Nvidia unveils RTX Spark ARM superchip while Intel bets on CPU resurgence for AI workloads

At Computex in Taipei, Nvidia announced its RTX Spark Superchip, an ARM‑based platform that combines a 20‑core Grace CPU with a Blackwell GPU featuring 6,144 CUDA cores. Built on TSMC's 3‑nm process, the chip delivers up to one petaflop of AI compute and can run large language models with 120 billion parameters locally, supporting a context window of one million tokens. Nvidia plans to ship the first systems in autumn 2026 with partners such as Microsoft, Dell, HP, Lenovo, ASUS and MSI, while noting that Windows‑on‑ARM kernel‑mode driver support is still pending. The launch pushed Nvidia’s share up about six percent, while Intel’s stock fell roughly 4.7 percent.

Meanwhile, Intel is highlighting a shift in AI data‑center workloads from pure GPU training to inference and agent‑based systems that demand more CPU involvement. The company frames this as a potential CPU renaissance, emphasizing its 18A and upcoming 14A process nodes for its expanded foundry business. Intel sees CPUs moving from a 1‑to‑8 CPU‑to‑GPU ratio toward parity as AI models become more autonomous, and it is positioning its new manufacturing capacity as a counter‑balance to the high utilisation of rivals such as TSMC.

Both moves underline a broader industry trend: competition over the next generation of AI‑centric silicon, with Nvidia targeting the ARM‑based PC market and Intel reinforcing its CPU and foundry roadmap to retain relevance in AI‑heavy data‑center deployments.