At Computex in Taipei, Jensen Huang unveiled not another data-center accelerator but a chip designed for laptops: the RTX Spark, Nvidia's first integrated PC platform chip combining CPU and GPU.
For three decades, Nvidia has made graphics cards; now it has built a complete PC chip with both CPU and GPU on one package. The RTX Spark was officially introduced on June 1 at Computex.
A chip that brings data-center capabilities to laptops
The hardware specs are striking:
- 20-core Arm CPU paired with a Blackwell GPU (6,144 CUDA cores)
- Connected via Nvidia's NVLink C2C chip-to-chip interconnect
- 128GB LPDDR5X unified memory with up to 300 GB/s bandwidth
- AI performance up to 1 petaflop — one quadrillion floating-point operations per second
The 128GB unified memory is the real game-changer. Typical laptop memory tops out at 8 to 24GB, limiting local large-model AI. With 128GB unified memory, a laptop can run substantial models locally, enabling agentic AI — assistants that autonomously break down tasks — directly on the device without sending data to the cloud.
Nvidia's positioning is clear: turn Windows into an "AI operating system that can work on its own," with the official goal to "redefine the PC."
Direct challenge to Apple and Intel
The RTX Spark runs on Windows on Arm. It will launch this fall in over 30 laptops and 10 desktops from partners including Dell, HP, Lenovo, Asus, and MSI, with the thinnest design at 14mm. Notably, Microsoft is also on the list, planning a new Surface Ultra powered by the RTX Spark.
This is significant because Microsoft has been pushing Copilot+ PCs powered by Qualcomm's Snapdragon. Now it is placing a separate bet on Nvidia for its flagship Surface. At the same Computex, Qualcomm announced the Snapdragon C, targeting affordable entry-level AI laptops, with Acer's Aspire Go 15 already adopting it.
Nvidia goes high-end with 1 petaflop; Qualcomm goes low-end with budget machines. The PC landscape, once dominated by Apple M-series, Intel, and AMD, now sees Nvidia and Qualcomm entering the fray.
Nvidia has not disclosed pricing, but said the first wave targets the high-end, with lower-memory versions for mid-range and budget segments later.
Jensen Huang wants to win both cloud and edge
In the same keynote, Nvidia announced that its data-center Vera CPU is already in full production, shipping this fall, with OpenAI, Anthropic, and SpaceX as early customers.
The strategy is clear: on the cloud side, Nvidia sells rack-scale monsters like Vera Rubin; on the edge, it now reaches into every laptop with the RTX Spark. One foot in AI factories, the other in consumer pockets.
What is the moat? Not benchmark scores. Apple's M-series wins on efficiency and a tightly integrated ecosystem. Nvidia's advantage is CUDA — a platform developers have used for nearly two decades, making switching costs prohibitively high. The RTX Spark's real test is whether developers and consumers will pay a premium for local agentic AI on laptops.
That answer will come when the first machines ship this fall. But one thing is certain: Nvidia is no longer content selling only the card inside the chassis.
Sources: Tom's Hardware, Engadget, CNBC, TechRadar, Digit; CocoLoop