As AI data centers continue to expand, discussions about NVIDIA’s moat are evolving. TechCrunch noted that investors previously focused on whether cloud providers’ in-house chips would erode NVIDIA’s GPU dominance, but following the latest earnings report, the market has shifted its attention to a new question: who can efficiently operate an entire AI system.
The focus of competition has shifted to the system.
The article argues that AI computing power deployment is moving toward higher power and larger clusters, and focusing solely on GPUs is no longer sufficient to explain the competitive landscape. As data centers grow larger, the coordination between data, storage, networking, and computation becomes increasingly complex. While GPUs remain a core component, system capabilities beyond GPUs are increasingly determining efficiency.
This is at the heart of NVIDIA’s current narrative shift. Even as major cloud providers like Amazon and Google continue advancing their own chip development, NVIDIA is building a more comprehensive infrastructure ecosystem around its GPUs. The article summarizes this as: if the GPU is the engine, then the rest of the system is the entire car.
Vera Rubin covers more stages
TechCrunch reports that NVIDIA is advancing the Vera Rubin architecture. This architecture includes not only the Rubin GPU but also the Vera CPU and rack-level systems designed for storage, networking, and other components. The article suggests that the significance of these products lies in the fact that they are not directly responsible for generating tokens, but rather in minimizing performance overhead outside of the GPU.
Among these, the Vera CPU is described as a crucial component in data scheduling. Jason Hardy, Vice President of Storage Technologies at NVIDIA, stated that the amount of memory that can be accommodated in a single server or computing platform is limited, making it essential to deliver data to the GPU at the right time to maximize system efficiency.
Data movement becomes the new battlefield
The article cites Hardy as saying that, in certain related operations, the Vera CPU can provide a speedup of approximately 3x. This means hardware resources such as flash memory can be utilized more efficiently, and system bottlenecks may be reduced.
The article also notes that similar issues are not unique to NVIDIA's ecosystem. When OpenAI previously introduced the Jalapeño chip, it also listed reducing data movement and communication latency as key goals. The approach aims to keep entire workloads within a single interconnected system to minimize losses from cross-module data transfers.
According to TechCrunch, this indicates that the competition in AI infrastructure is entering a new phase. Companies are no longer just competing on processor power, but also on how data flows, how storage is optimized, and how networks avoid congestion. In other words, the competition is shifting from “who has stronger GPUs” to “who can make the entire system operate more efficiently.”
The article argues that this change does not mean NVIDIA has already secured victory; it still faces competition from other chip manufacturers and major cloud service providers. However, at this stage, NVIDIA remains ahead in system integration and orchestration capabilities, which is a key reason the market is reassessing its AI advantages.