Nvidia's real AI edge is not the chip. It's everything around it.

Key points

• Nvidia's market cap grew roughly 10 times between January 2023 and mid-2025, then levelled off as rivals built competing chips.
• Nvidia is now rolling out its Vera Rubin architecture, a full system pairing the Rubin GPU with a dedicated CPU, a fast-inference accelerator, and specialised storage and networking hardware.
• The Vera CPU, the system's data-traffic manager, delivered up to 3x speed improvements in internal tests, according to Nvidia VP of storage technology Jason Hardy.
• OpenAI's own custom chip, called Jalapeño, takes a different but related approach: shrinking how much data needs to move around in the first place.
• The new competitive battleground is data orchestration, making every part of a giant AI system work together efficiently, not just making faster processors.

For the past year, the story on Nvidia was straightforward: rivals were catching up. Amazon, Google and others had started designing their own chips, GPUs (the specialised processors, short for graphics processing units, that do the heavy number-crunching AI needs), and suddenly the company that had made a fortune being the only serious supplier looked mortal. Its shares had been drifting sideways for months.

Then came Wednesday's earnings call, and investors started paying attention to a different question. What exactly is Nvidia selling, beyond the chip itself?

What does Nvidia actually sell now?

Nvidia now ships entire systems, not just processors. Its new Vera Rubin architecture bundles the Rubin GPU with the Vera CPU (a chip that manages how data flows around the system), the Groq 3 LPX inference accelerator (hardware that speeds up the moment an AI model produces an answer), plus matching storage and networking gear.

Think of it this way. A GPU is the engine in a car. Nvidia is now selling the whole car, the gearbox, the cooling system, the fuel lines.

The reason this matters is scale. Today's AI data centres consume power measured in gigawatts, roughly the output of a small power station. At that size, keeping every component fed with the right data at exactly the right moment is a genuine engineering problem, and inefficiency is expensive.

"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 VP of storage technology, told TechCrunch AI. "We saw upwards of 3x improvement in these operations, where the Vera CPU is allowing for acceleration."

Three times faster data delivery means the GPU sits idle less. Idle GPUs cost money.

Is anyone else solving this differently?

Yes. OpenAI's custom chip, called Jalapeño, sidesteps the traffic problem by designing one large connected system where data barely needs to move at all. "We designed Jalapeño to minimize data movement and communication delays," the company wrote in a recent blog post. Same goal, different route.

What does this mean for ordinary people?

Faster, cheaper AI responses, eventually. Every improvement in data-centre efficiency tends to filter down as lower costs or better performance in the apps and services people use every day, whether that is a chatbot answering a question or a tool drafting an email.

Nvidia still faces real competition. Building a rival to the Vera Rubin system is no simpler than building a rival GPU. But the company has moved the game to terrain where it currently holds a commanding lead, and where winning requires more than a single impressive chip.