AI Stack — The Silicon Railroad

From the mine to the mind — every layer of the AI economy, live. The audited value chain from energy and raw materials up to the models and apps, priced by the daily worker feed.

Researched & built with QAI proprietary data · QAI Research Framework · QAI Financial Harness · Claude Fable 5 and Opus 4.8 modelsHow it's built →
Where value sits

Two answers now, and July separated them. PRICING power still sits where it did: ASML's EUV monopoly, TSMC leading-edge plus CoWoS, the Cadence/Synopsys EDA duopoly, the SK Hynix/Micron/Samsung HBM oligopoly, Nvidia's CUDA/NVLink lock-in, and firm interconnected baseload under long-dated PPAs. EQUITY value in July accrued somewhere else — to the hyperscaler-distributors bundling AI into seats customers already pay for (Microsoft +24.6%, Amazon +15.3%), to the accelerator duopoly, and to the de-rated application layer bid up as money rotated out of hardware. The middle of the stack — memory, wafer-fab equipment, interconnect, thermal, merchant GPU rental — was sold as one basket regardless of any individual name's pricing power. Pre-revenue option plays still carry theme exposure and no present pricing power: uranium equities fell ~11% in July while U3O8 spot held ~$86/lb, the cleanest illustration of the distinction.

Top bottleneck

Firm energized power + multi-year grid interconnection — the stack's true binding constraint; new baseload can't be plugged in faster than queues, NRC licensing, and enrichment allow, so incumbents with interconnected assets price at scarcity.

Highest conviction

Nvidia (NVDA) — owns the accelerator layer via the CUDA + rack-scale NVLink moat, taxing every training and inference run regardless of which lab or cloud wins.

The AI economy, layer by layer

11 layers, from energy (L1) up to apps (L11), each sized by the combined market value of the public companies in it. Open a layer or sub-stack for QAI’s take; click any company for its conviction, bull vs bear and value-chain role.

all 432
criticalhighmedium

Only US-listed public names are sized; private / foreign names are shown for context but not counted — so a bottleneck layer can read smaller than it really is.

Layer size · market value vs the largest layer72%
Layer size · market value vs the largest layer51%
Layer size · market value vs the largest layer27%
Layer size · market value vs the largest layer10%
Layer size · market value vs the largest layer15%
Layer size · market value vs the largest layer22%
Layer size · market value vs the largest layer36%
Layer size · market value vs the largest layer100%
Layer size · market value vs the largest layer16%
Layer size · market value vs the largest layer7%
Layer size · market value vs the largest layer15%

Live prices & fundamentals from the QAI worker feed. Foreign / pre-IPO names shown as context, not investable calls. Not investment advice.

QAIInsights