Log in to CS Atlas
Creative Strategies · Atlas
CS Forecast Stack
The AI industry runs on a chain of dependencies most forecasts treat separately. We model it as one connected system.
Nineteen models share a single demand framework, covering the full economic stack: semiconductor fabrication capacity, compute infrastructure, data center power, token economics, AI factory unit economics, and end-market revenue, from frontier labs down to enterprise deployment.
End marketsAI Revenue & Enterprise Adoption
EconomicsToken & AI Factory Economics
InfrastructureData Center Power
SiliconFab Capacity & Compute
Why one stack
Every dependency moves together.
What that means in practice: when advanced packaging or HBM supply changes, compute availability changes with it. When compute changes, power demand and cost per token change. When cost per token changes, the enterprise adoption math changes. The industry moves as one system, and so does our forecast.
Four layers, one forecast
From fab capacity up to the enterprise.
Fab Capacity & Compute
The wafers, tools, packaging and chips the build-out runs on.
Equipment demand is built from the foundry and memory capacity plans and rebased to reported results. The foundry model carries the wafer balance by node and the CoWoS-class packaging that accelerators and server CPUs compete for. The Accelerator model builds GPUs and custom ASICs product by product and caps shipments at what packaging, HBM and power can support. Memory sizes DRAM, NAND and HBM bits, capacity and pricing against that build. Networking follows the deployed fleet across scale-up, scale-out and optics. Server CPU sizes accelerator hosts, general-purpose servers and the agent-native racks that carry out the work the accelerators plan. PC Silicon covers client compute and the AI PC transition.
Data Center Power
Whether the hardware can be switched on.
Powered Shell compares the facility power the fleet needs with the power that can be delivered, and that ceiling feeds back into accelerator shipments. The DC Capacity Tracker follows roughly 250 campuses at the operator level. AI Power 800V and Liquid Cooling size the power and cooling content per megawatt as racks get denser.
Token & AI Factory Economics
What the fleet produces and what it costs to run.
Tokenomics sizes token demand and agent workloads, and tests whether the fleet is actually used. Gigawattonomics, our AI factory model, measures revenue per watt, tokens per watt and capital per megawatt. The installed base tracks accelerators in service by cohort, and AI Server Buyers splits the accelerator market across hyperscalers, neoclouds and private buyers.
AI Revenue & Enterprise Adoption
Who pays for AI, and whether it pays back.
Enterprise AI Spend sizes what companies spend on AI, from seats and APIs to embedded software, agent platforms and readiness services, how much of it is new budget, and how much compute it funds. The AI Revenue Tracker follows the money from the frontier labs, to the clouds and neoclouds that rent them capacity, to the applications on top, using a dated ledger of public disclosures. It reads the same capacity and economics as the rest of the stack, so it can test whether the labs can fund the capacity they have contracted. Enterprise Agentic Data sizes what companies spend to make their data ready for agents and what the system-of-record software vendors now charge agents for access, checked against the agent workloads in Tokenomics. The AI Industry TAM rolls the whole stack up and reads it back against each source model.
One change, every model
Raise hyperscaler capex and follow it through the stack.
The contract updates accelerator shipments and the installed base.
Memory, networking, CPU hosts and foundry wafers reset to the larger fleet, and equipment demand follows.
Powered Shell tests the fleet against deliverable power and caps shipments where the ceiling binds.
Tokenomics and Gigawattonomics re-run on the fleet that actually deploys.
The AI Revenue Tracker picks up the new capacity and provider receipts.
A reader who holds one model sees either the cause or the effect. The stack shows both.
How we build it
Models built from research, not reverse-engineered to fit it.
Grounded in research
Every model starts from our own research: the reports we publish through The Diligence Stack, conversations with the operators, suppliers and engineers building this infrastructure, campus-level capacity tracking, and a dated ledger of public company disclosures. We use third-party data to check our work. The house view is ours, with the assumptions shown.
Triangulated forecasts
We fold pricing and supply signals in as they arrive, but the product is a forecast that holds together, not real-time price or SKU tracking.
Transparent by design
Every release recalculates each workbook, ties it back to the contract and a set of economic rules, and ships with a variance and change log showing what moved and why.
On a schedule
Official releases are quarterly, timed to the earnings window. Monthly watch notes say what came in and which way it leans. A model changes between releases only on a material break.
Questions the stack answers
Ask them on Atlas.
Access
Explore it on Atlas, or license the models.
On Atlas
Atlas subscribers explore the stack through model dashboards and the Atlas research agent, which answers questions from the same models and the research behind them.
Model Cores
The Excel workbooks are licensed in two Cores, Compute and Data Center Power, with add-ons for Fab & Supply, Economics (Tokenomics, Gigawattonomics and the AI Revenue Tracker) and PC Silicon.
Full Stack
The complete model suite, custom scenario capability, and extended analyst access.
Model structure as of the October 2026 V1 release. Links shown are structural; forecast values are available to subscribers.