Accelerator demand
Total accelerator demand, then GPU vs ASIC / TPU.
Accelerator demand is collected from the server bill of materials and channel mix, then split between merchant or captive GPU systems and ASIC / TPU systems. The page provides total accelerator demand and the GPU versus ASIC / TPU view from 2020 onward.
Companies considered include NVIDIA and AMD for merchant GPUs, hyperscaler and other captive ASIC programs, and ASIC / TPU providers such as Google TPU, Amazon Trainium, Microsoft Maia, Meta MTIA, and Intel Gaudi. These are demand categories and do not represent a claim about each company’s reported revenue.
Total accelerator demand
Year-sheet GPU/accelerator BOM plus Hyperscaler direct (GPU + ASIC).
Download PNG · Through 2025
Accelerator demand: GPU vs ASIC
GPU = accelerator BOM × (1 − ASIC%). ASIC share is the yellow driver on CPUAccRAMDemand.
Download PNG · Through 2025
Last model update:
How to interpret this data
Use this demand view with the corresponding Supply and Coverage pages to see whether market availability is keeping pace with modeled spending.
Related data
Frequently asked questions
What does this page measure?
Total accelerator demand, then GPU vs ASIC / TPU.
Can JP Data provide a custom company, product, region, or scenario cut?
Yes. Request custom data with the required market boundary, products, companies, geography, and time horizon.