AI cluster demand model

Chip Sense / Tools / AI cluster demand model

Estimates only. These tools are illustrative what-if calculators for teaching and writing — not forecasts, quotes, fab plans, or legal/compliance advice. Defaults are curated teaching values; replace them with your own inputs.

AI cluster demand model

Bill-of-materials chain for teaching: clusters → GPUs → HBM → packaging slots → implied logic wafers. Directional demand under your assumptions — not a hyperscaler forecast.

Methodology

How to use

Review Assumptions on the left. Amber badges are teaching estimates; they become Your input when you edit.
Set how many clusters and GPUs per cluster for your what-if deployment.
Adjust HBM stacks/GB and yields to match the SKU you have in mind.
Read implied GPUs, HBM, packages, and logic wafers — directional only.
Read Estimated results on the right (~ means estimate). Use Reset or Methodology anytime.

Assumptions

Edit any field — values you change are treated as your input, not Chip Sense published facts.

ClustersEstimateTeaching default of 1 cluster — set to your what-if deployment size.GPUs per clusterEstimateTeaching default resembling an 8-GPU HGX-class board — replace with your SKU.HBM stacks per GPUEstimateIllustrative HBM stack count for a high-end AI GPU class.HBM capacity per stackEstimateGBIllustrative GB per HBM stack.Package yieldEstimatefraction 0–1Teaching advanced-packaging yield default (same seed as packaging cost tool).Logic dies per waferEstimateSeed from the Yield estimator good-die geometry, or enter your DPW.Logic die yieldEstimatefraction 0–1Teaching default die yield (70%). Override with Yield estimator output.

Estimated results

Outputs use ~ / ranges on purpose. Basis: your current assumptions.

GPUs required

Estimate

~8GPUs

Clusters × GPUs per cluster.

HBM stacks

Estimate

~48stacks

GPUs × stacks per GPU.

HBM capacity

Estimate

~0.750TB

Stacks × GB/stack ÷ 1024.

Packages needed (w/ pkg yield)

Estimate

~8.7packages

GPUs ÷ package yield (scrap/rework teaching factor).

Implied logic wafers

Estimate

~0.229wafers

GPUs ÷ (DPW × die yield). Ignores multi-die packages and HBM wafer demand.

Pair with Yield and Packaging cost for essay screenshots.