Prior to this week, the prevailing narrative surrounding Nvidia was that it thrived as the singular provider of cutting-edge GPUs, reaping significant rewards amid the rapid growth of the AI sector. However, with major players like Amazon and Google beginning to develop their own chips, Nvidia’s status as a market leader has been called into question, prompting concerns about the sustainability of its competitive edge.
This story has been compelling and largely accurate. Following a tenfold market capitalization increase from early 2023 to mid-2025, Nvidia’s stock has exhibited a more restrained pattern in the past year as fears surrounding GPU competition have mounted.
A fresh perspective has emerged following the company’s recent earnings report on Wednesday, prompting investors to recognize that Nvidia’s strengths extend beyond just GPUs. As the demand for AI computing reaches gigawatt levels, managing orchestration has become an increasingly intricate task. Unsurprisingly, Nvidia has engineered much of the advanced hardware essential for this task, giving it a significant advantage in the systems that support GPUs, even amidst rising competition in the GPU market.
Though computing is often discussed as a commodity, operating a large-scale data center efficiently remains a complex challenge, which intensifies as deployments expand and accelerate.
Rack by Rack
This complexity is evident when examining Nvidia’s product offerings. The company is presently introducing its Vera Rubin architecture, which integrates the Rubin GPU with various other components, such as the Vera CPU and the Groq 3 LPX inference accelerator, along with dedicated systems for storage and networking.
In recent discussions with Nvidia representatives, I discovered that these systems perform surprisingly well. Like the Rubin GPU, they are highly specialized; however, instead of solely focusing on processing, they enhance the overall efficiency of the operations surrounding the GPU. If the GPU acts as the engine, these components serve as the essential systems that keep it running smoothly.
The Vera CPU specifically targets the orchestration of data. “Vera is crucial because there is a limit to the memory that can be allocated in a single server or compute platform,” stated Jason Hardy, Nvidia’s VP of storage technology.
As data centers scale their computing power, memory requirements have also increased, contributing to the success of companies like Micron in this second wave of technological growth. However, relaying that data to the GPU efficiently poses challenges. As organizations strive to maximize performance metrics like tokens-per-watt, the need for effective data traffic management becomes evident.
“We’ve observed up to a threefold improvement in these processes, thanks to the acceleration provided by the Vera CPU,” Hardy mentioned. “Now we can fully utilize our flash storage without encountering bottlenecks.”
Similar challenges have been recognized by others outside of Nvidia. For instance, when OpenAI developed its Jalapeño chip, a core objective was to circumvent these issues altogether by limiting data movement.
“Our design for Jalapeño was aimed at minimizing data transfer and communication lags,” the company detailed in a recent post. “Its extensive domain ensures that the entire workload stays within one connected system, thus reducing data movement and enhancing efficiency throughout the process.”
This strategy takes a different angle, conducting workloads entirely within a single integrated chip, thereby avoiding the need for extensive data transfers. Nevertheless, the underlying principle remains the same: enhancing efficiency through intelligent traffic management rather than simply increasing processing power. This sets the stage for a new level of competition in the infrastructure landscape.
This refined focus on data orchestration doesn’t guarantee victory for Nvidia. The company will face competition from other chip manufacturers and hyperscalers, similar to its GPU challenges. However, the competitive landscape has shifted; building a rival GPU may matter less than ensuring that entire systems operate efficiently.
In these early stages, Nvidia appears to maintain a substantial advantage.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
