For much of the AI boom, Nvidia’s edge was primarily attributed to its state-of-the-art GPUs. As the industry scaled, those GPUs generated substantial profits. In recent years, however, hyperscalers such as Amazon and Google have begun developing their own chips, prompting investors to question how durable Nvidia’s lead really is.
Nvidia’s market capitalization rose roughly tenfold between early 2023 and mid‑2025; over the past year its stock has followed a more modest path, partly because of concerns over GPU competition. Since the company’s earnings report on Wednesday, a new narrative has emerged: Nvidia’s advantage reaches well beyond its GPUs.
The challenge: data movement and orchestration at megascale
As AI compute grows toward the gigawatt scale, operating data centers efficiently becomes increasingly complex. Even if compute is treated as a commodity, running a megascale data center at peak efficiency is difficult — and as deployments grow in size and speed, the problem intensifies.
A central issue is getting memory-resident data to the GPU at the right time. That is not straightforward, particularly for teams trying to lower tokens-per-watt.
What Nvidia is selling
Nvidia is rolling out its Vera Rubin architecture, which pairs the Rubin GPU with complementary components such as the Vera CPU, the Groq 3 LPX inference accelerator, and similar racks designed for storage and networking.
These systems are highly specialized. Whereas the Rubin GPU focuses on model compute and token processing, the accompanying hardware is designed to ensure everything outside the GPU operates as efficiently as possible. If the GPU is the engine, these components are the rest of the car.
The Vera CPU is especially aimed at orchestrating data. "Vera is important because there’s only so much memory that you can put in a single server or any sort of compute platform," Jason Hardy, Nvidia’s vice president of storage technology, said. He added that in some operations they have seen "upwards of 3x" improvement thanks to the Vera CPU enabling better use of flash storage without creating bottlenecks.
Alternative approach: minimize data movement
Other companies pursue a different solution. When OpenAI developed its Jalapeño chip, a major design goal was to minimize data movement and communication delays so the whole workload could remain within a single connected system. That reduces the need for complex orchestration by keeping the domain large and self-contained.
Both approaches share the same logic: improve overall efficiency with smarter traffic control instead of relying solely on more processor cycles. But they open different layers of infrastructure for competition.
Competitive implications
This emphasis on data orchestration does not automatically guarantee Nvidia’s success. The company will face competition from other chipmakers and hyperscalers at this orchestration layer just as it has on GPUs. However, early signs suggest Nvidia currently holds a significant lead because it supplies multiple hardware components and integrated system designs rather than only standalone GPUs.
If future gains in efficiency depend as much on moving and orchestrating data as on raw GPU performance, Nvidia’s broader systems play could prove to be a durable advantage.
