Nvidia's AI advantage is moving beyond the GPU

What Nvidia is doing can be seen by looking at what it actually sells. The company is rolling out its Vera Rubin architecture — a generation of systems pairing the Rubin GPU with a collection of other units: the Vera CPU for data orchestration, the Groq 3 LPX inference accelerator for specialized inference workloads, and full racks optimized for storage and networking. TechCrunch writer Russell Brandom, after a week of conversations with Nvidia folks, reached a surprising conclusion: these systems are extremely specialized, but their job is not 'churning through tokens' — it is making sure everything outside the GPU runs as efficiently as possible. If the GPU is the engine, these new systems are the transmission, fuel system, and steering wheel — they determine whether the engine's power actually translates into the vehicle's speed.

The Vera CPU in particular focuses on the core challenge of data orchestration. Jason Hardy, Nvidia's VP of storage technology, explained to TechCrunch: 'Vera is important because there's only so much memory that you can put in a single server or any sort of compute platform.' As data centers scale up computing power, memory capacity has scaled up too — which is why companies like Micron have gotten rich in the second wave of the infrastructure boom. But the hard part is getting data to the GPU at the right time. As companies push tokens-per-watt lower and lower, they're realizing that data movement efficiency is the true bottleneck. Hardy revealed that 'we saw upwards of 3x improvement in these operations, where the Vera CPU is allowing for acceleration,' meaning 'now we can use our flash to its fullest potential, because we can get all that performance out of it without bottlenecking.'

Interestingly, the same logic appears outside Nvidia. When OpenAI developed its Jalapeño chip, a major focus was avoiding these challenges entirely by minimizing the amount of data that needs to be moved. In a blog post earlier this month, OpenAI said: 'We designed Jalapeño to minimize data movement and communication delays. 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.' It's a different approach — not optimizing data movement, but making data movement unnecessary. Yet the underlying logic converges: increasing efficiency through smarter traffic control instead of just more processor cycles. This opens up an entirely new layer of infrastructure for companies to compete over — a layer where 'building a more powerful GPU' matters less than 'making the entire system run efficiently.'

Of course, this new focus doesn't automatically mean a win for Nvidia. Just as it has competed in the GPU market, Nvidia will have to compete with rival chipmakers and hyperscalers. But the competition has moved to a new layer: here, building a rival GPU matters less than being able to make the entire system work efficiently. And in this layer, at least in the early stages, Nvidia looks to have a commanding lead. For investors, this means Nvidia's moat is deeper than imagined — it is not just a chip company, but a systems company turning 'the AI data center' itself into its product. For the industry, it means the focus of AI infrastructure competition is shifting from 'whose chip is faster' to 'whose entire system is smarter' — and this race has only just begun.

For AI practitioners and investors, the core message of this article is that the 'compute is power' narrative is upgrading. Over the past three years, the core of AI infrastructure competition was 'whose GPU is more powerful'; now, as single data centers consume gigawatts and the complexity of memory and data scheduling rises exponentially, the competitive focus has shifted to the system layer — whoever makes every watt in a data center produce more intelligence controls the pricing power of next-generation AI. Nvidia's Vera Rubin and OpenAI's Jalapeño prove the same thing from two directions: AI's next bottleneck is not the chip itself, but the flow of data between chips. This competition over 'system intelligence' will determine the final shape of the AI infrastructure landscape for years to come.