Nvidia unveils Vera Rubin systems designed for gigawatt-scale data orchestration and next-generation AI training workloads.

Nvidia’s latest Vera Rubin architecture is rolling out as a tightly integrated system rather than a standalone GPU. The coordinated stack comprises the Rubin GPU, the Vera CPU, the Groq 3 LPX inference accelerator, and purpose-built racks for storage and networking. This deployment marks a strategic pivot toward data orchestration as AI compute scales into the gigawatt range.

According to Jason Hardy, Nvidia’s vice president of storage technology, the Vera CPU addresses a fundamental bottleneck: moving data efficiently across increasingly massive memory pools. “Vera is important because there’s only so much memory that you can put in a single server or any sort of compute platform,” he said. As data centers expand their computing power, memory capacity has grown in parallel, fueling growth for suppliers like Micron during the infrastructure boom’s second wave. Yet delivering that data to the GPU at optimal timing remains technically demanding. With efficiency targets tightening—particularly around tokens-per-watt—precise traffic direction has become critical.

Hardy reported “upwards of 3x improvement in these operations, where the Vera CPU is allowing for acceleration.” He added: “So now we can use our flash to its fullest potential, because we can get all that performance out of it without bottlenecking.”

Other firms are tackling the same challenge through divergent hardware strategies. OpenAI’s recently disclosed Jalapeño chip, for example, aims to eliminate data movement entirely. In a blog post earlier this month, the company stated: “We designed Jalapeño to minimize data movement and communication delays.” It achieves this by housing entire workloads within a single connected system. “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,” the post continued.

Separately, sources report that Nvidia will acquire Hugging Face for $12.9 billion.

The emergence of specialized data-orchestration systems signals a structural shift in AI infrastructure competition. While hyperscalers—including Amazon and Google—have developed their own chips over the past few years, Nvidia’s advantage is no longer confined to GPU dominance. Instead, the battleground has expanded to system-level integration: ensuring storage, networking, memory, and compute operate in concert. That layer of infrastructure is now open for competition—not just among chipmakers, but across full-stack system designers. Though Nvidia faces rivals on this new front, early evidence suggests it holds a commanding lead in building and deploying these coordinated architectures at scale.