Jensen Huang has never walked onto a stage without a leather jacket or a superlative. At Computex 2026, he brought both — plus a chip that might actually deserve the hype.
NVIDIA unveiled the RTX Spark Superchip: a 70-billion-transistor, TSMC 3nm monster that fuses a Blackwell RTX GPU (6,144 CUDA cores with fifth-generation Tensor Cores) to a 20-core NVIDIA Grace ARM CPU via NVLink-C2C chip-to-chip interconnect. The whole package gets 128GB of shared LPDDR5X RAM with up to 300 GB/s of memory bandwidth. NVIDIA rates it at up to 1 petaflop of AI compute. That is, to put it plainly, a lot of math for something you can carry in a bag. Built in partnership with MediaTek, the RTX Spark is the foundation of a full Windows-on-Arm platform targeting laptops and mini-PCs. ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI have all committed to Spark-powered devices arriving in fall 2026.
The key architectural story is unified memory: the CPU and GPU share one big, fast pool of RAM, obliterating the "do I load this model into system RAM or VRAM?" headache that haunts every developer running local AI inference. At 128GB, you can run genuinely large models without chunking, swapping, or praying to the GPU gods. NVIDIA also dropped a three-generation roadmap: a Vera Rubin Spark with LPDDR6 is next, followed by a Rosa Feynman Spark beyond that. The naming convention suggesting they plan to do this for a while.
This matters because it signals the end of "cloud-only" as a default assumption for AI workloads. A petaflop in a laptop means local inference on models that would have required cloud API calls a year ago. For developers, it means testing agentic workflows on hardware that actually mirrors production. For enterprises with data sovereignty concerns, it makes on-device AI a credible architectural choice rather than a trade show promise. And for anyone who has suffered through cloud API latency mid-demo, the appeal of running everything locally at this speed is visceral.
If Apple Silicon redefined what a laptop could feel like, RTX Spark is NVIDIA's argument about what a laptop should be capable of. Fall 2026 is going to be a very interesting time to buy a PC.
Source: Tom's Hardware