Nvidia Grace Blackwell shipments reportedly increased 27% from last month

Nvidia’s Grace Blackwell shipments reportedly jumped 27% month-over-month, a figure that surfaced amid a broader picture of Nvidia racing to keep up with what appears to be near-insatiable demand from cloud providers and AI infrastructure builders.

What Grace Blackwell actually is, and why it matters

The Grace Blackwell platform pairs Nvidia’s Grace CPU, built on Arm architecture, with its Blackwell GPU. These chips are deployed in rack-scale systems like the GB200 NVL72, massive liquid-cooled server configurations designed for data centers running large-scale AI workloads. The kind of hardware that companies like Microsoft, Google, and Amazon need to power everything from chatbots to autonomous driving research.

Production rates for Blackwell rack systems have reportedly reached hundreds of units per week, with manufacturing yields improving over time. Cumulative shipments of Blackwell GPU packages are estimated to be in the low millions by late 2025. Analyst projections suggest Blackwell GPUs could account for more than 70% of Nvidia’s high-end market share by 2026.

The demand picture behind the numbers

Nvidia’s data center segment has reflected this dynamic in its financial results. The company reported 13% sequential growth in hyperscale revenue during Q2 of its fiscal year 2027, driven largely by increasing Blackwell demand.

Nvidia CEO Jensen Huang has previously acknowledged that demand for Blackwell products outstrips the company’s ability to deliver them. No Nvidia earnings transcript has verified the specific 27% month-over-month increase in Grace Blackwell shipments.

What this means for the competitive landscape

Nvidia’s ability to ramp Blackwell production has significant implications for the broader AI chip market. AMD, Intel, and a growing cohort of custom silicon efforts from hyperscalers themselves are all vying for a share of the AI accelerator pie. Nvidia’s head start in both hardware capability and software ecosystem, primarily through its CUDA platform, creates switching costs in enterprise AI infrastructure that are enormous, both in dollar terms and in engineering time.

The liquid-cooling requirement for these advanced rack systems is reshaping data center design. Traditional air-cooled facilities can’t accommodate GB200 NVL72 deployments, which means the Blackwell ramp is pulling forward billions in data center construction and retrofit spending across the industry.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.