Datacenter Industry Model
Critical IT power capacity for every datacenter that matters: who is building, where, how many megawatts, and whether the power is actually there. Built from more than five thousand tracked facilities for hyperscalers, semiconductor vendors, and investors.
- Facilities
- 5,000+
- Companies
- 200+
- Coverage
- 2017–2032
- Resolution
- Site-level
Art by Axelle Wyrozumski
The AI buildout, measured in megawatts.
The SemiAnalysis Datacenter Industry Model tracks current and forecast datacenter critical IT power capacity across colocation and hyperscale facilities, focused on the demand created by AI accelerator deployments. More than five thousand datacenters are followed individually through property records, permits, power usage, FOIA requests, and satellite imagery, with computer vision models accelerating the reads on size, capacity, timelines, and progress.
On top of the facility database sits the demand side: accelerator shipments and install base by operator, the power each deployment draws, and a supply-demand reconciliation that shows which operators and regions run short of power, and when.
Who uses it, and what it decides.
The buyers of this model, and the calls they make with it.
- Hyperscalers & AI clouds
- Competitive analysis of every other buildout: who is adding capacity where, self-build versus leased, and how their own pipeline compares against the power it will need.
- Semiconductor & equipment vendors
- Supply chain planning against the facility pipeline: where accelerators, power equipment, and cooling will actually land, and which sites slip.
- Public & private market investors
- Positioning across datacenter, power, and equipment names with capacity, capex, and supply-demand by operator and region, grounded in site-level evidence rather than announcements.
Four modules, one power ledger.
Each module is built from primary research and reconciled against the others, so capacity, demand, and capex stay consistent.
Capacity by operator, site by site
Historical, current, and forecast datacenter capacity at the datacenter, site, cluster, and region level, built on more than five thousand individually tracked facilities and over two hundred companies, with coverage from 2017 through 2032. Not just who operates every megawatt, but who actually uses it, quarter by quarter:
- AI labsOpenAI and Anthropic megawatts, broken into training and inference at the building levelEnd users
- HyperscalersCapacity split into self-build, leased, and GPU cloud rented, with the end customer behind each megawatt: OpenAI, Anthropic, or internal workloadsBy end customer
- Neoclouds100+ neoclouds tracked site by site, with the end customers renting each cluster: Microsoft, Meta, OpenAI, and moreSite by site
- ChipmakersNvidia, AMD, Google, and Broadcom, with their backstopping of neocloud and lab capacity quantified at the building levelBackstops
Self-build and leased megawatts with future plans for Microsoft, AWS, Meta, Google, Oracle, Apple, CoreWeave, Tesla, x.AI, and more, plus critical IT capacity of cities and major buildings across the USA, Canada, China, Singapore, Malaysia, Australia, India, Europe and the UK, and the Middle East.
Where the constraint sits
Supply and demand, reconciled. Accelerator shipment forecasts, deployment plans, and install base for more than fifty companies, with the all-in power each accelerator draws, are set against the facility pipeline to answer the question the buildout turns on: which link binds first?
- Power
- Datacenters
- Memory
- Logic
- Advanced packaging
Chip supply is compared against datacenter supply to locate the bottleneck, and AI lab demand against contracted capacity to size how many megawatts OpenAI, Anthropic, and Meta still need to lock up before supply and demand reach equilibrium. Rollups run across the United States, North America, Asia Pacific, China, and EMEA, with grid tagging and the AI share of US power generation behind it.
Delivery, measured quarter by quarter
For every hyperscaler and neocloud, promised megawatts are scored against delivered megawatts each quarter, showing who is on time, who is late, and whether delivery is accelerating or decelerating:
On time or late
Promised versus delivered megawatts, operator by operator, every quarter
Accelerating or decelerating
Whether each operator’s delivery pace is speeding up or slowing down, and what that does to the forecast
Colo economics
The economics of datacenter colocation, an early-stage module pricing the leased half of the buildout
Capex, at the industry level
Industry-wide datacenter capex excluding IT equipment, forecast and broken down into power, cooling, and facilities, so the buildout maps to the money behind it.
From bare land to live load.
Every tracked facility advances through the same lifecycle in the model. Each stage is read from a different evidence stream, so a site is measured long before it is announced.
-
Land & permits
Parcels assemble, zoning shifts, and filings hit county records long before a shovel does.
- Property records
- Permits
-
Ground broken
Change detection on frequent satellite passes flags grading, foundations, and steel going up.
- Satellite passes
- Computer vision
-
Shell & fit-out
Rooflines, generator yards, and cooling plant sizing reveal the design capacity from orbit.
- Imagery measurement
- Design capacity
-
Energized
Substations, transformers, and interconnect queues put megawatts on the meter.
- Utility filings
- FOIA power data
-
Live IT load
Deployments land and the site joins the install base, reconciled against accelerator shipments.
- Install base
- Power draw
Every one of the five-thousand-plus tracked facilities carries this lifecycle in the model, with capacity, timeline, and operator attached.
How the model is built.
Every megawatt is observed at the site, then tested against the demand that wants it.
-
Observe every site
Property records, permits, FOIA requests, and computer-vision-accelerated satellite imagery track each facility’s size, capacity, and timeline.
-
Build the capacity ledger
Sites roll up to clusters, operators, and regions: self-build versus leased megawatts, with history and forecast to 2032.
-
Reconcile power with demand
Accelerator shipments translate into power demand and are set against the pipeline, flagging the operators and regions that run short.
Research that ships with the model.
Model subscribers receive the update notes, webinars, and analysis published against each release. A sample of recent coverage:
Common questions.
Anything not covered here, ask the team directly through the form below.
What does the Datacenter Industry Model include?
Four reconciled modules: site-level capacity by operator across more than five thousand tracked facilities, power demand and supply by geography, industry capex with the equipment stack behind it, and the economics and energy analysis on top. Data runs from 2017 with forecasts through 2032.
How do you track five thousand datacenters?
With publicly available evidence: property records, permits, power usage, FOIA requests, and frequent satellite imagery, with computer vision models accelerating the reads on size, capacity, timelines, and construction progress.
How is the model delivered?
As an Excel workbook with dashboard access, including one year of quarterly updates, an onboarding call with the team to walk through the model and methodologies, and ad-hoc calls for questions that come up in use.
Is it part of the SemiAnalysis newsletter subscription?
No. Industry models are separate institutional offerings and are not included with the annual newsletter membership.
Can it show which operators run short of power?
Yes. The supply-demand analysis sets each operator’s facility pipeline against the power its estimated accelerator shipments will draw, by hyperscaler and by region, including the ones that come up massively short.
Models that pair with this one.
Pairing depends on the decision in front of you. The capacity tracked here hosts the hardware and economics the other models cover.
Accelerator & HBM Model
Demand forecastThe accelerator shipments that create the power demand these facilities are racing to host.
View model
AI Cloud TCO Model
Cost modelThe rental economics and cost of ownership of the clusters deployed into this capacity.
View model
Energy Model
Power modelThe generation, fuel mix, and tariffs behind the load these datacenters put on the grid.
View modelGet the Datacenter Industry Model.
Start with the sales team. They come back with scoping, licensing, and pricing for your mandate.
- Scoped to your use case
- Onboarding and ad-hoc analyst calls included
- Custom research engagements available
