Hajra Naz
Before this week, a dominant story was Nvidia. It went something like this. For the first few years of the AI boom, Nvidia was the only real source for state-of-the-art GPUs. That position became incredibly profitable as the industry scaled rapidly. In recent years, though, things changed. Major hyperscalers, like Amazon and Google, started building their own chips. Nvidia was no longer the only option available. This shift led many investors to ask a real question. How durable is Nvidia’s advantage, actually?
This story is compelling and mostly accurate. Nvidia’s market cap grew roughly 10x between the start of 2023 and mid-2025. Since then, growth has slowed noticeably. Over the past year, shares have followed a more modest trajectory. Much of that slowdown reflects growing concerns about GPU competition.
A New Narrative Emerges
Since Nvidia’s earnings report on Wednesday, a new narrative has taken shape. Investors are starting to realize something important. Nvidia’s advantage extends far beyond GPUs alone. As AI compute scales into the gigawatt range, orchestrating that compute has become genuinely complex. It’s no surprise that Nvidia has built much of the state-of-the-art hardware needed for this task. This gives the company a major advantage. That advantage exists in the systems surrounding the GPU itself, even as competition on GPUs alone intensifies.
There’s been plenty of talk about compute becoming a commodity. Still, operating a megascale data center at peak efficiency remains incredibly difficult. As deployments grow bigger and faster, that challenge only becomes harder to solve.
Read More: Amazon just tripled its Nvidia chip orders over rising AI demand
Rack by Rack
You can see this trend by looking at what Nvidia actually sells. The company is currently rolling out its Vera Rubin architecture. This system pairs the Rubin GPU with several other components. That includes the Vera CPU. It includes the Groq 3 LPX inference accelerator. Similar racks handle storage and networking too.
Over the past week, conversations with people at Nvidia revealed something surprising. These surrounding systems are extremely specialized, much like the Rubin GPU itself. But instead of processing tokens directly, they focus on something different. They ensure everything outside the GPU runs as efficiently as possible. If the GPU functions like an engine, these systems represent the rest of the car.
The Vera CPU specifically addresses the challenge 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. That’s part of why companies like Micron have profited significantly during this second wave of infrastructure growth. Still, getting data to the GPU at exactly the right moment isn’t simple. As companies push to lower tokens-per-watt, they’re discovering just how important smart traffic direction really is.
“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.”
Read More: Nvidia Reportedly Moves Toward Hugging Face Acquisition
A Similar Challenge Beyond Nvidia
This same underlying problem shows up outside of Nvidia too. When OpenAI developed its Jalapeño chip, a major design goal was avoiding this challenge entirely. The company focused on minimizing how much data needed to move around in the first place.
“We designed Jalapeño to minimize data movement and communication delays,” OpenAI wrote in a blog post earlier this month. “Its large domain allows the entire workload to remain within one connected system, minimizing data movement and helping the complete request stay fast and efficient from beginning to end.”
This represents a different approach. Instead of directing data efficiently, OpenAI avoids unnecessary movement altogether. This happens by keeping workloads within one integrated chip. Still, the underlying logic remains the same. Efficiency improves through smarter traffic control, not simply more processing power. This shift opens up an entirely new layer of infrastructure. Companies now have a new arena to compete within.
Nvidia’s Early Lead
This new focus on data orchestration doesn’t guarantee Nvidia an automatic win. The company will need to compete here too. That includes rival chipmakers. It also includes major hyperscalers, just as it has with GPUs before. Still, the nature of competition has shifted to a new layer. Building a rival GPU now matters less on its own. What matters more is making the entire system work efficiently together.
At least in these early stages, Nvidia appears to hold a commanding lead.
