The story on Nvidia has been simple for the past year: GPUs are becoming a commodity, hyperscalers are building their own silicon, and the company’s 10x market cap run between 2023 and mid-2025 was never going to last. That narrative took a hit this week.
Nvidia is rolling out its Vera Rubin architecture, which pairs the Rubin GPU with a collection of specialized systems including the Vera CPU, the Groq 3 LPX inference accelerator, and dedicated storage and networking racks. None of these are token-churning GPUs. They’re infrastructure for moving data efficiently at megascale — and according to Nvidia, that’s where the real bottleneck has shifted.
Jason Hardy, Nvidia’s VP of storage technology, told me the Vera CPU delivers “upwards of 3x improvement” in data orchestration operations. “Vera is important because there’s only so much memory that you can put in a single server or any sort of compute platform,” Hardy said. The pitch: as data centers push toward gigawatt scale, getting data to the GPU at the right moment matters as much as the GPU’s raw horsepower.
OpenAI is attacking the same problem from a different angle. Its Jalapeño chip, detailed earlier this month, minimizes data movement by keeping entire workloads inside one connected system. “Its large domain allows the entire workload to remain within one connected system,” the company wrote. Different approach, same underlying logic — efficiency through smarter traffic control rather than more processor cycles.
Nvidia still has to fight for this new layer of the stack. But for now, the competition has moved to a place where building a rival GPU matters less than making the entire system work together. That’s a game Nvidia has been quietly playing for years.
