unveiled alongside Isaac GR00T, an open reference platform for humanoid robots. For robotaxis, Nvidia introduced Alpamayo 2 Super, a 32-billion-parameter vision-language-action model targeting Level 4 safety. These announcements broaden the company’s AI footprint well beyond data centers and desktops, though the RTX Spark remains the brightest headline.

The knock-on effects for competitors are stark. Intel, despite showcasing a strong roadmap — Xeon 6+ (Clearwater Forest) on its own 18A process, the E835 Ethernet platform, and a preview of the Crescent Island data-center GPU — saw its stock punished as analysts assessed the RTX Spark threat. The new Nvidia chip attacks both pillars of Intel’s business: consumer PC processors and inference-optimized server CPUs. Intel shares have more than doubled since the start of the year, but the weekly decline of 13 percent and an RSI of 18.9 signal heavy overselling. Morningstar analyst Brian Colello warns that investors may already be pricing in a best-case scenario while competitive and execution risks persist.

Elsewhere at Computex, Marvell Technology received a celebrated endorsement from Jensen Huang, who called the custom-chip specialist “the next trillion-dollar company.” The stock surged 24 percent pre-market to a record 222.65 euros, driven by Nvidia’s two-billion-dollar investment earlier this year and Marvell’s own forecast of over ten billion dollars in annual custom-chip revenue by fiscal 2029. Yet even with a market cap near 192 billion dollars, the trillion-dollar target remains distant.

SK Hynix chose the event to announce plans to double its wafer capacity within five years, with 2026 capital expenditure set to significantly exceed last year’s 30.2 trillion won. The company has also applied to list American Depositary Receipts in New York. Chairman Chey Tae-won cited AI-driven memory scarcity lasting until 2030 as the rationale, though new fabs require over five years to come online. SK Hynix controls roughly 57 percent of the HBM market and 32 percent of global DRAM production, and its market capitalization crossed one trillion dollars for the first time last week.

Infineon, meanwhile, hit a new all-time high of 85.52 euros, doubling its value since January. The catalyst is the upcoming opening of its Smart Power Fab in Dresden on July 2 — a five-billion-euro investment that will produce energy-efficient power supplies for AI data centers, integrating with Nvidia’s MGX platform using gallium-nitride and silicon-carbide components. Infineon’s revenue from the AI data-center segment surged from 250 million euros in 2024 to over 700 million last year, with a target of 2.5 billion by 2027.

The Computex frenzy has drawn a clear line between companies riding the AI infrastructure wave and those caught in its undertow. Nvidia’s PC offensive places it in the former camp, but the stock’s muted reaction hints at the market’s growing caution toward valuations that already price in substantial growth. For Intel, the question is whether an oversold RSI and a credible roadmap can outweigh the existential threat from a competitor that now makes chips for the machine on your desk, not just the servers down the hall.