• At GTC Taipei, NVIDIA unvieled the highly anticipated PC platform,  the RTX Spark Superchip, delivering up to 1-Petaflop of AI performance, and the Full CUDA and RTX Ecosystem to the Windows PC platform
• NVIDIA's entry into the AI PC market has the potential to reshape what has become a relatively mature PC industry by by creating a new category of local AI Inference Machines.
• According to Counterpoint's Arm Laptop Tracker and Forecast, Arm based laptops are expected to account for a 33% share of the global laptop market by 2030.

At GTC Taipei, NVIDIA unveiled the highly anticipated RTX Spark Superchip for PCs, ending months of industry speculation . RTX Spark combines  NVIDIA's Blackwell-based GPU with an Arm-based  CPU co-developed with MediaTek. The device is expected to handle a variety of AI workloads locally in NVIDIA's CUDA-based ecosystem. This can be seen as an attempt to convert the PC into a Local AI Inference Machine, rather than simply providing a new ARM-based chip.

Full-fledged Challenge to X86 PC

Until now, the PC market has been dominated by x86 with Arm-based devices serving primarily niche segments. The most significant milestones in Arm’s expansion into the PC market has been Appple’s successful transition to its M series chip and Qualcomm's entry through the X-series.

Apple was the first to announce its x86 split with the announcement of the 'M1' chip in 2020 and has now successfully transformed its entire Mac lineup into its own ARM-based Apple Silicon (M-Series).

While Apple has successfully taken  some share from x86 through its own macOS ecosystem, Qualcomm startegy is more distruptive because it is targeting the Windows PC x86 market directly. Qualcomm has formed a strong strategic partnership  with Microsoft through the launch  of its Snapdragon X Elite and Snapdragon X Plus platforms based on its own powerful architecture (Oryon CPU). With Microsoft specifying laptops with Qualcomm chips as the standard reference for Copilot+ PCs, the Windows ARM ecosystem instantly rose to the mainstream.

Although Arm based PCs settled into the market thanks to the performance of Apple and Qualcomm, Intel and AMD's x86 (a combination of discrete GPUs) remained firmly intact in the areas of high-performance desktops, AAA-class gaming, professional deep learning development and 3D graphics. It was difficult to completely replace this heavy workload with Qualcomm's Adreno or Apple's built-in GPUs. Unlike its competitors, Nvidia has retained its top-tier Blackwell architecture for the new chip(RTX Spark). NVIDIA's entry into the the AI PC market has the potential to rechape what has become a relatively mature PC industry by creating a new category of local AI Inference Machines.

It can argued that  Apple's M-series devices were among the first to demonstrate the PC industry’s transition towar becominng a Local AI Inference Machine. Apple's M-series based devices are now the most realistic and powerful alternative to "local AI" workstations based on their hardware structural strengths (unitifed memory architecture) and strong open-source ecosystem support.

Wth a system architecture that leverages many of the advantages that have made Apple silicon successful for AI workloads, including  a unified memory architecture a high performance GPU and efficient power management, the RTX Spark has the potential to become one of the most compelling platforms for local AI comouting. Combine these advantages with NVIDIA’s AI software stack, including CUDA. TensorRT and its extensive developer community, the RTX Spark could emerge as a  premier deice for running LLMs, AI agents and other generative AI workloads directly on the PC.

Why ARM matters in AI PC era

As AI computing evolves, the expectation is that AI workloads will shift from cloud to local devices. In implementing this local AI workload, the Arm camp has a clear advantage over x86.

• Arm based architectures leverage low-power and high-efficiency mechanisms that dramatically reduces heat and power consumption during local AI inference workloads.
• High-performance Arm chips that are disrupting the market, such as Nvidia's RTX Spark and Apple's M-Series, have a few things in common: the CPU, GPU, and NPU use a unified memory structure that shares the entire memory within a single chip. This will lead to directly processing data with ultra-high bandwidth of hundreds of GB/s, demonstrating overwhelming response rates when loading and running multibillion parameter Large Language Models (LLMs) or real-time Generative AI apps.

The problems of application compatibility and performance limitations of the existing ARM family are gradually being resolved, and the ARM camp is gradually taking some share from x86.

Why NVIDIA is different from current Arm players

NVIDIA's entry into the market is expected to be an opportunity to accelerate this change. The advantages of RTX Spark over the existing Arm camp are as follows.

• Performance optimized for local AI workload based on high hardware performance (GPU performance, memory bandwidth)
• Access to a wider software ecosystem compared to a closed ecosystem called Windows on Arm
• High interworking with existing CUDA ecosystems can minimize software bottlenecks in AI workloads
• NVIDIA can connect with various product ecosystems (AI servers, CUDA SW, Foundation Model)

What needs to be proved by new chips

There are still some challenges that need to be addressed in order for these NVIDIA attempts to achieve meaningful results for the market.

• The maturity of the Windows on Arm ecosystem that can properly utilize the existing software in the Windows market will have a great influence on the utilization of this device.
• Entry to the right price segment in light of the higher costs. Adoption of high-performance hardware  will inevitably lead to high prices. In an environment where overall PC prices are rising, the key will be in which areas NVIDIA's chip-equipped devices can be positioned and provide a meaningful differentation.  
• Popularization of Local AI Inference. In order for the device not to remain as a niche market target device for developers or some AI-related tasks, local AI Inference work will need to be popularized. This will require many changes, such as the maturity of the software ecosystem and changes in consumer behavior. This is expected to take some time