Nvidia’s AI advantage is expanding well beyond the graphics processor, and investors are only now grasping the full scope of that edge. For years, the prevailing view of the company was straightforward: during the early stages of the AI boom, Nvidia was the sole supplier of cutting-edge GPUs, a position that grew enormously profitable as the industry scaled. More recently, hyperscalers such as Amazon and Google began designing their own chips, prompting many on Wall Street to question how durable Nvidia’s lead could really be.
That concern had merit. After its market capitalization grew tenfold between the start of 2023 and mid-2025, Nvidia’s stock settled into a more modest path over the past year, weighed down by worries about mounting GPU competition. But a different picture emerged following the company’s earnings report on Wednesday, as investors recognized that Nvidia’s strengths reach far past the GPU itself.
Orchestration Becomes the New Battleground
As AI compute grows into the gigawatt scale, coordinating all of that hardware has become an increasingly complex challenge. Nvidia has built much of the state-of-the-art equipment required to manage it, giving the company a significant edge in the systems surrounding the GPU even as rivals close in on the processors themselves. Despite frequent talk of compute becoming a commodity, running a megascale data center at peak efficiency remains extremely difficult, and that challenge only intensifies as deployments get larger and faster.
Inside the Vera Rubin Architecture
Nvidia is currently rolling out its Vera Rubin architecture, which pairs the Rubin GPU with a range of additional units, including the Vera CPU, the Groq 3 LPX inference accelerator, and comparable racks for storage and networking. These are highly specialized systems, but rather than processing tokens, they ensure that everything outside the GPU operates as efficiently as possible. If the GPU is the engine, these components form the rest of the car.
The Vera CPU is centered on the problem of 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,” said Jason Hardy, Nvidia’s VP of storage technology.
As data centers have scaled up computing power, memory capacity has grown alongside it, which is part of why companies like Micron have profited during the second wave of the infrastructure boom. Yet delivering that data to the GPU at the right moment is not simple, and as firms push to lower tokens-per-watt, the importance of directing that traffic is becoming clear.
“We saw upwards of 3x improvement in these operations, where the Vera CPU is allowing for acceleration,” Hardy said. “So now we can use our flash to its fullest potential, because we can get all that performance out of it without bottlenecking.”
A Shared Industry Problem
Versions of the same challenge appear elsewhere in the industry. When OpenAI developed its Jalapeño chip, a major goal was to sidestep these bottlenecks by minimizing how much data needs to be moved. “We designed Jalapeño to minimize data movement and communication delays,” the company said.
Source
Image: techcrunch.com
