Nvidia Goes PC-Native — RTX Spark Lands, Marvell Soars
At Computex 2026 in Taipei, Nvidia CEO Jensen Huang introduced the RTX Spark superchip — a Windows PC processor co-developed with MediaTek that pairs a 20-core Arm CPU with a Blackwell GPU delivering 1 petaflop of on-device AI. The same week, Huang declared Marvell Technology the “next trillion-dollar company,” sending its stock surging over 22%. These two announcements put Nvidia at the centre of both personal computing and AI data centre infrastructure simultaneously — a position no chip company has previously occupied.
Laptops and compact desktops powered by RTX Spark are expected from Microsoft Surface, Dell, HP, Asus, Lenovo, and MSI this autumn — with Acer and Gigabyte to follow. Learn about the broader Nvidia Taiwan investment context here.
Inside the RTX Spark Superchip
Click each component below to understand what makes this chip different from everything else in a Windows laptop right now.
RTX Spark internal architecture — schematic view
One Chip, Three Workloads
RTX Spark was designed around three distinct use cases. Select a category to see the specific capabilities and performance figures confirmed by Nvidia.
RTX Spark was specifically designed to run AI agents on-device — no cloud subscription required. Jensen Huang’s phrase was “meter free”: agents running 24/7 without per-query costs. With 1 petaflop of FP4 compute and 128GB unified memory, the chip handles models up to 120 billion parameters with 1 million tokens of context. OpenShell enforces privacy so agents can’t send data to cloud models without user permission.
Adobe is rebuilding Photoshop and Premiere specifically for RTX Spark, targeting 2× faster AI and effects performance. The chip renders 3D scenes exceeding 90GB using OptiX and DLSS, supports 4K AI video generation via ComfyUI, and decodes 12K 4:2:2 video with its dedicated Blackwell hardware decoder. Substance 3D Painter and Stager also run natively. DLSS 4.5 Ray Reconstruction with a second-generation transformer model arrives in Blender 5.3 and dozens of other apps.
The Blackwell GPU’s 6,144 CUDA cores — comparable to a desktop RTX 5070 — target 1440p resolution at over 100 frames per second with full ray tracing, DLSS, and Reflex active. RTX technology is supported in over 1,000 games and applications. DLSS 4.5 Ray Reconstruction arrives in dozens of games. XBOX, KRAFTON (PUBG/NARAKA), NetEase, Riot Games, and Remedy Entertainment are confirmed RTX Spark platform partners. Check out the HyperX Omen RTX 5070 Valorant edition for more RTX gaming context.
Marvell: The Chip Behind the Cloud’s Backbone
On June 2, Huang joined Marvell CEO Matt Murphy on stage at Computex in Taipei and called the company the “next trillion-dollar company” — a statement that sent Marvell’s stock up more than 22% in premarket trading that morning, with some reports citing gains as high as 24.7%, to a price of $273.70. That would add over $47 billion in market capitalisation if the gains held.
Nvidia invested $2 billion in Marvell earlier in 2026 to help customers more easily combine Marvell’s custom AI chips with Nvidia’s networking gear and processors. Marvell specialises in designing high-performance chips for cloud computing, AI infrastructure, 5G carrier networks, enterprise networking, and automotive systems. Its data centre business now accounts for over 75% of total revenue — up from less than 10% in 2016.
In its fiscal Q1 2027 earnings released in May 2026, Marvell posted record quarterly revenue of $2.418 billion, up 28% year-over-year. The company also forecast that its custom chips business alone would surpass $10 billion in revenue in fiscal year 2029, as hyperscale cloud providers accelerate AI data centre buildouts. Marvell’s stock was up over 158% year-to-date as of the Computex announcement. At its current market capitalisation of approximately $192 billion, the company remains well below the $1 trillion mark Huang referenced.
The surge in AI adoption has increased demand for the kind of interconnect and custom ASIC chips Marvell designs — chips that link thousands of processors together inside data centres so they can train and run AI models at scale. See related context on AI infrastructure security challenges here.
How Nvidia Got Here
The RTX Spark and Marvell endorsements didn’t come out of nowhere. Here’s the path that led to Computex 2026.
Who’s Building RTX Spark Machines
Confirmed OEM partners for autumn 2026 launches — laptops and compact desktops. Acer and Gigabyte join the lineup after the initial wave.
From Silicon to Stage — Computex 2026 Covered
This piece covered Nvidia’s RTX Spark superchip announcement at Computex 2026 in Taipei, including its core specifications — a 20-core MediaTek-collaborated Grace CPU, a Blackwell GPU with 6,144 CUDA cores, 128GB of unified LPDDR5X memory, and 1 petaflop of FP4 AI compute. The confirmed OEM partners, the Adobe partnership, and the Microsoft collaboration on Windows security primitives and the OpenShell runtime were also covered.
The piece also addressed Jensen Huang’s public endorsement of Marvell Technology as the “next trillion-dollar company,” Marvell’s Q1 FY2027 revenue of $2.418 billion (up 28% year-over-year), the $2 billion Nvidia investment in Marvell made earlier in 2026, and the resulting 22%+ premarket surge in Marvell’s stock on June 2. Marvell’s custom chips business was forecast to exceed $10 billion in revenue by fiscal year 2029.
For more on Nvidia’s presence in Taiwan and its AI infrastructure investments, see this earlier piece on Nvidia’s Taiwan plans. For the latest in hardware, explore coverage of Samsung’s One UI 8.5 AI rollout and the Oura Ring 5 launch.
