All topicsCOMPUTE INDEX.

Understanding the numbers

GPU, rack, and fleet definitions, plus the difference between announced and operating capacity.

Questions in this topic

What is the difference between B200, GB200, and NVL72?How much announced AI compute is actually operational?

Questions and answers

Hardware definitions

What is the difference between B200, GB200, and NVL72?

B200 is a Blackwell GPU. A GB200 Grace Blackwell Superchip combines two Blackwell GPUs with one Grace CPU. A GB200 NVL72 rack connects 72 Blackwell GPUs and 36 Grace CPUs.

The index counts GPUs, not racks or CPUs: one GB200 NVL72 rack contributes 72 GPUs. Its B200/B300 and GB200/GB300 charts separate standalone GPU deployments from Grace-Blackwell systems.

Blackwell chart · Grace-Blackwell chart · NVIDIA specifications

Read this answer

Index cutoff · October 1, 2026

How much announced AI compute is actually operational?

Only the “Deployed” segment is classified as operating in this index. “Installing” means hardware is being racked or energized; “Announced” is future capacity. The full length of a stacked bar includes all three.

Read the deployed segment to compare operating GPU counts. An announced delivery date does not establish completion, and deployed hardware does not reveal utilization. The index does not measure an industry-wide conversion rate from announcements to live capacity.

Deployment status chart · Data sources

Read this answer

Charts and sources

  1. Blackwell chart
  2. Grace-Blackwell chart
  3. NVIDIA specifications
  4. Deployment status chart
  5. Data sources

Citation estimates use inputs through September 1, 2026; fleet counts have an October 1, 2026 cutoff. The research-topic sample covers January 1 to June 1, 2025. See each answer for its period and limitations.

All data sources