Nvidia’s edge shifts from GPU to system orchestration

The old narrative cracked. The story that dominated the market until this week was simple: Nvidia rode a GPU monopoly in the first AI boom, then hyperscalers Amazon, Google and others started building their own chips. The stock, which jumped 10-fold between early 2023 and mid-2025, moved onto a more moderate trajectory in the past year amid fears of competition for the processing core. Quarterly reports released on Wednesday changed the picture: the real advantage does not sit in the accelerator silicon but in the layer that manages everything around it.

What Nvidia now markets is the Vera Rubin architecture, a complete package that includes the new GPU alongside a Vera CPU, the Groq 3 LPX ASIC accelerator, and dedicated shelves for storage and networking. Instead of pushing more tokens, these components solve the genuine gigawatt-scale problem: moving data without a bottleneck. If the GPU is the engine, the rest of the system is the chassis, gearbox and tires.

The Vera CPU was designed specifically for memory orchestration. Jason Hardy, vice president of storage technologies at Nvidia, says, “As data centers increase compute power, memory capacity also grows, which propelled Micron in the second wave of infrastructure.” But delivering data to the GPU at the right moment is another story. Hardy reports an improvement of up to three-fold in orchestration operations thanks to the CPU, allowing flash to be used at full potential without choking the pipeline.

The problem is the same, but the solution differs. OpenAI unveiled this month the Jalapeño chip, which tries to eliminate the need to move data in the first place: all workload stays inside a single integrated circuit with a large domain that holds the request from start to finish. This is a strategy of avoiding traffic rather than managing it. In both cases the logic is identical – efficiency through smart traffic control, not just extra clock cycles – and it opens an entirely new infrastructure layer to build on.

The shift does not guarantee an automatic win for Nvidia. The company will have to compete for the orchestration layer against players building their own integrated solutions and against storage and networking suppliers that see the same opportunity. The advantage earned in GPUs buys time and early entry, but the game has moved from accelerators to systems, where the rules are completely different.