NVIDIA has sold off its 'last bits' of stake in ARM this week, coming a long way from actually acquiring the company a few years ago.
NVIDIA's x86 Deal With Intel Is an 'Indirect Indication' That ARM Alone Cannot Handle the Agentic AI Buildup
NVIDIA's partnership with ARM has been pivotal in the modern-day AI buildout, as the company's CPU architecture has enabled it to deliver impressive offerings under the Grace Hopper and Blackwell lineups. More importantly, ARM will also play an essential role in NVIDIA's upcoming Vera CPUs, as the importance of such processing units has been rising significantly. Bloomberg reports that NVIDIA sold out its remaining stake in ARM, worth $140 million, according to the latest SEC filings, but interestingly, this move also comes at a time when ARM's role in the future of the AI race is being questioned.
For those unaware, CPUs are gaining immense traction in recent times, given inference, especially agent-based, as shown with such workloads, the primary focus shifts away from GPU compute to more CPU-focused tasks, such as tool calls, API requests, memory lookups, and orchestration logic. This pivot is evident in the fact that both Intel and AMD are reporting immense hyperscaler demand for their data center CPUs, driven by the aggressive growth of the CPU TAM. At the same time, the role of ARM's architecture is also being talked about, and here's why GF Securities says:
GFHK noted in their Q&A following their February report that ARM-based CPUs have relatively weak momentum in AI servers, attributing this to lower GPU scheduling efficiency compared to x86. They indicated that companies including NVIDIA plan to develop x86 CPU solutions in response.
Now, there are various reasons why x86 is claimed to be superior to ARM when the scope is limited to agentic workloads. But one of the bigger bottlenecks comes from the fact that, with agents in particular, the focus is more on 'single-threaded burst speeds' than on the advantages of multi-core configurations, given that the former prevents the GPU from sitting idle while waiting for instructions. In an agentic environment where millions of microtasks are executed per second, even a slight delay can cause a noticeable bottleneck.
Another major advantage, apart from the technicals on ARM vs x86, is that several enterprise data centers have their ecosystems focused on x86, including firmware stacks, virtualization layers, and years of software compiled. This is another major reason why Intel/AMD see tremendous demand for their server products in recent times: hyperscalers are in an upgrade cycle right now, which is why, for NVIDIA, an x86 server rack makes sense as well.
Well, if you think this is an opinion, it is backed by the fact that NVIDIA is exploring the x86 route as well, and in their latest partnership with Intel, the idea is to bring an x86-equivalent solution to market by integrating it into NVLink-fused server racks. Of course, the ARM stake situation has little correlation with how NVIDIA plans to capitalize on the growing CPU TAM. It is reported that NVIDIA's decision to offload its remaining ARM stake is a pure 'financial' play and has little to do with the company's overall product strategy.
For now, Vera CPUs are entirely ARM-based, but diversification of CPU offerings towards Intel's x86 ecosystem could occur with Feynman or later.
About the author: Muhammad Zuhair is a hardware and technology reporter for Wccftech, specializing in the semiconductor industry and the complex interplay between technology, manufacturing, and geopolitics. His coverage focuses on the corporate strategies and technological roadmaps of industry giants like TSMC, NVIDIA, Samsung, and Intel. Zuhair's expertise lies in deconstructing complex topics such as fabrication nodes (e.g., 2nm process), the economic impact of policies like the CHIPS Act, and the strategic development of AI infrastructure from NVIDIA, AMD and Intel.
Follow Wccftech on Google to get more of our news coverage in your feeds.
