Map the AI value chain from power to applications
A practical research page for separating durable AI infrastructure demand from cyclical inventory, pricing pressure, and application monetization risk.
Framework only, not financial advice.
Drilldown
Where AI infrastructure demand flows
Click any source or destination to drill into the estimate. Public hyperscaler capex, cloud commitments, and private AI lab cluster estimates are modeled separately.
One pool, four source types
This is the top-level demand number before it is routed through operators and suppliers. The cards below show what is inside L1, and what must not be counted again.
Method: use a single AI CAPEX demand pool built from company-specific 2026 AI/cloud infrastructure estimates, cloud compute commitments, and directional cluster estimates. L2 to L4 are routing and attribution layers; they do not add new spend on top of L1.
L2 is the order route, L3 is allocation of the same pool, and L4 is supplier pass-through inside L3 revenue.
Read the chain as one money-flow system
Start with who creates AI demand, then separate who places orders, who receives the first dollar, and which upstream suppliers benefit inside that dollar.
Direct spend paid to accelerator and AI server vendors. Foundries, HBM, and advanced packaging are upstream cost pass-throughs inside this revenue, not separate add-ons.
Layer boundaries prevent double-counting: L2 routes L1 demand, L3 allocates the same pool, and L4 is supplier pass-through inside L3 revenue.
| Companies | Role | AI range | Confidence |
|---|---|---|---|
Microsoft | US hyperscaler | $150-170B | public guidance |
Alphabet | US hyperscaler | $140-160B | public guidance |
Amazon | US hyperscaler | $130-155B | public guidance |
Meta | US hyperscaler | $105-130B | public guidance |
OpenAI / Stargate Private | AI-native lab | $80-120B | compute commitment |
Anthropic Private | AI-native lab | $45-80B | cloud commitment |
Oracle | AI cloud capacity | $45-58B | guidance + estimate |
xAI Private | AI-native lab | $30-55B | cluster estimate |
ByteDance Private | China AI cloud / apps | $24-32B | analyst estimate |
Alibaba | China cloud | $15-22B | multi-year plan |
Tencent | China cloud / internal AI | $11-17B | run-rate estimate |
Tesla | AI / autonomy | $8-13B | public guidance |
Baidu | China AI cloud | $5-9B | analyst estimate |
DeepSeek Private | China AI lab | $2-8B | opaque estimate |
Mistral AI Private | Europe AI lab | $2-6B | opaque estimate |
MiniMax Private | China AI lab | $1-4B | opaque estimate |
Moonshot AI Private | China AI lab | $1-4B | opaque estimate |
Zhipu AI Private | China AI lab | $1-4B | opaque estimate |
L2 answers who converts L1 AI CAPEX demand into real purchase orders. It is an execution path, not incremental spend.
L3 is a parallel allocation of the L1 AI CAPEX pool. These categories add up to L1; L4 upstream pass-through is shown separately and is not additive.
routing layer, not extra spend