The SemiAnalysis AI Datacenter Model is used to understand current and forecast datacenter critical IT power capacity for both colocation and hyperscale datacenters with a focus on the demand driven by deployments of AI accelerators into datacenters. We track over 1,500 datacenters and their deployments with publicly available information including but not limited to property records, power usage, FOIA requests, and satellite images. We are training a CNN to accelerate the frequent satellite imaging of datacenters to expand our tracking to every datacenter across every country.
The model can be used to understand capacity constraints by major player, key considerations driving location decisions for datacenter deployment, economics of AI Accelerator clouds and datacenter colocation, and capex requirements for facility level data.
Our AI Datacenter model includes the following:
Accelerator unit shipment forecasts through 2027, Accelerator deployments, install base and AI datacenter requirements by major Hyperscalers and AI clouds.
All-in power requirements per Accelerator unit across various major chips/models.
Forecast AI datacenter critical IT power demand, PUE, utility power, annual power consumption for AI datacenters by major geography (US, North America, Asia Pacific, China, EMEA) and major region.
Current and forecast datacenter capacity by major hyperscaler by datacenter, site, cluster, and region, as well as forecast major hyperscaler datacenter and IT capex, based on a database with over 1,100 individual datacenter facilities tracked.
Forecast industry-wide datacenter capex (excluding IT equipment) and breakdown of capex by major category and subcategories (i.e. Power, Cooling, Facilities).
AI Supply/Demand Analysis broken down by major hyperscaler as well as for each major geographic region. This analysis will examine the extent to which datacenter supply currently under development will meet the power demand from estimated accelerator shipments. Certain hyperscalers will be massively short power in our estimation.
For major geographies, % of power generation from AI and non-AI datacenters, power generation fuel mix and overall power generation.
Datacenter Anatomy and key suppliers of major datacenter equipment (i.e. Generators, Transformers, UPS, Chillers, CDUs, CRAH/CRAC/PAHUs).
AI/GPU Cloud Total Cost of Ownership Analysis.
Power and carbon emissions to carry out training and inference for various AI LLM models.
Survey of global industrial electricity tariffs.
Power cost scenario analysis for major AI accelerator deployments.
Solar power economics by region/country.
The data is provided for 2023 to 2030.
The model will also include one year of quarterly updates, an initial call with SemiAnalysis to explain the model and methodologies employed, as well as subsequent ad-hoc calls to answer any questions that arise from the use of the models.