Nvidia ожидает выручку $20 млрд от Vera Rubin в Q3: рекордный запуск
Nvidia has officially confirmed that its new Vera Rubin platform is set to become the company's fastest-ramping product in history, with shipments expected to generate approximately $20 billion in revenue during the third quarter of fiscal year 2027. This milestone would represent roughly 20% of Nvidia's total data center revenue for the period, marking a significant shift as the company transitions from its Blackwell architecture to the next generation of AI infrastructure.
Record-Breaking Revenue Ramp
During Nvidia's earnings conference call with financial analysts and investors, Chief Financial Officer Colette Kress provided specific guidance on the upcoming quarter. She stated that Vera Rubin hardware sales are projected to account for about 20% of data center revenue in Q3 FY2027. Given Nvidia's total revenue projection of around $108 billion for this quarter, the Vera Rubin segment alone is expected to contribute approximately $19.87 billion. This aggressive growth trajectory signals that the market demand for the new architecture has already been validated by major hyperscalers and system OEMs.
What Constitutes a 'Vera Rubin' System
The Vera Rubin platform is not a single chip but a comprehensive ecosystem designed for large-scale AI data centers. The core components driving this revenue include the new Vera CPUs, Rubin GPUs, and BlueField 4 DPUs (Data Processing Units). Nvidia has already begun ramping up production of these units earlier in the year. Notably, Microsoft was among the first major cloud providers to deploy commercial systems based on this architecture, specifically utilizing the VR200 NVL72 rack-scale servers.
Unit Volumes vs. Revenue Value
While the revenue figures are staggering, the actual volume of units shipped presents a more nuanced picture. If Nvidia were to ship only the high-end VR200 NVL72 racks, which are estimated to cost between $5 million and $7.8 million per unit depending on configuration, the company would need to supply roughly 2,550 to 3,975 machines to hit the $20 billion target. These massive racks contain approximately 91,700 to 143,100 Vera CPUs and 183,500 to 286,100 Rubin GPUs combined.
However, these numbers are rough estimates because Nvidia will not be selling only pre-built racks. The company plans to sell MGX servers as well as individual Vera CPUs and Rubin GPUs separately. Consequently, the total unit shipment volume for these key components will likely exceed the count of rack systems alone. Nevertheless, even with hundreds of thousands of units rather than millions, the high price points of leading-edge AI hardware allow Nvidia to generate tens of billions in revenue from relatively limited physical volumes.
Market Context and Strategic Shift
Data center revenue has historically accounted for around 92% of Nvidia's quarterly sales. The introduction of Vera Rubin aims to maintain this dominance while expanding the addressable market with new architectural optimizations. The platform represents a strategic pivot, moving away from the Blackwell era to a solution that integrates CPU and GPU capabilities more tightly for inference and training workloads.
The confirmation of purchase orders from every major hyperscaler suggests that the supply chain is ready to handle this surge in demand. For industry observers, this quarter marks a critical inflection point where Nvidia's revenue mix diversifies beyond just Blackwell-based systems, establishing Vera Rubin as the new standard for enterprise AI infrastructure.
Гика
